Cultural Evolution in 

the Digital Age
Cultural Evolution in 
the Digital Age
A L B E RT O   AC E R B I

1
 3
Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom

Oxford University Press is a department of the University of Oxford.

It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© Oxford University Press 2020
The moral rights of the author have been asserted
First Edition published in 2020
Impression: 1

All rights reserved. No part of this publication may be reproduced, stored in

a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available

Library of Congress Control Number: 2019949971

ISBN 978–​0–​19–​883594–​3

Printed and bound by

CPI Group (UK) Ltd, Croydon, CR0 4YY
Oxford University Press makes no representation, express or implied, that the
drug dosages in this book are correct. Readers must therefore always check
the product information and clinical procedures with the most up-​to-​date
published product information and data sheets provided by the manufacturers
and the most recent codes of conduct and safety regulations. The authors and
the publishers do not accept responsibility or legal liability for any errors in the
text or for the misuse or misapplication of material in this work. Except where
otherwise stated, drug dosages and recommendations are for the non-​pregnant
adult who is not breast-​feeding

Links to third party websites are provided by Oxford in good faith and

for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.
 Contents

Acknowledgments vii
List of Figures ix
Introduction xi

1. A growing network for cultural transmission 1

2. Wary learners 21
3. Prestige 49
4. Popularity 71
5. Echo chambers 96
6. Misinformation 117
7. Transmitting and sharing 156
8. Cumulation 186
9. Conclusion 213

References 221
Index 237
 Acknowledgments

I wrote this book while working at the Department of Philosophy & Ethics at the
Eindhoven University of Technology. I thank Krist Vaesen who allowed me to
carry on my research with great freedom. I also want to thank the colleagues and
friends that read and provided helpful feedback on chapters, or part of chapters, of
the book manuscript: Elisa Bandini, Andrew Buskell, Mathieu Charbonneau, Max
Derex, Pierre Jacquet, Ángel Jiménez, Daphné Kerhoas, Simon Kirby, Siobhan
Klaus, Hugo Mercier, Alex Mesoudi, Helena Miton, Olivier Morin, Oleg Sobchuk,
Joe Stubbersfield, Peter Turchin, Radu Umbres, and Sacha Yesilaltay.
 List of Figures

1.1 Layers of grouping 4

2.1 The logic of Rogers’ models 27
2.2 Cards used in Asch’s experiment 35
3.1 Relative frequency of the hashtag #kiki on Twitter 50
3.2 Elvis Presley stages a polio vaccine shot before appearing on
The Ed Sullivan Show 63
4.1 Count of YouTube views of the most popular 1,000 compositions
of Johann Sebastian Bach 73
4.2 Count of given male baby names in the US in the past 100 years 74
4.3 The logic of conformity and random copying 77
4.4 Distribution of ratings of movies and TV shows on Amazon 92
6.1 A schematic representation of a transmission chain experiment 138
7.1 The scribble and the star in Sperber’s thought experiment 160
7.2 The box and the possible actions in the overimitation experiment 166
7.3 The equality versus equity meme 173
7.4 A schematic representation of a transmission chain experiment
simulating online transmission 180
8.1 Aspects of cumulative cultural evolution 187
8.2 How an “alien” language evolves 192
8.3 Average Elo rating of the top-​100 chess players through time 199
8.4 Rise of the Machines, from Saturday Morning Breakfast Cereal 205
 Introduction

The Walkman effect

New technologies, quite understandably, can be worrying matters. Academic

books often mention the Greek philosopher Plato on the topic. In the dialogue
Phaedrus, Plato has Socrates—​here his alter ego—​telling a story about the Egyptian
king Thamus. When offered the invention of writing for his country, Thamus
considers that:

If men learn this, it will implant forgetfulness in their souls; they will cease to
exercise memory because they rely on that which is written, calling things to re-
membrance no longer from within themselves, but by means of external marks.
What you have discovered is a recipe not for memory, but for reminder. And it is
no true wisdom that you offer your disciples, but only its semblance.

Socrates, and, we infer, Plato, agrees with Thamus. Writing has the “strange quality”
of presenting words “as if they have intelligence, but, if you question them,” Socrates
adds, “they always say one and the same thing.”1
Almost two millennia after Phaedrus, the Italian Renaissance poet and scholar
Francesco Petrarca gave a straight, though indirect, answer to Socrates. When
books are questioned Petrarca writes, “they answer to me, and for me they sing
and speak; some reveal nature’s secrets, others give me excellent pieces of advice on
life and death, still others tell me their own and others’ endeavors, bringing to my
mind the ancient ages.”2
Today we may feel closer to Petrarca than to Plato, but writing has indeed been
a radical innovation that has changed, and still is changing, the world we live in.
Anthropologists such as Jack Goody reflected deeply on how the diffusion of
writing technologies allowed the reorganization of societies in what we consider
their modern forms. Bureaucratic states, implementing control through imper-
sonal authority that extends in time and space beyond the limited reach of face-​
to-​face interactions, would not have been achievable, following Goody, without
writing. Record keeping in large groups would be out of the reach of our limited
cognitive abilities. The same holds for complex, universally applicable, laws sys-
tems. Many other examples are possible.3
xii Introduction

At the cognitive, individual level, writing was also revolutionary. Writing, to-
gether with graphic representations such as lists or tables, provides indeed the op-
portunity to free memory resources—​Plato’s protestations notwithstanding—​or to
disengage language from immediate usage. Specialized forms of logic, such as the
syllogism (of which the textbook example concerns, quite fittingly, Socrates: All
men are mortal; Socrates is a man; therefore Socrates is mortal), can only, again ac-
cording to Goody, be the product of a literate society.
It is not intuitive to grasp the importance of these transformations—​as your
reading of this book attests. Still, the contemporary proliferation of to-​do lists, ap-
plications, and instructions may provide a hint. An internet search can quickly re-
veal thousands of results, with titles ranging from “The To-​Do List Secret Revealed”
to “What Successful Leaders’ To-​Do Lists Look Like,” and “The Life-​Changing
Magic of Tidying Up Your To-​Do List.” A book, The Checklist Manifesto: How to
Get Things Right, suggests that a wise usage of checklists is, if not the solution, then
a key component of the solution to the increasing complexity and interconnected-
ness of tasks which characterize contemporary society. Socrates would probably
have remained skeptical.4
On other occasions, our expectations or worries seem with the benefit of
hindsights, somewhat misplaced. In 1984, the term “Walkman effect” was coined
in an article that appeared in the scientific journal Popular Music. The author men-
tions an interview in which “young people” were asked “whether men with the
Walkman [ . . . ] are losing contact with reality; whether the relations between eyes
and ears are changing radically; whether they are psychotic or schizophrenic,” and,
in a dramatic climax, “whether they are worried about the fate of humanity.” It is
easy to dismiss with a smile the misplaced anxiety expressed by the interviewer.
Walkman, or their contemporary equivalents, digital audio players (now re-
placed, in turn, by multifunctional smartphones), seem quite innocuous devices.
However, the same, or similar, concerns are expressed today regarding fears over
smartphones and social media, and the same questions are still resonating. Are we
losing contact with reality, by interacting only through screens and instant mes-
saging applications? Is the intensive use of smartphones changing the relations be-
tween eyes and hands?5
It is hard to imagine how these concerns will appear in the future. While in-
formed speculation is useful, we cannot know for certain how the diffusion of
digital media will be considered by the next generations. Will it be considered as
important as the diffusion of writing, or perhaps more so? Or, on the contrary, will
be seen as an interesting, but after all peripheral and minor, event, similar to the
diffusion of the Walkman? We cannot know for certain whether its effects on our
societies, and, ultimately, on ourselves, will be undesirable and harmful, as Plato
thought about writing, or beneficial, as books were for Petrarca. In fact, trying to
answer these questions might not be the best way to proceed. There are plenty of
 Introduction  xiii

books out there warning us about the dangers of the digital age, or to a lesser ex-
tent, enthusiastically cheering for the digital revolution.
The goal of this volume is at the same time more modest and more ambitious.
More modest because, frankly, I  do not have a ready-​to-​use answer. Of course,
I have my personal opinions, and, as they will emerge clearly in the following chap-
ters, it may be better to disclose them now. First, I think the transformations we are
witnessing now are extremely important. However, one of the main ideas explored
here—​if not the main idea—​is that these transformations can only be appreciated
when taking a long view that situates them in a broad cognitive and evolutionary
context. Second, I am more of a cheerer than a doomsayer. I believe, nevertheless,
that I appear so only because doomsayers are in the majority now. If cheerers were
vocal and relatively more numerous in the first years of the diffusion of personal
computers and of the internet, the default mood today seems quite gloomy: the
principles of the founding fathers have been betrayed. Google is evil, to say nothing
of Facebook. As a consequence, I realized that having a critical, but not necessarily
negative, attitude toward current digital developments is enough to be considered
a cheerer. So be it.
The goal is more ambitious because this book proposes that modern evolu-
tionary approaches to culture and cognition can provide a series of tools that
allow us to interpret, understand, and possibly influence, some of the tech-
nologies that characterize our digital age and how they are changing our cul-
ture. I write today of yesterday’s innovations, for tomorrow’s readers. Given the
peace of digital innovations, the goal is that the tools presented here could be
applied in general. Although I will use, of necessity, contemporary examples,
I will always try to relate them both to past technologies and mainly, general
characteristics of human cultural evolution and cognition. I  will not try to
guess what the equivalent of the contemporary iPhone will be in ten years, but
my hope is that the strategies discussed here will be useful to think about it,
whatever it may be.
Moreover, research on digital media is scattered through many disciplines, with
diverse methodologies, and, often, radically different take-​home messages. I do not
pretend I have given justice to this diversity, but I will discuss the research that
naturally fitted with my perspective. For the field of cultural evolution, though,
I tried my best to present a comprehensive picture—​through my personal lens of
course. However, cultural evolution is a fast-​evolving discipline, and many studies
that should have been prominently featured in this book have been published after
I had concluded writing and many more will appear in the following years, or even
months.
More ambitious because the cultural evolutionary approach that this book en-
dorses is a patchwork of different theories. Although I have the feeling that I did
not add much to them, as they have been mainly developed by other researchers
xiv Introduction

sometimes with no connection to one another, perhaps mixing them together

itself constitutes an original endeavor. Yes, culture can be thought of as an evo-
lutionary process in which cultural traits (be they folktales, political systems,
or YouTube videos) reproduce and compete by being copied more or less fre-
quently. However, sometimes “transmission” is just a loose metaphor that diverts
our attention from the important role that individual human minds and external
supports have in the process, and we do not need to find cultural equivalents for
any biological evolutionary feature. Yes, we are influenceable individuals, as the
last social media craze you can think about neatly demonstrates, but social in-
fluence is not everything and, more often than not, we are wary learners, and we
evaluate various cues before changing our behavior. In the digital world or not,
some things just tend to spread more than others, whatever the efforts of advert-
isers or policy makers. Not only do I think that this synthesis can make better
sense of what we observe, but I  also think that it provides a coherent theory
of cultural evolution. Even if this book is not intended as an exposition of this
theory, I will not avoid discussing some aspects of it when they are relevant for
the topic at hand.
Finally, more ambitious because there are few broad attempts to use the tools
of cultural evolution and cognitive anthropology to investigate recent or con-
temporary culture. Cultural evolutionists, for understandable reasons, prefer to
dedicate their efforts to long-​term, putatively universal, phenomena such as the
evolution of language, religion, and rituals, human large-​scale cooperation, kin-
ship, and so on. Cognitive and evolutionary oriented research on the effects of
present-​day digital technologies on cultural transmission and evolution is, with
very few exceptions, surprisingly missing. I think the risk is worth taking. For cul-
tural evolution to be a mature science of human behavior, it needs to be possible
to apply it both to hunter-​gatherer populations and to contemporary teenagers.
It should be able to say something about the thousand-​year-​long processes of lin-
guistic change, as well as about the spread of Grumpy Cat.
Along the way, I will introduce you to a few hopefully interesting, and perhaps
surprising, ideas. We will consider how digital media make, at least potentially, an
enormous amount of cultural traits accessible to an enormous amount of people,
and the possible consequences on cultural evolution of this hyper-​availability. I will
try to convince you that social influence is not an uncontrollable and inescapable
force, but on the contrary, the daunting job is sometimes to change people’s minds
at all. I will discuss how prestige and popularity are not as important as you may
think, and how echo chambers might be stronger offline than online. We will see
that the diffusion of misinformation did not necessarily worsen with social media,
and that understanding why we find some stories appealing can help us to under-
stand why online misinformation spread. We will explore the new modalities by
which cultural traits can be transmitted in the digital age and how they relate to
other forms of transmission. And yes, Grumpy Cat.
 Introduction  xv

Old planes and new phones

Driverless and electric cars are the up-​and-​coming revolution today, but good-​old-​
fashioned cars, where someone needs to sit and hold the steering wheel, did not
change much in the last century. Measuring innovations is not an obvious task, and
so researchers from the University of California, Los Angeles, decided simply to
count how many new models appeared, and when they were discontinued, using
more than one century of automobile production data in the United States. They
explicitly modeled the dynamics of the automobile industry as a process of species
competition and extinction, where each model of car was analogue to a species. In
this case, this was useful only because it allowed the researchers to analyze the data
with computational techniques developed for biological evolution, but even if we
are not going into details, this counts as the first evolutionary model applied to cul-
ture, or technology, that is mentioned in this book. We will see many more of them.
Among other findings, they reported that from the 1980s, the “extinction rate”
of American cars has been higher than the introduction of new models, and that
the average lifespan of each species-​model has also increased. This means that
there are fewer, and older, models that now dominate the market, and that innov-
ation and experimentations are becoming increasingly limited relative to the past.6
The dishwasher we bought one year ago seems precisely the same machine
that my parents used when I was a kid. While I am not too keen to investigate if
there has been any exciting change under the hood, it is reasonable to believe that
dishwashing machine technology has not changed too much in the last thirty, or
fifty, years. More worryingly, I just discovered that the plane flying me from Bristol,
in the United Kingdom, to Amsterdam and back fortnightly (an Airbus A319) has
been operative from 2006 and is based on a project developed in the late 1980s.
The slightly annoyed flight attendants to whom I ask for reassurance not only tell
me that this is perfectly fine, but that their fleet is younger, on average, than other
airlines. In fact, the most noticeable difference between boarding the same plane
now or twenty years ago, from the passenger’s point of view, would be that pilots
now log their records in a shiny iPad, instead of writing them in a paper notebook.
These examples are, of course, cherry-​picked, but compare this to what happens
for digital technologies. In 2006, the year of the inaugural flight of my beloved A319,
which I want to consider state-​of-​the-​art technology, the iPhone did not even exist.
The first generation iPhone, which gave the initial momentum to the worldwide
diffusion of smartphones, was presented to the public one year later (I also invite
you to search for images of it, as it looks, well . . . old). In the same year, an unfor-
tunate senior analyst commented skeptically on Facebook’s public opening: “I do
not understand why they don’t want to be a college network . . . They’d get higher
[advertising rates] than MySpace, I guarantee you.”7 In any case, Facebook is old
news, and smartphones are, according to many, just a taste of an impending future
of wearables and ubiquitous connectivity.
xvi Introduction

From one side, then, the technological advance of digital tools is proceeding
with a pace that seems faster—​probably orders of magnitude faster—​with respect
to other technological domains. However, what is more important is how this ad-
vance directly impacts our lives. I admit I have been unfair with the automobile
and aeronautic industries’ research and development sectors. Lots of impressive
work is being done that is only vaguely understandable to most of us, such as ex-
perimenting with new materials to reduce the weight of aircraft. Less weight means
less fuel, and this translates into huge savings for airline companies, and in theory,
in the long run, in a reduction in consumption of carbon-​based resources, which
should be good news for everybody (I said “in theory” because the reduction in
costs for airline companies results in an increase of travelers, so that, all things con-
sidered, is not clear that the net usage of fuel would actually decline). Anti-​lock
braking systems (ABS), which prevent automobile wheels from locking, and as
a consequence prevents cars skidding when braking, were only a mandatory re-
quirement in all vehicles from the early 2000s in Europe, and even more recently in
the United States.
True, ABS can save your life, but its effect on day-​to-​day interactions appears
somewhat limited, if we compare it with the likes of email, social networks, or
smartphones. I could provide—​and I will do throughout this book—​several figures.
(here is one: the Pew Research Center reported a 2015 survey according to which
the majority of daily conversations between US teenagers happened through text
messaging. Non-​digital, in person, contacts were in the fourth position, preceded
by instant messaging and interactions through social media websites8), but allow
me to introduce you to a relatively low-​tech and perhaps unexpected example. I am
writing this sentence on my laptop, using my editor of choice. Thinking about it,
after finishing school, I never wrote anything, besides quick notes and scribbles,
with pen and paper. Has this had an effect on my writing?
It has been noted how using a digital editor blurs the boundaries between
composing and reviewing. The revision of text is simultaneous to the process of
writing. This goes down to the level of sentences; for each sentence I write, I gen-
erally start with a few words and then I add some others, usually changing the pre-
vious ones in a continuous interactive process. Long sentences written in one go
are rare pearls (there is also generally some mistake in them). Grammatical errors
and typos are automatically highlighted, and, if I am stuck somewhere, I can al-
ways ask for synonyms or antonyms of a word. This process would be simply im-
possible with a mechanical, analogic, typewriter. Toward the end of the day, for
an “old-​style” revision I save my draft as a pdf file, so that I cannot change the text
while reading it.
On top of this, digital editors, together with the structure of files and folders
in personal computers and other devices, provide writers with an easy-​to-​grasp
global view of their manuscripts. Searches for specific words can be performed on
an entire file or, actually, across different files, for example the chapters of a book.
 Introduction  xvii

Pieces of writing can be instantly and cost-​free moved from one place in the text
to another. A  practically unlimited number of previous versions can be stored
and quickly recovered. I keep a “bits and pieces” folder where I save paragraphs or
bigger sections that I wrote but then decided to cut. I do not need to know where
exactly they will go, but I know that I can find them with a few keystrokes and use
them when I need to. None of this was impossible before the digital age, but it was
certainly more difficult and not as natural as it is today. And I am not even touching
my internet browser.9
In sum, digital technologies are not only developing faster than technologies
in many other domains, but their effects on our daily life—​even for deceptively
simple tools, such as editors and word processors—​seem deep and persistent. I do
not need to work hard to convince you that this is the case. I will, though, have
to persuade you that cultural evolution and cognitive anthropology, and the long
view they deliver, can provide an interesting and enlightening way to look to those
effects.

The long view

The reach of the internet is vast: as of June 2018, more than 55 percent of the world’s
population was online. In Europe and North America, approximately nine-​in-​ten
adults use the internet daily. This figure goes to ten-​in-​ten when considering indi-
viduals between 18 and 50 years old. The growth rate of social media is even more
impressive. Facebook declared 2.2 billion monthly users worldwide in the first
quarter of 2018. In the same period, they reported 185 million “daily active users”
in the US and Canada, which is roughly 85 percent of the population between 18
and 69 years old. Americans spend, on average, 3.5 hours a day online. Just go back
20 years, and only 3 percent of the population had web access.10
Writing in 1685, Adrien Baillet, best known as a biographer of philosopher
Descartes, lamented the unfortunate situation brought about by Gutenberg’s in-
vention of the press—​that happened around one century and a half before he was
writing—​and its diffusion:

We have reason to fear that the multitude of books which grows every day in a
prodigious fashion will make the following centuries fall into a state as barbarous
as that of the centuries that followed the fall of the Roman Empire.11

Baillet, thinking about it, was quite right. The diffusion of the printing press gen-
erated an “early modern information overload” that had to be coped with a series
of inventions and refinements of cognitive tools. Alphabetical back-​of-​the-​book
indexes, thematic catalogues of books, or the practice of taking notes while reading
are all examples of these coping strategies. On the other hand, long-​standing
xviii Introduction

intellectual traditions, such as the Art of Memory, gradually lost importance, as

they were no longer of use.12
New information technologies require other technologies to support them, or
they create new habits. As they adapt to us, we, as individuals and as a society, adapt
to them. We are the fellows of Baillet, a few centuries after. Not everybody agrees
that looking at these changes in a long perspective is a good idea. The legal scholar
Cass Sunstein, one of the most vocal and respected critics of the social and political
consequences of the diffusion of digital media, wrote:

I will be comparing the current situation [ . . . ] to a communication system that

has never existed—​one in which existing technological capacities and unimagin-
able improvements are enlisted to provide people with [ . . . ] substance, fun, di-
versity, challenge, comfort, disturbance, colors, and surprise.13

I am doing more or less the opposite. I will be comparing the current situation to
past communication systems and cultural phenomena in other domains, and I will
discuss it in relation to some general cognitive properties and to some expected
characteristics of cultural evolution. I will try to understand the features that make
cultural evolution in the digital age special, and what they really change.
One of these features is availability. The diffusion of the internet and of devices
that offer easy and cheap online access, such as your smartphone, create an unpre-
cedented growth of the network in which cultural transmission is possible. If our
networks of friends or close contacts may not have changed too much (at least yet),
the networks in which we exchange information have changed, and radically so.
We can have access potentially to more and more information than we ever had.
It is difficult to think of something that is not online—​we will see examples of spe-
cialized (“exoteric” programming languages) or just weird (“Queen Elizabeth is a
vampire”) pieces of knowledge throughout the book. What are the consequences
for cultural evolution?
The other side of the coin of availability is reach. In the same way as we can ac-
cess online a profusion of information, everybody can, again potentially, spread
their ideas cheaply to millions of people in a click. This resembles a slippery cliché
(and, as we will see in Chapter 4 “Popularity,” you and I are in fact very unlikely to
succeed in reaching these millions of people), but it is literally true: each time you
tweet, for example, your words can effectively be read by everybody else and some-
times this happens to the likes of us—​even though it might be, more than anything
else, a frightening experience.
A different aspect is opacity:  as many have observed, our digitally mediated
interactions often lack transparency. Think fake social media profiles, Twitter bots,
anonymous comments and reviews of everything you can imagine. Unlike the situ-
ation with most analogic, offline, interactions, we often do not know with whom we
are interacting or even if we are interacting with another human. In the same way,
 Introduction  xix

algorithms and recommendation systems tune which posts we see in our news-
feed, or which results we receive in our Google searches, generally without us being
aware of it. I will not even try to provide a complete picture of these changes, but,
as above, I will try to understand how some of them impact cultural transmission.
A feature seemingly opposed to opacity is what we can call explicitness.
“Seemingly” because it is not too far-​fetched to suggest that explicit and quanti-
fied online information became successful, at least in part, as an answer to opacity.
If we cannot be sure with whom we are interacting, there might be some value in
knowing what they do, what they like, and what other people think about them.
The same goes for pieces of information: we can know exactly how many other
people “liked” or generally reacted to a post, and practically everything online is
rated, scored, commented, and has a number on it. Top lists of virtually everything
exist and, often, the top-​rated (by humans or algorithms) items are the only ones to
which we can have access.
Finally, fidelity. If, in the popular image, cultural evolution suggests that ideas
and behaviors spread by replicating gene-​like from individual to individual, prac-
titioners tend to be more circumspect. Cultural transmission is often compared to
the Chinese Whispers game, where players in line whisper to each other a sentence
and the final result is usually very different from the starting message. You start
with “I love eating toasted cheese and tuna sandwiches” and finish with “I like plan-
ting trees and learning languages.” This metaphor works up to a point, because to
be considered cultural something needs to remain stable when passing through
different individuals, and, as we will see throughout the book, understanding why
this is the case is often the most interesting ingredient of an explanation of a cul-
tural trend. Different media provide different support for faithful transmission.
Digital media create a cheap, fast, and accessible way to make highly faithful cul-
tural transmission possible.
Of course, the way in which these features changed in the digital age is a matter
of degree. All of them can be thought of as being on a continuum in which digital
age technologies tend to push them on the more available, more opaque, more
faithful side, and so on. Circulation of books, analogic media like radio and tele-
vision, or the increased facility of traveling, all influenced the availability of in-
formation. Plenty of other ways, besides using digital media—​language itself is an
obvious one—​exist to increase the fidelity of cultural transmission. Still, I believe
these features provide a map to think about the changes brought about by our on-
line life on cultural evolution in the long run, and I will keep on referring to them
in the following pages.

It is easy to forget about it, but each time we gaze at our smartphone we are only
seconds away from an incredible amount of information. In Chapter 1 (“A growing
network for cultural transmission”), I  will discuss how digital media increased,
and are still increasing, the network in which cultural transmission can occur.
xx Introduction

Some research shows that our social networks did not change radically in the past
few years. We may have the same number of friends we had before the advent of
Facebook, family matters as it always did, and we still tend to befriend people who
work with us, live in our neighborhood, and whom we greet in the park. I think this
picture is broadly correct (and, if anything, reassuring). What happened, however,
is that digital connections, together with our tendency to share information for
limited or null gains, made cultural transmission relatively detached from these
networks. We exchange information with many other individuals, who are not part
of the good-​old trusted networks of family and friends.
While this situation generates obvious advantages, it also encourages strong
worries. The focus of many of these worries seems to be that, faced with much more
available information coming from opaque sources, we would end up being easily
manipulated. In Chapter 2 (“Wary learners”), we will start to dive deeper in cul-
tural evolution research. Albeit there are many differences, a common idea is that
human cognition is specialized for processing social interactions, communication,
learning from others. There is a popular caricature of evolutionary psychology, for
which we are adapted to our ancestral environment, the African Pleistocene sa-
vanna, or something similar. Perhaps a better caricature is that our environment
is neither the savanna nor the forest, but other people. From this perspective, I will
propose that we should at least be suspicious of accounts that propose that we are
too gullible, and I suggest instead we can be considered wary learners.
But, of course, listening to what others say and copying what they do is in
many cases our best option. Drawing on the background developed in the second
chapter, we will return to examine specifically online phenomena. Chapter  3
(“Prestige”) will scrutinize the role of influencers, celebrities, and famous people.
Cultural evolutionists talk about prestige bias in this regard: one can make use of
signs of deference, respect, or simply check from whom other people are learning,
and choose those individuals as cultural models. This tendency gives us today, in
large and opaque networks of cultural transmission—​the story goes—​the celebri-
ties “famous for being famous,” if not the danger of radical proselytism from charis-
matic leaders. We will see, however, that experiments and data tell something more
nuanced: celebrities’ influence works only in specific conditions and it is far from
being a blind force. Recent internet trends, such as the rise of micro-​influencers,
figures who are expert in their domain and who can engage in direct relationship
with their followers, are consistent with this picture.
The tendency to copy the majority will be the topic of Chapter 4 (“Popularity”).
We will explore various nuances of cultural herd-​behavior, and whether they are
important or not online. We will see how the rich-​get-​richer phenomena that are
considered a feature of internet popularity are also common offline and that they
are to be expected every time we copy others, even when we do not prefer ma-
jority cultural traits: as for celebrities, our tendency to copy the popular opinions
may have been overestimated. We will also discuss how digital technologies permit
 Introduction  xxi

radically new forms of popularity advertisements, from the real-​time quantifica-

tion of “likes” in social media to the explosion of consumer reviews, or top-​lists of
virtually everything: what is the effect of the explicit quantification of popularity
online?
Social media have recently been touted as the biggest threat to democracy. In
Chapter 5 (“Echo chambers”) I will focus on the phenomenon of echo chambers,
trying an evaluation from the broad perspective of cultural evolution. It has been
noted that individuals associate on social media in communities of like-​minded
people, where they are repeatedly exposed to the same kind of information and,
even more importantly, they are not exposed to contrary information. How strong
are echo chambers? What are their effects on the flow of online information? We
will see that although the formation and existence of echo chambers is consistent
with our cognitive and evolutionary approach, individuals are exposed online to a
considerable amount of contrary opinions: in fact, against current common sense,
to more diverse opinions than is the case in their offline life. As a consequence, the
increase of polarization, which many link to a more informationally segregated
society, could also have been overestimated, or, in any case, may be due to other
motifs than our social media activity.
We will then look at another of the big questions that understandably concerns
many of us. Is our current digital age a post-​truth era? Did fake news, conspiracy
theories, and the like find in digital and, in particular, social media, their perfect
niche, and are they thriving out of control? In Chapter 6 (“Misinformation”), I will
take a broad view of misinformation: the spread of factually false claims is as old
as cultural transmission itself, and to assess the real danger represented by social
media we need to understand what kind of cognitive triggers are activated by suc-
cessful information, online or offline. We will explore in detail how some cultural
traits possess features that make them particularly well suited for retention and
transmission, conferring them a selective advantage relative to other traits. All
other conditions being equal, a shorter and rhymed lullaby will spread more easily
than a longer and unrhymed one, and we need to work hard to convince children
to eat vegetables, but not pizza. Unfortunately, all other conditions are rarely equal.
Despite that, some psychological tendencies are stable and common enough to
allow us to discern their effects among the clutter of forces that influence cultural
dynamics, and to predict which ideas or behaviors are more likely to spread in cer-
tain situations.
From this perspective, misinformation can be manufactured building on fea-
tures that make it attractive in an almost unconstrained way, whereas true news
cannot, simply because it needs to correspond to reality. Misinformation can be
designed to spread more than real information (whether this is a conscious process
or not). Misinformation is not low-​quality information that succeeds in spreading
because of the shortcomings of online communication. Quite the opposite: misin-
formation, or at least some of it, is high-​quality information that spreads because
xxii Introduction

online communication is efficient. The difference is that quality is not about truth-
fulness, but about how it fits with our cognitive predispositions. Online fake news,
is, from this perspective, not much of a political and propagandistic phenomenon,
but is more similar to the diffusion of memes, urban legends, and the like.
In Chapter 7 (“Transmitting and sharing”), we will examine in more detail the
processes that underpin cultural transmission. The chapter builds from the idea
that it is useful to consider copying as an active process in which individuals pay
particular attention to certain traits, succeed in acquiring and remembering them,
decide whether to transmit them or not, and in turn reproduce them. I will sug-
gest that digital technologies provide several fidelity amplifiers, which offer a quick
and cheap way to spread cultural traits with high fidelity, in comparison with more
traditional forms of cultural transmission. Whereas the success of most digital
content is still due to the possibility of being transformed and remixed, social
media sharing is quick and effortless as pressing a button. Are there consequences
on what kind of content is more likely to spread? (If you make it to here, you will
finally find Grumpy Cat.)
Chapter 8 (“Cumulation”) will consider what cultural evolutionists call cumula-
tive cultural evolution, that is, the idea that culture increases in complexity: more
traits, more efficient, building on previous innovations. I will suggest that this is not
a necessary outcome, and that different domains show different signs of cumula-
tion. What are the consequences of the fidelity and the hyper-​availability provided
by digital media for the cumulation of culture? How does this influence cultural
change at large? A possibility is that online transmission allows for more cumula-
tion in domains where it was limited before. Not surprisingly, it also allows for the
stockpiling of useless information—​you may call it junk culture—​that, paradoxic-
ally, would be more easily discarded in less effective systems of transmission and
storage. How do we find and retain good information, and get rid of the rest?
In the Conclusion, I will reflect again on the main message of the book: to under-
stand the current impact of the digital environment we need to take a long view,
and to recognize the cognitive and evolutionary underpinnings of the changes and
challenges we face today. This does not mean that these changes and challenges
are not novel, or that they are unimportant, but that we need a solid background
to fully understand them. While somewhat encouraging, at least with respect to
many contemporary accounts, the long view defended here does not imply that we
should unquestioningly accept the status quo, or that we should not actively work
to change what we believe are the negative aspects of our digital interactions. Quite
the contrary: only knowing which aspects are genuinely problematic, and why, can
help us to understand them. A cultural evolution perspective can suggest which
realistic modifications to our online lives could make them better, and hopefully
make society as a whole better too.
 Introduction  xxiii

Notes

1. Hackforth (1972)
2. Petrarca (1951). My translation
3. Goody (1977)
4. Gawande (2009)
5. Hosokawa (1984)
6. Gjesfjeld (2016)
7. http:// ​ w ww.bloomberg.com/​ n ews/​ a rticles/​ 2 006-​ 0 9-​ 1 1/​ f acebook-​ o pening-
​the-​doors-​wider
8. http://​www.pewinternet.org/​2015/​08/​06/​teens-​technology-​and-​friendships/​
9. Kirschenbaum (2016)
10. Data from: https://​en.wikipedia.org/​wiki/​Global_​Internet_​usage
http://​www.pewinternet.org/​fact-​sheet/​internet-​broadband/​
https://​investor.fb.com/​investor-​news/​press-​release-​details/​2018/​Facebook-​Reports-​
First-​Quarter-​2018-​Results/​default.aspx
http://​w ww.digitalcenter.org/​wp-​content/​uploads/​2013/​10/​2017-​Digital-​Future-​
Report.pdf
11. Blair (2003)
12. Eisenstein (1979)
13. Sunstein (2018)
 1
A growing network
for cultural transmission

Digital Hutterites?

A few years ago, in 2010, I spent my New Year’s Eve in Batu Putih, a small village in
the north of Sulawesi, in Indonesia. As is often the case, I had to pose for an endless
series of pictures with all the participants of the party and, at the same time, drink
palm wine simulating appreciation (mixing it with Coke seemed quite useful).
But another thing was puzzling me: when I was introduced to young people, al-
most all wanted to know my family name. As a relatively young anthropologist, my
thoughts were wandering on the importance of kinship in traditional societies or
on the fact that cultural differences required some form of respectful distance. In a
few hours, however, I had the simple but perhaps surprising answer. Other guests
needed my family name to find me on Facebook, and (that’s the word) friending
me. Some days after, when I gathered some information—​on the web, of course—​I
discovered that in the previous month (November 2009) Indonesia was estimated
to be the second fastest growing country for Facebook users, both by absolute
number and by percentage difference. Today Indonesia has around 130 million
users, and it is the third country for Facebook penetration, together with Brazil,
and after the US and India.1
If you are around my age, you are probably on Facebook too. How many friends
do you have? The median, calculated a few years ago, was 200 (however, as you are
reading this book, my bet is that you will have more). Other estimates are some-
what lower, pointing to a number between 100 and 200. It has been proposed that
this quantity is not accidental, but it results from cognitive constraints. Primates,
including humans, have unusually larger brains in respect to other species, and
they have unusually complex social lives. The social brain hypothesis suggests that
there is a relationship between these two facts: primates evolved their large brains
to manage the intricacies of the social interactions within their groups. When
Toshi the chimp decides whom to approach today, he needs to remember that
Lee attacked him yesterday, but Kaky did not. Jojo shared some of her food with
him a few days ago. Perhaps Toshi even needs to remember that Loulou, while not
interacting with him directly, was aggressive with his friend Kaky. Processing, con-
tinuously updating, and retaining all this information is a cognitively demanding

Cultural Evolution in the Digital Age, Alberto Acerbi. Oxford University Press (2020) © Oxford
University Press.
DOI: 10.1093/oso/9780198835943.001.0001
2  A growing network for cultural transmission

task. The bigger the social group, the bigger the cognitive effort, and the bigger the
brain needed for it.2
Anthropologist Robin Dunbar considered the size of the neocortex—​an area
of the brain in mammals that is considered evolutionarily recent, and devoted to
high-​order functions such as sensory perception, generation of motor commands,
and, in humans, language—​or, better, the neocortex ratio (how big is the neocortex
with respect to the total volume of the brain) in various species of primates. He
found that the neocortex ratio was correlated with the average size of each species’
group: primates with a higher neocortex ratio live in bigger groups. In general, pri-
mates have bigger neocortex ratios than all other animals, and, among primates,
humans have the highest ratio: our neocortex is 30 percent larger, with respect to
the whole size of the brain, than the neocortex of any other primates. Extrapolating
the correlation between neocortex ratio and group size, Dunbar estimated that a
species with our brain should live in groups of around 150 individuals (plus or
minus 50), a figure now universally famous as “Dunbar’s number.”3
The size of several human communities coincides with this number. Dunbar
analyzed data from around 20 contemporary hunter-​ gatherer populations,
and he found that, at what he defined as an “intermediate level of grouping”—​
corresponding to the village, or to the self-​defined and culturally significant (this is
who we are) notions of “clan” or “group”—​these populations comprised on average
148 individuals (or 156, or 135, depending on the criteria by which different popu-
lations are included). Neolithic villages in Mesopotamia have been estimated to be
around this same size. Even modern organizations are arranged similarly. Armies,
as in Dunbar’s example, were historically divided into semi-​independent units,
from the Roman Centuria (which means it was composed of 100 individuals) to
the Second World War Company, a group comprising between 100 to 200 or 250
soldiers.
Hutterites, a group of Christian Anabaptists (such as the Amish or the
Mennonites), live in small and self-​sufficient communities, based on agriculture
for subsistence and are resistant, so far, to most modern cultural innovations.
When a Hutterite community becomes too big—​Hutterites still have an average
of five children per family—​it splits forming a daughter community. Dunbar and
colleagues analyzed around 100 of these fission events from a century of history
of two Hutterite colonies in South Dakota, and they found that the average size
of the colonies at each event was 166 individuals (plus or minus 26). In other
words, Hutterites self-​limit their communities to a size that corresponds well to the
threshold found by Dunbar.4
As the popular story goes, our primate brain constraints the number of rela-
tionships we can track and thus the size of the groups we can live in. Our Snapchat
app and busy Facebook feed did not change much the situation: the size of our
circle of online friends is the same as hunter-​gatherer’s clans or Hutterites’ com-
munities. A few years ago, a tabloid reported that the Swedish tax authority had
 Layers and time  3

planned to reorganize its units with an upper limit of 150 employees, according to a
“leaked internal report.” The reorganisation was “partly shaped on studies of apes”
and, according to the tabloid, “employees [were] not flattered by the comparison.”
I could not find out what happened afterwards with the Swedish tax authority, but
Dunbar’s number reached an uncommon popularity for an academic finding, and
150 became a threshold with quasi-​miraculous powers. The story, however, as
Dunbar himself pointed out several times, is more complicated.5

Layers and time

As noted above, the number 150 was found, for hunter-​gatherer populations, at a
specific level of grouping, corresponding to the village or clan. This is a meaningful
threshold, but, as Dunbar noted from the beginning, just one of the possible levels
of organization that can be considered. Contemporary hunter-​gatherers show
complex population organizations with different level of groupings. Smaller units,
sometimes referred as bands in the ethnographic literature, count between 30 and
50 individuals, while bigger units, or tribes, goes from 500 to more than 2000 indi-
viduals. These levels of grouping vary from society to society—​not all populations
have all of them—​and, importantly, they change depending on the situation: an
individual can act as a member of a smaller group in particular circumstances, for
example when sharing food, but as a member of a bigger group in others, for ex-
ample in case of warfare.
In fact, in further analyses of the structure of hunter-​gatherer societies and of
other populations with different organization, grouping is described as a series of
“hierarchically inclusive layers” with a consistent scaling, where each successive
group is three or four times larger than the previous one. You can think about
it, as in Figure 1.1 below, as a series of successive concentric circles. One or two
persons, that is, you and possibly an intimate partner, compose the central circle.
After that, there is a circle of around five individuals, the intimate friends, or the
“support clique” you would tell most secrets to, and ask for help in times of dis-
tress. Fifteen or 20 individuals represent the best friends or “sympathy group,”
people you contact and see often, and with whom you can have a relaxed conver-
sation. Around three times more, and we get the 30–​50 individuals composing
the hunter-​gatherer bands, or what Dunbar calls “good” friends, say the people
you would invite for sure to your wedding. Then there is the number, the 150 (plus
or minus 50) members of the village, the company, your casual friends, or, as we
saw earlier, your Facebook friends. The circles extend outside this level. A first
further grouping, at around 500 individuals, denotes the acquaintances, people
you know, but with whom you do not have meaningful relationships—​you prob-
ably would not say hello to when meeting one of those in the street—​or, according
to some ethnographic accounts, the megaband. Lastly, around 1500-​ –​
2000
4  A growing network for cultural transmission

1–2
5
15–20
30–50
150
500

1500–2000

Figure 1.1  Layers of grouping.

Reprinted from Trends in Cognitive Sciences, 22(1), R.I.M. Dunbar, The Anatomy of Friendship,
pp. 32–​51, doi.org/​10.1016/​j.tics.2017.10.004 © 2017 Elsevier Ltd., with permission from Elsevier.

individuals constitute tribes or, in modern western societies, it represents “prob-

ably the number of faces we can put names to.” Hence, 150 is only one of the pos-
sible thresholds and various other groups, with various sizes, are more important
depending on the nature of the relationships, or the specific activities we are
carrying out with their members.6
Another direction in which the story gets more complicated is that cognitive
factors, as described by the social brain hypothesis, are only one of the possible
constraints of groupings. Social relationships, especially in the innermost layers,
require investing time to be maintained. Even if our brain could remember thou-
sands of individual identities and manage all the information about the relation-
ships among them, we would still need time to meet, greet, and interact with our
contacts. Dunbar mentions an example that may be familiar to most readers, and
that is part of the plot of countless comedies (I confess the one I have in mind is
Old School with Will Ferrell): when we enter into a serious couple relationship we
dedicate less time to our close friends. In fact, Dunbar and colleague have shown
 Virtual networks  5

that passing from being single to being in a relationship often implies losing one
of the members of the five individuals’ support clique. Put simply, there is no time
for everybody. The same happens when going outwards in the layers. Each of them
is associated with a typical time signature, and if we do not interact enough with a
person, the relationship will deteriorate, and the person will figuratively move to
the adjacent more external layer.7
In this respect, the digital age could have changed everything. Social media, un-
like offline interactions, allow connecting at the same time with a virtually infinite
number of people. When we tweet, update our status on Facebook, or post a pic-
ture on Instagram or a video on YouTube, this reaches instantly and effortlessly
all our contacts, and potentially even more individuals. We can copy and paste
Christmas greetings to hundreds of our friends and family members in WhatsApp
or Snapchat in a matter of minutes. Facebook relieves some of the cognitive load
needed to maintain relationships by reminding us, amongst myriad other things,
the birthdays of our friends. In fact, there is at least an extension for Chrome (the
web browser developed by Google) that automatizes the whole process, and posts a
message of birthday wishes on the wall of the birthday girl or boy in the appropriate
day, without the need of any action from us. Digital communications, and social
media in particular, could stretch the circles of our groups, by easing time and cog-
nitive constraints, and potentially allowing the maintenance of larger numbers of
meaningful contacts. But then, why do we still have only 200 Facebook friends?8

Virtual networks

Several researchers have analyzed various kinds of social media networks, and,
more broadly, online communities, drawing on the background just described. Are
they special in this respect? As you may suspect, the short answer is: no, online
communities broadly match the structure of offline, old-​style, communities.
Until 2009, Facebook was organized around the concept of networks. A network
could have been a university or a high school (educational network), a company or
an organization (work network) or, finally, simply determined by the geographical
location (a town or a region). As for default settings, all personal data were access-
ible by the individuals in the same network. In sum, one did not need to be a friend
of someone else to see all their posts and photos—​as well as their address and con-
tact information—​as long as they were part of the same network. Add to this that
crawling (using a program that automatically and systematically goes through
various web pages and extracts the information they contain) the social media was
not restricted. In other words, having access to, say, the network of the Harvard
University or the city of New Orleans would have allowed downloading all the per-
sonal data of all the members of the network, unless they had changed their default
settings (which is uncommon today, and more so at the time).9
6  A growing network for cultural transmission

Besides putting in perspective our (justified) contemporary worries about

privacy, this set-​up allowed researchers to extract complete data on Facebook
interactions in large groups. Taking advantage of this, Dunbar and colleagues were
able to check not only the size of individual networks (how many friends one had),
but also the number of times each person was interacting with each of the others,
thus determining, approximately, in which of the circles described above the con-
tacts were. They found that there were similarities with the layered structure de-
scribed above. There were one or two friends who were contacted twice a day or
more (corresponding to the central circle), around five contacted about once (the
support clique), and so on until the 50 “good friends” who were contacted occa-
sionally, once every month.
Surprisingly, the analysis of Facebook networks revealed that the contacts were
less than in offline life: the most relevant 150-​individuals-​layer was missing, as well
as, perhaps less surprisingly, the more external ones. Dunbar and colleagues argue
this is because early Facebook users were on social media to keep in contact with
people they knew well, which would correspond to the inner grouping layers. The
tendency of friending less close individuals, casual friends or acquaintances, would
begin only later, especially as an effect of the widespread diffusion of Facebook,
following the public opening (2007), and more explicitly with the “People You May
Know” feature, that Facebook introduced in 2008. As a possible support for this
conjecture, an analysis of two more recent samples—​from a 2015 survey—​found
that the distributions of the Facebook network sizes of single individuals have
averages around 150, and they are thus consistent with the predictions of the social
brain hypothesis. Data from Twitter tell a similar story.10
Other researchers investigated the network structure of a virtual world, the
MOORPG Pardus (MOORPG stands for massively multiplayer online role-​playing
game), which has more than 400,000 registered users. Players roam “in a techno-
logically advanced but war-​torn universe,” and they have various options to obtain
fame and success. Characters can specialize in trade, construction of spaceships or
buildings, extraction of raw materials, and of course contraband, war, and many
other choices that one can pursue in the open-​ended game. Players interact with
each other and create small and informal groups of friends who provide mutual
support. At a higher level, however, they can organize themselves into “alliances,”
groups that are formally recognized in the game. Above that, there are “factions.”
Factions are pre-​defined and limited to three (“Empire,” “Federation,” “Union”),
but their size is not. Finally, there is the entire population of each “society” of the
game. Alliances are the most important level of organization: an alliance allows
players to operate with joint forces in war, trades, and in all the other activities the
game involves. Alliances have their private chat channels, and can organize taxes
collection from members to fund common projects. They do not have a predefined
size, but the researchers found that the maximum number of members was 136,
 Virtual networks  7

which immediately made them think to Dunbar’s number. Their average size, how-
ever, was found to be around 25, so in this case sensibly lower than the expected
number, or, perhaps, closer to the estimates for the inner layer of the “sympathy
group.”11
There are a few caveats to the idea that digital networks replicate the struc-
ture of offline networks. The Pardus analysis shows that the actual sizes do not
match exactly with Dunbar estimates. Related—​and this is a general problem,
no matter whether we are considering offline or online groups—​it may be hard
to precisely test the theory precisely with empirical data. First, the margins of
error are large: think 150 plus or minus 50. Second, this is associated with skewed
distributions: network sizes are not normally distributed, that is, they do not ar-
range themselves nicely around the mean. Generally, there are very few individ-
uals highly connected, and many more individuals who are not. If one adds, on
top of this, that there are several layers of grouping and that the scaling ratio can
be variable, it intuitively looks that it is often possible to arrange the empirical
data to fit the predictions. The main result of the Pardus analysis—​the existence
of hierarchical layers of grouping—​may also be due to the fact that this structure
is hardcoded in the game, at least for some of the levels, such as alliances and
factions.
Another difficulty concerns the size of networks in the most used social media,
such as Facebook or Twitter. Dunbar used early data, or relatively small surveys,
which mostly seems to confirm his prediction. Other estimates—​I cited the me-
dian of 200 at the beginning of the chapter—​indicate that the average number of
Facebook friends was 342 (in 2013) or 338 (in 2014). Dunbar notes that these num-
bers are the result of the age distribution on Facebook: the users of Facebook are
(or, at least, they were few years ago, before Instagram and Snapchat) younger than
the average population, and young individuals tend to have more friends and con-
nections than older ones. This is correct, but it is also possible that the average net-
work size of Facebook has indeed grown through the years, both because of the
continuous explicit efforts from the company (like the above mentioned “People
You May Know” feature) or, more interestingly, because we developed different
ways to use social media that go beyond the social bonds implied by Dunbar’s
number (more on this later).12
Overall, leaving aside the simplistic idea that there is a miraculous number
of 150 individuals that constraints all our social interactions, online and offline,
it is sensible that various factors, such as time or cognitive constraints, limit the
number of meaningful social relationships we can be involved with. It is also rea-
sonable that we can describe the social networks we are part of as having a layered
structure, with sizes limited at each level by these constraints. Likewise, our digital
social networks are influenced by time or cognitive constraints, but we need to dig
further to understand their relationship with offline social networks.
8  A growing network for cultural transmission

The global village

There are other aspects through which digital media influence the access to cul-
tural information. Even if our online social networks are not bigger than the offline
ones, we can choose their members free from spatial constraints. The hypothesis
that media would render geography unimportant for social interactions predates
the digital age (McLuhan’s notion of global village goes back to the last century, the
1960s) and, as we will see in a moment, is not completely accurate. Nonetheless,
digital media did greatly increase the ease of maintaining long-​distance social re-
lationships of various kinds (as an Italian married to a French woman, living in
England, with a job in the Netherlands, I may personally know something about
that), and, importantly from the cultural evolution perspective, of forming com-
munities purely based on interests.13
Space did not become completely irrelevant. The majority of our contacts in
digital social networks live close to us. The distribution of Facebook contacts is
inversely correlated to their geographical distance: many of your Facebook friends
live in your city, if not in your neighborhood. The same happens for email: we write
more emails to people that are spatially close to us. The majority of our phone calls
are to people in our surroundings. Analyzing a mobile phone call dataset from
2007 that contains calls between more than 30 millions users, researchers found a
negative correlation between calls and distance: “emotionally closer friends” also
tend to live closer. In sum, we use mobile phones more to supplement our local,
offline, connections than to keep in contact with people geographically distant.14
These results rightly cast doubts on the idea—​never really put forward by
anyone, I suppose—​that geography would not matter at all in the digital age. After
all, digital communications can be used to enhance and support our face-​to-​face
interactions. Our offline social networks have a strong influence on what our on-
line social networks look like. We write email after email to our colleagues, we
phone our friends in the neighborhood, and our busiest social media contacts are
the same people we meet in the workplace or at school. However, we can, also and
with ease, keep in contact with other individuals that we do not meet regularly
face-​to-​face.
A recent study checked this intuition. The researchers analyzed 11 populations,
possibly culturally diverse (Afghanistan, France, Netherlands, Japan, Rwanda,
Singapore, South Korea, Turkey, United Kingdom, United States, and a “large
western country” for which mobile phone data agreement prohibited public dis-
closure) and different media, such as mobile phones and Twitter. They calculated
the “tie range” between individuals as the “second-​shortest path between two
nodes. To illustrate this, imagine that Alice and Bob follow each other on Twitter.
The first-​shortest path between the two is the path that directly connects them. The
second-​shortest path is how we can connect, in the same network, Alice and Bob
without passing from that direct link. If both Alice and Bob follow Charlie, and he
 Cultural transmission and social networks  9

follows back them, their tie range will be two. But imagine now Alice and Donna.
They are old friends from their hometown, or they share a specific, specialized,
interest—​say they are both enthusiasts of Italian neorealist movies—​and they do
not have any other contact in common: there is no Charlie connecting them. How
many links do we need to pass through to go from one to the other? This number
will be their tie range. (You may have heard of the “six degree of separation” idea,
that is, everybody is at a maximum of six connections away from each other: the
logic is the same).
The researchers measured the strength of the connections as the total volume of
the phone calls and the number of Twitter direct mentions. Not surprisingly, they
found that short-​range ties are also the strongest ones. The Alices and the Bobs
contact themselves with the highest frequency. In line with what we said earlier, the
strength of the connections decreases when the tie range increases, but only up to
a point. Above tie range four, the strength started to increase again. Thus, we have
intense relationships with individuals who are close to us, or with whom we have
many connections (the two characteristics tend to go together), but also with few
individuals with whom we do not share other connections, like the Alices and the
Donnas of the previous example. In general, the number of connections depends
on distance: there are more short-​range than long-​range connections—​one tends
to befriend people who are in physical proximity to them—​but there is a U-​shaped
relationship between distance and the intensity of the contacts.15
A similar suggestion comes again from Twitter data, but with a different analysis
that focused explicitly on geographical distance. With data from six million user
profiles, researchers found that the spatial distribution of Twitter contacts (how
far from you are the people you follow, the people that follow you, the users you
interact with on Twitter through answers, mentions, retweets, and so on) is a bi-
modal distribution, a distribution with two distinct peaks. One peak is at around
ten miles from the user considered: these are, again, the friends and contacts phys-
ically close to you, the people leaving in or nearby the same town and who you
meet on a daily or weekly basis. However, the second peak is at around 2,500 miles.
What is this about? These are contacts who live far away, perhaps friends or col-
leagues with whom you can keep in contact through digital media, but also people
you never met in real life, such as celebrities, politicians, important figures in your
professional field. On Twitter, in fact, they not need to be people at all, but they are
often news organizations, brands, sport teams: everything related to what you may
be interested in.16

Cultural transmission and social networks

There is an interesting idea here. If social media like Facebook were originally used
to keep in contact with friends—​the people we meet down in the street and with
10  A growing network for cultural transmission

whom we shake hands or call once in a while—​now they are also used to exchange
information with people we will definitely not meet up with for their birthday.
It is telling that Facebook recently lamented what they called “context collapse.”
Facebook users are sharing less about themselves and posting more generalist
content, from memes to news or, for example, in my peculiar corner of the social
media, scientific articles. Some speculate that “On This Day,” a feature that pro-
poses to share “memories,” that is, old posts and pictures shared in the same day in
the previous years, was introduced to counteract this trend.
However, one should indeed expect this effect if social media are used, also,
to gather information from individuals who are not in the circle of the people
with whom we interact in everyday life. This type of usage is more evident in
platforms such as Twitter that from the beginning was considered, or adver-
tised, not as a friends’ social media, but as the place to “follow your interests” or
“discover what’s happening right now, anywhere in the world.” Recent estimates
suggest that the average number of Twitter followers is around 700. Given that
there is no upper limit to the number of people one can be followed by, the dis-
tribution of followers on Twitter is even more skewed than the distribution of
friends on Facebook (Katy Perry has 107 million followers as of February 2019),
thus this average may not be particularly representative. However, even ignoring
all the celebrities with more than 100 thousand followers, the average user has
more than 450 followers. Whichever way you look at it, is far from 150 plus or
minus 50.17
Overall, from a cultural evolution perspective, what matters is not the number of
friends or close, meaningful, contacts we have, but the number of possible sources
from which we can extract information. It is possible that the digital age did not
change the former, but it did, certainly, change the latter. When Robin Dunbar and
colleagues describe our social networks, they have in mind “people whom we make
an effort to maintain contact with, and to whom we feel an emotional bond,” “long-​
lasting” relationships that “involve close attention to the partner” as well as “trust
and obligation, combined with a willingness to act prosocially.” You may be ex-
cused if that is not how you feel about the people you follow on Twitter. The close
contacts are not, however, the only people we learn from in the digital age. Dunbar
may be right: it is not that in the digital age the structure of our social networks
changed. What matters, from our perspective, is that cultural transmission grew
more and more independent from that structure.18
A social media such as Facebook might be thought of as a hybrid between a
circle of friends and a network for exchange of information, and different people
use it in different ways. Others, like Twitter, as discussed above, look more like
a network for exchange of information: less spatial constraints, more contacts,
more (when successful) exchange of generalist information. But more than so-
cial media, think about Wikipedia or, even better, if you have had the occasion to
 Cultural transmission and social networks  11

write a line of code, GitHub or Stack Overflow. GitHub is a repository for open
source code. In 2017, it hosted 57 million projects. If you need to write a pro-
gram, it is highly likely that someone else already solved a similar problem, and
you can find it on GitHub. Stack Overflow is a website that features questions and
answers on various topics related to computer programming: the 10,000,000th
question was asked in 2015. Virtually any question concerning programming
can be found there and, more often than not, a usable answer to it, or a snippet
of code to directly copy-​and-​paste into your project. The cover of a (fake) book
called “Essential copying and pasting from Stack Overflow” is a widespread
internet meme, but, in fact, many introductory programming courses include
now—​and for very good reasons—​classes on how to search and find reliable code
online.19
This is, of course, the essence of cultural transmission: there is no need to re-
invent the wheel. If a problem is difficult or impossible to solve by yourself, try and
see if someone else already did it. Incidentally, digital microcosms reproduce typ-
ical features of offline cultural evolution. For example, as we will see at length in the
next chapter, it is important that individuals do not always blindly copy others: they
also need, from time to time, to experiment by itself. However, in Stack Overflow,
many tend to merely copy-​and-​paste the fragments of code they need from the
answers in the website. While the final code can work overall, it has been shown
that security issues arise due to the difficulty of adapting the same code to different
problems. Researchers analyzed 1.3 million Android applications, and they found
that more than 15 percent of them contained security-​related code snippets from
Stack Overflow. Practically all of these (97.9%) included some parts of code that
were estimated as insecure.20
A different angle to look at the same phenomenon is specialization. Wide ac-
cess to cultural information means that whatever topic interests you, you can
find someone else who is also involved in it. You may have used GitHub or Stack
Overflow, but there are also large communities for people that are into esoteric
programming languages, that is, languages “designed to experiment with weird
ideas, to be hard to program in, or as a joke, rather than for practical use.” These
include Whitespace (a language that uses only whitespace characters: space, tab,
and return) or Grass (a language that uses only the characters W, w, and v, so that
“programs in Grass are said to look like ASCII art of grass”). Online communi-
ties form around less geeky, although specialized, interests. Researchers mapped
the vast world of Reddit, isolating more than 15,000 subreddits, topic-​specific
forums in which the entries are organized. Topics include swords, calligraphy,
transhumanism, historical costuming and so on—​I am picking randomly those
for which I  understand the meaning:  others are “RTLSDR (the low-​cost soft-
ware defined radio (SDR) community”) or “MS3TK (a place for fans of Mystery
Science Theater 3000).”21
12  A growing network for cultural transmission

Information donors

An important, and often overlooked, aspect that makes possible the growth of the
networks in which cultural transmission happens is that we, as humans, seem to
love to share information. Kevin Kelly, co-​founder of the magazine Wired when
I probably did not even know about the existence of the internet, recently wrote:

We all missed the big story. Neither old ABC nor startup Yahoo! created the con-
tent for 5,000 web channels. Instead billions of users created the content for all the
other users. There weren’t 5,000 channels but 500 million channels, all customer
generated. The disruption ABC could not imagine was that this “internet stuff ”
enabled the formerly dismissed passive consumers to become active creators.

Kelly continues:

Users do most of the work—​they photograph, they catalog, they post, and they
market their own sales. And they police themselves; while the sites do call in
the authorities to arrest serial abusers, the chief method of ensuring fairness is a
system of user-​generated ratings. Three billion feedback comments can work won-
ders. What we all failed to see was how much of this brave new online world would
be manufactured by users, not big institutions. The entirety of the content offered
by Facebook, YouTube, Instagram, and Twitter is not created by their staff, but
by their audience. Amazon’s rise was a surprise not because it became an “every-
thing store” (not hard to imagine), but because Amazon’s customers (me and you)
rushed to write the reviews that made the site’s long-​tail selection usable.22

Leaving aside the possibly overoptimistic position, especially about the self-​
regulation of the digital world (as I write, the necessity for companies like Facebook
and similar to filter and, in fact, censor, users’ posts is one of the most debated topic
in the news), these reminiscences are interesting because they emphasize how sur-
prising the shift from “passive consumers” to “active creators” was in the earlier
years of Internet, even for insiders such as Kelly.
From a cultural evolution perspective this behavior is surprising indeed. As we
will explore more in detail later on, cultural evolutionists think of social learners
as information scroungers, as opposed to individual learners, who are information
producers. With a predominance of scroungers, like the users of Stack Overflow
that copy-​and-​paste the snippet of code they need and do not contribute with their
own work, the system will simply not work. But this is not what generally hap-
pens. On the contrary, individuals everywhere seem to be happy to produce con-
tent, and share their knowledge, whether in form of code, reviews of restaurants
and Amazon products, or recipes of whatever you can imagine, for apparently no
gain.23
 Information donors  13

Exchanging information in social media can provide, up to a point, direct re-

wards. First, we may personally know many of the members of our digital social
networks and they could be able to reciprocate our altruistic sharing. If I comment
on a Facebook post from a friend about, say, where to have a nice dinner in Bristol,
they may be more willing to respond to later requests from me. If I retweet a col-
league looking for a new job, I expect they will be more likely to retweet me in the
future. Second, even when there is not the opportunity for a direct exchange, your
sharing activity can be useful to enhance your reputation. I was once invited to
give a quite informal lab seminar where I was introduced, first thing, as a person
who is worth following on Twitter by people interested in cultural evolution. I was
slightly upset but in fact, having a curated Twitter feed, sharing—​at least in my
circle—​interesting and on-​topic papers and news, is an effective way to get known
and potentially valued.
In bigger networks direct rewards are less important, or not present at all. Many
online communities like Stack Overflow implement explicit reputation points that
can be gained or lost according to the activity of the users. Answers voted up give
to the respondent ten reputation points, answer marked as “accepted” give 15, and
so on. Reputation points can also be lost: two points are lost, for example, if an
answer is voted down. Interestingly, voting down an answer costs one reputation
point to the voter, making negative feedback costly. Users can also place “bounties”
on questions. One can draw attention to a question that is not answered by placing
a bounty on it, that is, an arbitrary number of reputation points, that the “owner” of
the question will transfer to the user that will provide a correct answer. Reputation
points provide explicit advantages, or “privileges.” Users need reputation points,
for example, to be able to vote up (15 reputation points) and down (125) other
questions and answers. The current highest level (25,000 reputation points) allows
users access to site analytics. Practices like Stack Overflow’s reputation points—​
often referred to as gamification —​provide a seemingly effective way to encourage
engagement and, from our perspective, voluntary sharing of information.24
But that’s clearly not the whole story. Hundreds of thousands of videos are up-
loaded everyday on YouTube, in absence of any system of reward or reputation and
without any obvious direct return for the uploaders. Or consider Wikipedia. The
online encyclopedia, one of the most studied cases, does have a reputation system,
consisting in a sufficiently intricate—​at least for outsiders—​scheme of awards that
can be conferred to contributors. There is the “seldom awarded” Order of the day,
bestowed personally by Jimmy Wales, the co-​founder of Wikipedia, for “excep-
tional service to the community” to only five contributors so far (including Justin
Knapp, the first person to ever make 1,000,000 edits on Wikipedia, at the age of
thirty). Awards go down to the relatively common Barnstars, given for a variety of
general or more specific reasons (going from the Rosetta Barnstar for “outstanding
translation efforts” to the Recent changes Barnstar, for users “patrolling recent
changes,” and many others). However, in contrast to the Stack Overflow system,
14  A growing network for cultural transmission

the number of contributors that can aspire to receive one of these awards, including
Barnstars, is rather limited, especially compared with the total number of people
that contribute to editing Wikipedia. Also, a substantial amount of contributions,
estimated around one third is anonymous, so that by definition, they cannot have
any return in terms of reputation.25
Interestingly, the contributions of anonymous users are not, on average, of poor
quality. Analyzing a random sample of more than 7,000 contributors of French
and Dutch language Wikipedia, researchers found that anonymous users, they
call them the “Good Samaritans,” who edit Wikipedia pages without registering
and can be tracked only by their IP number, are at least as good as the edits of the
“Zealots,” reputationally-​motivated registered users. The researchers calculated the
“retention rate” of users, basically the proportion of characters of their contribu-
tions retained across all edits. We can understand this through an invented simple
case. if I add in a Wikipedia page the sentence: “There have been a number of dif-
ferent approaches to the study of cultural evolution,” retention rate will be 0 per-
cent if the entire sentence is deleted in the current version of the page, 100 percent
if it remained untouched, and, for example, 74 percent if the current version reads
“There have been various approaches to the study of cultural evolution” (51 charac-
ters retained over a total of 69).
The average retention rate calculated, in real Wikipedia pages, was 72 percent.
The main result was that non-​registered users had an average 74 percent retention
rate, higher than the retention rate of registered users, at 70 percent. This result,
however, could be skewed by the fact that many “Good Samaritans” contributions
are likely to be low-​cost amendments, such as the extemporary correction of typos,
for which registering would be indeed a waste of time: such changes are dispropor-
tionally likely to obtain a 100 percent retention rate. Still, when considering users
who make more than one edit, and who are thus unlikely not to be making only
typo corrections, the retention rate of non-​registered users continues to be better
on average. Zealots start to be better than Good Samaritans only for users who
made around forty edits. Of course, not all anonymous users are Good Samaritans.
Vandals do exist and anonymous users are more likely to vandalize articles in re-
spect to registered users (the contrary would be very surprising indeed), but the
association between anonymity and vandalism is not robust. In addition, and
importantly, the proportion of malicious edits in Wikipedia—​such as deleting
all content of a page, inserting vulgarities or text unrelated to the content—​is
generally low.26

Why do we edit Wikipedia?

In sum, why do we edit Wikipedia anonymously, and carefully? Why do we pro-

vide YouTube tutorials going from how to play blues guitar to how to tie a bow, or
 Why do we edit Wikipedia?  15

almost everything else you can imagine? Explanations along the lines of “because
it makes us feel good” are right, in a sense. However, they just move the target: in
evolutionary jargon they are proximate explanations, but we want to understand
the ultimate ones: why does it make us feel good?
There are a few suggestions that can shed light on this tendency. First, we may
act as if we are sharing information in the habitual population where this behavior
does indeed provide direct returns in terms of reciprocity or reputation. In many
cases, in fact, it could provide it. As we will discuss later, the popularity of YouTube
videos, as of many other cultural products, is strongly skewed, with very few videos
becoming popular, and the vast majority cumulating only a few views. The prob-
ability for a video I will upload to become successful is very low, but still not null.
I may hope that my tutorial on blues guitar will gain traction, and it will make me
the next YouTube celebrity or, at least, it will give me some credits among a small
circle of musicians.
Anonymous Wikipedia edits, as well as many small contributions to, for ex-
ample, Stack Overflow or similar, do not provide this prospect. We may still, how-
ever, gain reputation in our small offline networks through sharing afterwards the
knowledge we have obtained, or simply by reporting our activity. In other words,
while we edit Wikipedia pages anonymously, we can non-​anonymously tell it
to our friends, show them what we did, and explain to them how to do it them-
selves. Also, “Good Samaritans” contributions could be the result of experts’ edits,
people knowledgeable on a specific subject, be it an academic subfield or a little
known Japanese manga. Although not personally interested in reputation within
Wikipedia, such people may be personally interested in a fair and accurate repre-
sentation of their area of expertise in the online encyclopedia.
In addition, we may simply not be well equipped to process completely an-
onymous interactions, and, as a consequence, we act instead as if they would
produce reputational gains. Anonymous interactions are associated to modern
societies, especially where writing is widespread. In small, oral, societies, the oc-
casions for anonymous interactions are limited and, interestingly, they mainly in-
volve negative events, such as murders, ambushes, or accusations of wrongdoings,
such as witchcraft. This is not to say that positive anonymous interactions do not
exist at all in those societies, but that their occurrence is much less than what is
possible today. In this perspective, when anonymously editing Wikipedia, or an-
swering anonymously a stranger’s question on Stack Overflow, we would act as if
our interaction would produce reputation gains.27
Many have interpreted the surprisingly high levels of cooperation in eco-
nomic games—​surprisingly high, that is, if we subscribe to the Homo economicus
portrayal—​simply as a consequence of framing effects of the experimental condi-
tions, and in particular the difficulty of processing anonymous interactions. Here
is an example. One of these games is known as the “Dictator Game:” the experi-
menter gives you an amount of money, say 10 pounds, without asking anything in
16  A growing network for cultural transmission

exchange. You have to choose whether to give some share of this money to another
participant you do not know, and with whom you will not interact in the future.
What would you do? Do you give some money? Do you give all? Nothing? For the
Homo economicus description, according to which we are rational, self-​interested
agents, the Dictator Game is a no-​brainer. We should thank the experimenter and
leave with our money, without giving anything to the stranger.
However, experiment after experiment made clear that this is not what we do.
The distribution of sharing among American university students is bimodal: there
are as many people giving half of their money to the stranger as people not giving
anything. This is not an idiosyncrasy. The Orma people, semi-​nomadic shepherds
from Kenya do the same, but their lower mean is not zero, but around 20 percent
of their share. In an experiment among the Tsimané, indigenous people of Bolivia,
nobody chose not to share anything with the stranger, and the mean was around
30 percent.28
Again, what would you do? Perhaps the participants are uncertain of whether
or not the experimenter will reveal in the end that the stranger is someone they
know, or even a friend or a familiar. Perhaps the experimenter is just sorting people
as greedy or generous, and the generous ones will then receive a bigger prize. The
experimenter is explicit that the interactions are completely anonymous and there
will not be consequences, but how can you be sure?
Imagine a real life situation: you find 10 pounds in the street and pick it up.
Nobody sees you. After a few minutes you come across a passer-​by, who is un-
known to you. Do you give some money to them? Now the question seems easy to
answer. Experiments attempting to mimic more realistic conditions suggest that, if
anonymity is perceived as genuine, we become less generous. Researchers played
the same Dictator game, this time “at bus stops within one block of a major ca-
sino” in Las Vegas. Participants were just people waiting at the stop, and who were
informed only afterwards of having being the subjects of an experiment. A con-
federate sat at the stop, and then left for a while talking on his phone, at some dis-
tance and facing away from the participant (the idea is that the confederate would
not be aware of what would happen in the meantime). Confederate 2 then passed
by hurriedly, also talking on the phone and he pretended to remember having a
few spare casino chips in his pocket. He told the participants that, as he was late
for the airport and did not have time to cash the chips in the casino, they could
have them (the total amount, in chips, was 20 dollars). Confederate 2 left and, after
a moment, confederate 1 returned to the bus stop, supposedly unaware of what
had happened. The experimenters wanted to test how much the participants gave
to confederate 1, who for them was a stranger at a bus stop. All participants gave
the same amount: zero dollars. There was also a slightly different condition, where
confederate Accomplice 2, while offering the chips, also mentioned “I don’t know,
you can split it with that guy however you want” gesturing towards confederate
 Effective cultural population size  17

1. How much did the 30 participants in this condition give? Well, exactly the same
as above: zero.29
Is it possible that when we provide information online, free and anonymously,
we are (also) behaving as in the laboratory economic game experiments? We do
not need—​both in the lab and on internet—​to actually think that the interactions
are not anonymous, but we act as if they were not. Of course the mechanism is not
perfect (Wikipedia vandals do exist, and editors have an important role in keeping
the encyclopedia functioning), but it produces, together with other factors, an on-
line environment where the transmission of information is sufficiently reliable.
Together with a motivation for indirect, or possibly misplaced, reputational
gains, another important factor is that the cost associated with sharing information
is, in most cases, limited. As mentioned, some of the Wikipedia Good Samaritans’
contributions are typo corrections, for which registering would possibly be more
costly than the correction itself. In addition, if the cost is very low, even if the
probability of a future gain is very low, as in the case of YouTube videos, it may be
worth trying. Big costs only happen when the scroungers (individuals who only
learn from others and do not produce new information, as discussed above) gain
a particular advantage over us. However, helping an unknown person on Stack
Overflow, sharing my codes on GitHub, or making available information on my
special lasagne recipe does not present that risk. Sensitive, secret, information
is generally not shared, and when this happens it is often because of some error.
On the contrary, the cost of accepting information outside of the close network of
known, trusted people, may be more relevant, which explains, as we will see in de-
tail in the next chapters, why individuals are not that easy to persuade and deceive.

Effective cultural population size

Cultural evolutionists, drawing inspiration from population genetics, use the con-
cept of effective cultural population size to indicate the number of people who
share cultural information, and thus can be involved in the transmission of a
cultural trait. While there is uncertainty regarding the exact formal definition of
effective cultural population size, it suffices for our interest to distinguish two pos-
sible meanings of it. Effective cultural population size can indicate the number of
people actually involved in the transmission of a cultural trait, all the individuals
who, for example, are interested in historical costuming and are part of a transmis-
sion chain that involves the exchange of information about historical costuming.
Alternatively, effective cultural population size can indicate all the individuals one
can interact with, that is, the number of people who are potentially involved in the
transmission of a cultural trait. Both differ from census population size, the total
number of individuals in a population.30
18  A growing network for cultural transmission

Imagine you want to track the effect of population size for the spread of a cul-
tural trait, say a new cooking technique, in two societies, the “connected” and the
“isolated.” Both societies have the same census population size, 10,000 people, and
in both cases these persons live in 10 villages of 1,000 inhabitants each. The con-
nected, however, travel frequently from one village to another for trade and they
easily exchange information, whereas the isolated seldom travel, and when this hap-
pens their relations are hostile. The potential effective cultural population size of
the connected is much higher than those of the isolated and, as a consequence, the
actual effective cultural population size—​individuals directly involved in using the
new cooking technique, and talking about it—​will be too. Intuitively we would ex-
pect the cooking technique to spread faster among the connected. Not only that, but
among the isolated, only a few people will ever enter into contact with the new trait,
and there will be more risk for the trait to be forgotten, as not many people will know
or use it. Notice that effective cultural population size can also be bigger than census
population size if a society is involved in extensive exchanges with other societies.
Given these complications, it is not surprising that estimates of effective cul-
tural population size, besides controlled experimental conditions, have been
generally approximate (speaking of which, the precise quantitative data coming
from digital communities could represent a critical addition for research testing
hypothesis on the interactions of demography and cultural evolution). Examples
include estimates between three and 20 individuals involved in pottery in villages
in pre-​hispanic US Southwest, amounting to around 50 decorative pottery mo-
tifs, or ten to 52 in Neolithic settlements, amounting to a few hundreds of motifs.
Anthropologist Joe Henrich, in an influential paper on the relationships between
cultural complexity and population size, estimated 800 individuals involved in
the cultural transmission of technological tools in Tasmania at the time of the
European contact. Another research used a rough estimated of the effective popu-
lation size in late Pleistocene Europe of 3,000 individuals. Other works took a prag-
matic approach and used census population size as a proxy for effective population
size, with estimates going from a few hundred to 7,000 for North-​American for-
agers, or a few hundred thousand for Pacific Islands.31
When thinking in these terms, effective cultural population size has risen
steeply in the digital age, whatever domain we consider. No doubt this has been a
continuous process, lasting centuries and involving innovations such as writing,
printing, analogic media like radio or television, or the development of inter-
national trade and leisure travels. However, the easiness and convenience of
finding and transmitting information we can experience today are unprecedented.
Only one hundred years ago, if you were interested in historical costuming your
potential effective cultural population size would have been limited to your direct
connections, perhaps the few thousand individual Dunbar estimates we can “put a
face on.” Some of them could have pointed you to other people, or you might have
 Effective cultural population size  19

found books on historical costuming. The majority of people, in fact, could simply
not have been interested in historical costuming whatsoever, if not with important
investments of time and money, for example, travelling where other people could
provide information on it. The actual effective cultural population size you could
have access to would have possibly been equal to zero. Nowadays, potential ef-
fective cultural population sizes are in terms of millions or billions, and converging
with total census population sizes (when everyone is connected to the internet),
and this keeps actual effective cultural population sizes above zero for the majority
of people.

Let me now summarize where we have arrived so far. The digital age may not
have changed drastically, at least not yet, our relations with friends, families, and
other close contacts. However, cultural transmission became independent from
these trusted networks. The number of individuals with whom we can exchange
information—​the potential effective cultural population size—​is measured in bil-
lions, and they are reachable at a negligible cost and instantaneously. We are also
willing to spontaneously share information on these networks, without neces-
sarily being reciprocated and without any obvious return in terms of reputation.
Specialized digital communities or massive collaborative projects are all signs of
this change. In the last chapters of this book, we will ponder on what are the general
consequences of this increased access to cultural information, and we will specu-
late on possible long-​term trends. For now, however, it is interesting to consider
how this hyper-​availability, together with new opportunities, poses new problems
related to cultural evolution. One is that, when we can copy from everybody, it may
become more difficult to decide if and when we should do so.

Notes

1. https://​www.statista.com/​statistics/​268136/​top-​15-​countries-​based-​on-​number-​of-​
facebook-​users/​
2. Dunbar (1998)
3. Dunbar (1993)
4. Dunbar and Sosis (2018)
5. https://​www.thelocal.se/​20070723/​7972
6. Zhou et al. (2005); Hamilton et al. (2007); Dunbar (2018)
7. Burton-​Chellew and Dunbar (2015)
8. https://​chrome.google.com/​webstore/​detail/​birthday-​buddy/​ciljodcgjplloiacmjbngige
ihcgdheb?hl=en
9. Wilson et al. (2012)
10. Gonçalves et al. (2011); Dunbar et al. (2015), Dunbar (2016)
11. Fuchs et al. (2014)
20  A growing network for cultural transmission

12. http://​blog.stephenwolfram.com/​2013/​04/​d ata-​s cience-​of-​t he-​facebook-​world/​;

http://​www.pewresearch.org/​fact-​tank/​2014/​02/​03/​what-​people-​like-​dislike-​about-​
facebook/​, Dunbar (2018)
13. McLuhan (1964)
14. Goldenberg and Levy (2009), Jo et al. (2014)
15. Park et al. (2018)
16. McGee et al. (2011)
17. https://​k ickfactory.com/​blog/​average-​t witter-​followers-​updated-​2 016/​, https://​
en.wikipedia.org/​wiki/​List_​of_​most-​followed_​Twitter_​accounts
18. Dunbar (2018)
19. https://​github.com/​blog/​2345-​celebrating-​nine-​years-​of-​github-​with-​an-​anniversary-​
sale, https://​meta.stackoverflow.com/​questions/​302884/​10-​000-​000th-​question-​is-​
here, https://​memecollection.net/​copying-​and-​pasting-​from-​stack-​overflow/​
20. Fischer et al. (2017)
21. https://​en.wikipedia.org/​wiki/​Esoteric_​programming_​language, Olson and Neal (2015)
22. Kelly (2017)
23. Laland (2004)
24. https://​stackoverflow.com/​help/​whats-​reputation, Hamari et al.  (2014)
25. https://​en.wikipedia.org/​w iki/​Wikipedia:Awards, https://​e n.wikipedia.org/​w iki/​
Wikipedia:IPs_​are_​human_​too
26. Anthony et al. (2009), Viégas et al. (2004)
27. Hagen, E.H. and Hammerstein (2006), Boyer (2018)
28. Henrich et al. (2001)
29. Winking and Mizer (2013)
30. Henrich (2004)
31. Estimates from: Kohler et al. (2004); Shennan and Wilkinson (2001); Henrich (2004);
Powell et al. (2009); Henrich et al. (2016); Kline and Boyd (2010)
 2
Wary learners

Big data and big theory

Evolutionary and cognitive perspectives to culture have a long history, but a rec-
ognizable discipline, which has been initially developed by a relatively small group
of anthropologists, psychologists, and biologists, has started to emerge in the last
thirty years. It is now expanding (at least I am betting on it) inside the whole social
and human sciences. Why should one care about cultural evolution in the digital
age? This book would ultimately aim to provide an answer to this question, but, for
now, let me at least give you a few hints.1
The first is that cultural evolution provides both a quantitative methodology and
a solid theoretical framework for the study of culture. While details vary, and they
will be discussed at length in the next chapters, there is also a broad consensus
on some issues. Cultural evolutionists, unlike other scholars interested in human
cultural dynamics, have a taste for quantitative explanations, models, data collec-
tion, and experiments. Basically, they believe that the study of culture should be a
strictly scientific enterprise, and it should be conducted with the methodologies
developed for all other sciences.
This quantitative focus fits perfectly with the kind of data the digital age pro-
duces. We have now access to what, until a few years ago, would have been an in-
conceivable amount of data, especially for the social and human sciences. Most
of what we do leaves footprints that can be quantified and analyzed whether we
want it to or not. For a cultural evolutionist there is a bright side to this: we have
never, ever, had so abundant—​and readily accessible—​information on human be-
havior. Cultural evolution provides a privileged perspective to make sense of this
information.
Still, one needs to know what to do with these data. An article, appeared in
2011 in the prestigious journal Science, used data from Twitter to investigate
individual-​level mood changes in humans. Among the findings, the researchers
report that individuals tend to wake up “in a good mood that deteriorates as the
day progresses,” and that “people are happier on weekends, but the morning peak
in positive affect is delayed by 2 hours, which suggests that people awaken later.”
Another more recent study, using “over three and a half billion social media
posts from tens of millions of individuals from both Facebook and Twitter be-
tween 2009 and 2016” found that people are happier when weather conditions

Cultural Evolution in the Digital Age, Alberto Acerbi. Oxford University Press (2020) © Oxford
University Press.
DOI: 10.1093/oso/9780198835943.001.0001
22  Wary learners

are optimal then when they are not: “cold temperatures, hot temperatures, pre-
cipitation, narrower daily temperature ranges, humidity, and cloud cover are all
associated with worsened expressions of sentiment.” In sum, big data allowed
scientists to discover we sleep more at the weekend, and a good sunny day puts a
smile on our faces.2
Don’t get me wrong: these studies are useful. On the one hand, it is important
to back up anecdotal evidence with proper experiments or quantitative data. Is it
really true we are happier when is sunny or is it just a popular misconception? Do
people sleep more at the weekend or perhaps the sleep cycle of the working week-
days influence our Sunday mornings too? On the other hand, these results provide
a proof of concept. If we can extract meaningful signals from digital data about
patterns we already suspect, or know exist, such as people’s mood being influenced
by the weather, we can be more confident that surprising signals will be reliable too.
That being said, to interpret these signals and even more importantly, to decide on
which signals to look for, we need good theories.
Many computer scientists and physicists in recent years have put to use their
quantitative training and skills to analyze digital data on human behavior. This
tradition has produced many outstanding works—​some of which will be men-
tioned and discussed in the following pages—​and it is far from my intention to
undermine their importance. However, cultural evolution can provide a missing
theoretical background. Albert Einstein is quoted as having said that “everything
should be made as simple as possible, but not simpler.” (Did Einstein really say
that? In the chapter on Prestige I will discuss an experiment in which we evaluated
whether people preferred quotes associated with a famous author, even if these
quotes were associated with them at random.) Considering individuals as “social
atoms” or transmission of ideas as “contagion” may be useful, but it is important to
develop models drawing on solid theories. Cultural evolution provides, together
with the above-​mentioned methodological attitude to modeling and quantitative
analysis, a sophisticated view of human behavior. Cultural evolutionists are com-
mitted to developing hypotheses that are grounded in evolutionary theory and
cognitive science: big data and big theory.

Population thinking

When thinking about Darwin’s legacy, the “survival of the fittest” (an expression,
by the way, never used by Darwin, but coined by Herbert Spencer) generally comes
first to mind. Ernst Mayr, one of the most important evolutionary biologists of last
century, would disagree. According to Mayr, one of the greatest, and overlooked,
contributions of Charles Darwin to biology was population thinking. Before the
nineteenth century, species were considered as immutable entities, and individ-
uals were instantiations, possibly imperfect, of these entities. Not too dissimilar
 Population thinking  23

to the concept of platonic ideas, the pigeons I see from my window in Bristol are,
from this perspective, all expressions of an ideal type (the species Columba livia) of
which they are examples. Biology was the study of these abstract entities through
observations of their mundane counterparts.3
In the nineteenth century, through the work of Darwin and others naturalists,
the idea that the proper objects of study of biology were the individuals started to
take hold. After all, the pigeons out there are flying, eating, copulating, and even-
tually dying: not much action happens around the Columbia livia ideal type. This
may not seem too exciting but think again about the survival of the fittest. To even
formulate the hypothesis of natural selection, one has to move the level of analysis
to individual interactions, to the details of survival, reproduction, and mutation.
This does not mean that it is wrong to generalize about the behavior of Columbia
livia, but population thinking implies that individual interactions and differences
are real and causally important and that they explain population-​level properties,
and not vice versa.
The importance of population thinking is uncontroversial among cultural evo-
lutionists, but what does it mean to apply this framework to culture?4 Let us take
as a concrete example inspired by the anthropologist Dan Sperber, the tale of Little
Red Riding Hood. Little Red Riding Hood seems a legitimate object of study for
cultural evolution. But what are we studying exactly? Is our object the canonical
version of Grimm’s brothers? Is another one of the more than fifty versions recog-
nized in folklore indexes? Or perhaps all of them? What about the thousands of
YouTube videos, or the dozens of recent movies, inspired by the tale? What about
Carnival costumes, internet memes, and nursery rhymes? What about The Tiger
Grandmother, a tale from East Asia, in which the protagonist has to deal with a
tiger disguised as grandmother?5
Jusy as in the case of the biological species Columbia livia, the cultural trait
Little Red Riding Hood is an abstraction. For practical reasons we can, and
should, cut its borders (“I will not consider movies”) to make it more manage-
able, but the important thing is that Little Red Riding Hood, as any other cultural
traits, is the population of events in which the story is mentioned (me telling
it to my daughter), the population of artefacts that contains it (the books, the
YouTube videos), and everything else that is linked in chains of social transmis-
sion, including possibly these same paragraphs you are reading. As cultural evo-
lutionists would say, a cultural trait is a statistical aggregate of individual-​level
interactions.
We will see later how thinking in this way about culture makes it easier to grasp
some possibly counterintuitive ideas, such that cultural traits are fuzzy objects,
with arbitrary edges (as just said for Little Red Riding Hood) and, indeed, culture
itself is one of these fuzzy objects. Or, we should not assign causal properties to
“culture,” because what is casually active are individual interactions and physical
artefacts, as we said for Columbia livia. As Dan Sperber puts it:
24  Wary learners

what caused the child’s enjoyable fear was not the story of “Little Red Riding
Hood” in the abstract, but her understanding of her mother’s words [ . . . ] What
caused the story of “Little Red Riding Hood” to become cultural representation is
[ . . . ] the construction of million mental representations causally linked by mil-
lions of public representations.6

This focus compels developing sound hypotheses about what happens at the fine-​
grained level of individual interactions. When do we tend to copy others and when
we do not? What kind of cues we use to decide? What are the features of a specific
cultural trait that makes it successful, being present in chains that extend in time
and space? Why are some chains more robust than others?

A very social animal

Another idea that underpins all the approaches, albeit with different flavors, is that
evolution by natural selection is the theoretical framework we need to answer to
questions such as the ones posed above. Again, we will explore various details in
this and in the following chapters, but one important aspect, particularly relevant
for the study of our digital age is the idea that human cognition evolved in a social
environment and is adapted to process social information: to optimize learning
from others, to interpret communication, to detect manipulation attempts.
In many worried accounts of the negative effects of our hyper-​connected life, the
image of humans is one of isolated thinkers, who have to change their default set-
tings, so to speak, when they need to extract and assess information from others.
We are easily gullible and prone to overload when too much social information is
present. Here is a quote from a recent book (excellent in many respects) of philoso-
pher Michael Patrick Lynch:

A key challenge to living in the Internet of Us is not letting our super-​easy access
to so much information lull us into being passive receptacles for other people’s
opinion.7

In another valuable account (The Distracted Mind), human behavior, in particular

human information search and retrieval, is modeled as an individual foraging
problem, aiming to explain the costs and benefits of remaining in a current food
patch versus moving into a new one. Picture yourself going to pick mushrooms: you
find an area rich in, say, porcini. At the beginning you gather several of them but,
with time, you find fewer mushrooms. When does it make sense to move and look
in another area of the woodland? Food patches can be thought of as information
sources: when does it make sense to look for a different source of information? You
can already picture here the multitasker, short-​attention-​spanned, easily distracted
 The secret of our success  25

digital human, jumping from one side of the forest to the other without making the
most of the patches they visit:

Because of modern technology, we are no longer performing “optimally” in ac-

complishing our goal of foraging for information.8

This might, or might not, be correct, but the point is that humans are not soli-
tary foragers of information or passive receptacles that need to activate a special
modality when dealing with social cues. Humans are always part of chains of so-
cially transmitted information. The idealization of a lonely reasoner, who gets into
trouble with information from conspecifics, is not the best place to start a criticism
of digital media. A cultural evolution approach promises, if not a cautious opti-
mism, at least a starting point in which information from others is always present,
and evaluating whether it is useful or not is what we do by default.
Moreover, if we have at all a tendency to extract information from the behaviors
of others and to use them to modify our behavior, if systems of communications
evolved at all and remain stable, they should be, on average, advantageous for all
the individuals involved. It makes sense to start from a “presumption of good de-
sign,” for which our abilities to engage in communication and social learning are
functioning reasonably well: we are not too gullible and, conversely, we cannot per-
suade others too easily.
Of course, the fact that the cognitive mechanisms tracking and making use of
information are a product of evolution does not guarantee that they are efficient
now. As we saw in the previous chapter, contemporary digital age represents only
the last episode in the growth of the network in which cultural transmission po-
tentially happens. Today, many people have a cheap and immediate access to many
other people, and to the information they produced. It might be, that under these
conditions our cognitive mechanisms to process information are easily tricked,
that we are easily tricked. This is a question that is worth asking, and it is central to
this book.

The secret of our success

For the majority of cultural evolutionists, a basic premise of the theory is that cul-
ture is adaptive. Culture is, as the title of a book from anthropologist Joe Henrich
suggests, “the secret of our success.”9 Cultural evolution, in this perspective,
provides a buffer to store information that allowed humans to become the spe-
cies able to colonize different environments and to outperform bigger and phys-
ically stronger animals. From repeating Little Red Riding Hood to your children
to copying grandma’s lasagne recipe, from the piano lessons I took as a child to
the long list of references that closes this book, we have several intuitive pieces of
26  Wary learners

evidence that copying from others is the smart thing to do. However, should we
always do it? The answer is a clear “no,” but understanding why this is the case will
help us to proceed toward the discussion of the potential dangers of cultural influ-
ence in the digital age.
There are various reasons why we need to be selective. Later in the chapter, we
will consider in more detail the idea that others could be actively and consciously
misleading us. It is possible, and indeed common, that the interests of different
people are not aligned. Deniers of the role of human activity in climate change can
sincerely believe that there is not a significant increase in global temperature or
that even if there is, it is due to other causes. Alternatively, they can have a direct or
indirect interest in an oil company. Supporters of the anti-​vaccination movement
can sincerely believe that vaccines are unnecessary or even harmful. Otherwise,
they can have a chiropractic studio and they want to convince people that chiro-
practic techniques can better help prevent and treat infectious diseases.
Perhaps counterintuitively, however, social learning can be ineffective even
when interests are aligned and there is not intention to mislead others. Kevin
Laland called social learners, as mentioned above, information scroungers, as op-
posed to individual learners, who are information producers.10 Here is a simple
thought experiment: imagine a world in which there are two sources of food, red
fruits and blue fruits. Red fruits are poisonous, whereas blue fruits are edible and
nutritious. The population is composed of two kinds of individuals, “explorers”
and “copiers.” Explorers look around and try out different fruits until they find out
which one of the two is good. This is a costly process: it requires time, and perhaps
the red fruits make them sick for a while. Copiers, on the contrary, look at what
others animals are doing, and they do the same. In a population mainly composed
of explorers, copiers have an easy life. They do not pay the costs of exploration, and
they still gain reliable information on which fruits to eat. In this condition, copiers
will reproduce more than explorers, and they will become the majority.
Imagine, however, that the environment changes. Perhaps other edible red fruits
appear, or blue fruits become poisonous or they disappear because everybody ate
them. Perhaps a new species of fruits, yellow ones, begin to spread in the environ-
ment. With a population mostly composed by copiers, there will be problems, as
they will look around and copy the wrong, outdated, behavior (remember from
the previous chapter the Stack Overflow users that only copy-​and-​paste code and
end up with insecure software). The few explorers will be now advantaged. Formal
models have shown that, in this condition there is an evolutionary equilibrium in
which the population is composed of a certain optimal amount of copiers and ex-
plorers. It is important to realize that this means that the fitness of explorers and
copiers at this equilibrium is equal, which is the same as saying that social learning
is not more advantageous than individual learning.
A way to understand the logic behind this result is depicted in Figure 2.1. The
fitness of individual learners is constant, slightly below the optimal value as they
 The secret of our success  27

1.00

Individual learners

0.75
Fitness

0.50
Social learners

0.25

0.00

0.00 0.25 0.50 0.75 1.00

Frequency of social learning

Figure 2.1  The logic of Rogers’ model.

incur in the cost of exploring, whereas the fitness of social learners depends on the
frequency of social learners themselves. On the left side of the plot, the few social
learners successfully scrounge information from the many explorers, but on the
right side, there are too many social learners/​scroungers: they copy (from other
scroungers) outdated information, and their average fitness goes down. At the only
point of equilibrium—​where the two fitness lines intersect—​the fitness of social
learners is equal to the fitness of individual learners, bringing the conclusion that
social learning is unlikely to evolve, as it does not provide advantages.11
This result, dubbed “Rogers’ paradox.” from anthropologist Alan Rogers who
devised the original model, has generated a great deal of analytical and mod-
eling work in cultural evolution. Of course, as with all models, it depicts a greatly
simplified version of reality, but Rogers’ formulation is good because under-
standing what is wrong in this picture is exactly what we need to sharpen our in-
tuitions. The point is that, if the tendency to extract and use information from the
behavior of other conspecifics has evolved in the first place, something must be
different from the situation described in the imaginary world above.
First, the model presupposes that capacities of social learning and the capacities
of individual learning are two separate skills. It presupposes, in other words, that we
are switching from a zombie-​copy-​mode in which we mindlessly reproduce what
others do to an individual-​explorer-​mode in which we discard everything else that
is not what we try ourselves. This is in an unrealistic assumption. One of the solu-
tions of Rogers’ paradox suggests that information from others is useful, and that
28  Wary learners

a tendency to use it could evolve, providing that we are “critical” social learners.
Critical social learners copy others and, at the same time, evaluate the behaviors
they acquired. If it proves inadequate they revert to individual learning: they try
to scrounge, and if it does not work, they resort to explore. At equilibrium, critical
social learners have higher fitness than individual learners under a wide range of
conditions, suggesting that this form of mixed learning could evolve.12
The same point—​ individual and social learning should be considered
intertwined—​is highlighted by another problematic assumption in the model,
namely that the capacities of social learning have to evolve in a species that before-
hand was capable of individual learning. This, also, seems unlikely. The same cog-
nitive mechanisms we use to process information about the non-​social world can,
and do, process information about the behavior of our conspecifics. Whereas hu-
mans may have a stronger propensity to pay attention to what others do, in respect
to different species, there are no reasons why a species able to process non-​social
information would not be able to process a social one. I interpret the continuous
expansion of the range of species considered “cultural”—​a recent addition is fruit
flies, which copy from each other with whom to mate—​as supporting precisely this
idea: there is no reason to expect a species capable of general learning not to use
social information.13
The aspect that has been paid more attention, in Rogers’ model, is however that
social learning should not be indiscriminate. Many cultural evolutionists accept,
at least for modeling purposes, that the capacities for social learning and the cap-
acities for individual learning are separate, and that the former needs to evolve in
a species that possesses the latter, but they point out that social learners, in the
simplified world of the model, copy others at random. This does not need to be
the case: cultural evolutionists talk about “transmission biases” or “social learning
strategies.” I will use the two interchangeably, even though my preference goes to
social learning strategies, as the term “bias,” as used in psychology and economy
(and in common language), suggests some sort of faulty tendency (you are biased!
algorithms are biased!): in the original formulation, it is not us being biased, but
the process of transmission, that favors certain traits over others.
Social learning strategies are relatively simple, general-​domain (i.e. they can be
applied no matter what is the specific content of the cultural traits at stake) heur-
istics to choose when, what, and from whom to copy. Transmission biases are
rule-​of-​thumb principles such as “copy the majority,” “copy kin<” or “copy when
uncertain” that allow obtaining accurate information on average. Copying your
kin or your parents is hardly an infallible strategy, especially for specialist know-
ledge, but it works on average, in particular when you are young, as the interest of
your parents are generally aligned with yours and they have more knowledge than
you in most domains.14
Social learning strategies put a limit on generalized copying, but the critical ques-
tion is where to set the bar: to be evolutionary effective, social learning strategies
 Opaque customs  29

need to strike a balance between being parsimonious and sufficiently accurate.

How many individuals should you observe before deciding whether to try the red
or the blue fruit? Should you also try by yourself? How many times? If a strategy
is too accurate it will give you correct information, but its costs, be they cognitive
or, for example, temporal (when you finally take your decision, blue fruits are fin-
ished), may overtake its benefits. If a strategy is too simple, on the other hand, it
will be cheap and fast, but it will provide you with incorrect information too often.

Opaque customs

Besides being relatively cheap and fast, according to cultural evolutionists, there
is another reason for which social learning strategies cannot be too picky. For this
same reason, they also need to be, by and large, content-​independent, that is, they
should abstract from the specific features of the cultural trait we observe. Many cul-
tural practices are opaque to the individuals: not only by copying we save the time
we would have needed to experiment by ourselves, like in the case of the red and
blue fruits, but we also can discover techniques or learn skills that would have been
highly unlikely we would have figured out alone. In fact, we do not need to under-
stand what we are doing or why we are doing it, but, as Nike (or Shia LaBeouf)
would say, we have to “just do it.”
In these circumstances it is not important, or it is even detrimental, to try to
evaluate whether a cultural trait is good or not, or whether doing something is ad-
vantageous or not. Psychologists György Gergely and Gergely Csibra tell a well
known anecdote about a recipe for a ham roast, in their case “Sylvia’s recipe.” The
recipe includes a unique detail: at the beginning of the preparation, both ends of
the ham need to be cut. They continue:

One day, while her elderly mother happened to be visiting, she set out to make
her special ham for dinner. As her mother watched her remove the end sections,
she exclaimed “Why are you doing that?” Sylvia said, “Because that’s the way you
always began with a ham.” Her mother replied, “But that is because I did not have
a wide pan!”15

Sylvia does not know why she is cutting both ends of the ham. She just does it. She
copied her mother without bothering to understand the reasons behind her ac-
tions. Now, cutting the ends of the ham before cooking it is not a particularly useful
opaque cultural trait (at least, if we assume that Sylvia has bigger pans than her
mum), but, to remain on-​topic, cooking meat is. Cooking confers several advan-
tages: it makes meat more digestible; it kills the majority of potentially dangerous
parasites present at lower temperatures; and it may provide an overall energetic
gain with respect to eating raw meat. We do not have to know why cooking makes
30  Wary learners

meat more digestible: in fact, we do not have to know at all that it does. Again, we
just need to trust the people around us and do whatever they are doing.16
An interesting aspect here is that the cultural practice of cooking meat predates,
for example, the discovery of parasites and of their role for human health. While it
is possible that early meat cookers knew that cooking meat was good for them, it is
also possible that they did not, but the ones who were doing it looked more healthy
or prosperous, or they lived longer and had simply more occasions to be copied
than individuals that were not cooking meat. Even though cultural evolutionists
are wary of too-​easy analogies between the processes of cultural and biological
evolution (we will discuss this in many other parts of the book), many of them
strongly support the idea that cultural evolution, is, as biological evolution, a dumb
process, operating on (relatively) dumb individuals, that creates smart adaptations.
As in the famous Orgel’s second rule for natural selection: “cultural evolution is
smarter than you.”17
Joe Henrich, Robert Boyd, Pete Richerson, and colleagues have discussed sev-
eral examples where this happens, including food taboos, manioc processing, and
even divination, which may be interpreted as a randomization machine that al-
lows hunters not to return too often to the places they consider lucky, thus pro-
viding clues to preys to avoid them, or to deplete the resources in specific patches.
Netsilik and Copper Inuit people use clothes made of caribou skins to keep them-
selves warm in the Arctic winters. Caribou fur happens to be better suited than
other animal fur, such as, for example, bear, because of the microstructure of the
individual hairs, that consist internally of cells that fill with air, forming something
similar to insulating bubbles. When temperature may dip to –​50 degrees, a caribou
fabric can be crucial for survival. Not only individuals do not have to rediscover
that caribou is good for clothing every generation, but Inuit people do not need
to—​and they probably do not know anything about the microstructure of caribou
hairs. As in the case of cooking meat, they do not need even to realize that caribou
clothes keep them warmer, but they simply need to do what the others around
them do, or what they tell them to do.18
Maxime Derex and colleagues illustrated this with a clever experiment. Imagine
an inclined track with a wheel on it. The wheel has four radial spokes with a mov-
able weight on each of them. The goal of participants was to place the weights in
such a way that the wheel would go down the rail as quickly as possible. How would
you do it? The best tactic is not immediately obvious. There are two dimensions to
take into account. One is associated with the moment of inertia: just as ice-​skaters
who draw their arms in to spin faster, a wheel where the weights are close to the
center will go down more quickly than a wheel where the weights are placed to-
ward the ends of the spokes. The other one is related to potential energy. The higher
the center of mass of the wheel, the faster it will accelerate, so one should place the
weight of the spoke on the top (and only that) toward its end. The best approach is
to optimize both dimensions.
 What are social learning strategies?  31

Each participant was allowed to copy the solution of the previous one, simu-
lating a process of cultural transmission akin to the Chinese Whisper game (we
will discuss again this methodology and explore other examples later). Derex and
colleagues showed that the performance of participants increased along these
transmission chains, that is, the wheel set up by participants at the end of the chains
covered the track faster than the wheel of participants at the beginning of the
chains. The catch, however, is that participants did not understand the logic behind
it. Asked explicitly after the task whether, for example, a wheel with weights close
to the center will cover the track faster than a wheel with weights close to the ends
of the spokes, late participants were not better then early participants. Still, their
results were better: cultural evolution was smarter than them.19
Virtually all technologies we use today, and no doubt digital technologies, are
causally opaque to us. Should you update your operating system to the last version
or should you keep the one you currently have, which works fine? Are antiviruses
still useful for your computer? Are my data safe in the cloud, or should I save them
in an external physical hard disk? How does a hard disk work, by the way? These
are other examples of situations in which it is not easy to figure out individually
what is the best thing to do, and it may also be hard to understand pros and cons
of the various options. Again, using not-​too-​strict, domain-​independent, rules to
choose someone else to copy may be an effective strategy.

What are social learning strategies?

Social learning strategies are important modeling devices. The majority of models
of social transmission, often developed by physicists or computer scientists, as-
sume relatively simple forms of social influence, in which individuals copy others
based on reaching a certain threshold of exposure or, similarly, their probability to
copy is proportional to the number of times one is exposed to the trait. If more than
a certain number of your friends does something, or if you see more than a certain
number of times a particular advertisement, or a post in a social network, you will
be influenced by it. “Complex” contagion is more sophisticated: it involves a non-​
monotonic relation between the number of exposures and the probability of being
influenced. In other words, the more is not always the better when, for example,
too much exposure can be detrimental and inhibit transmission. In any case, these
types of models generally assume that the only feature determining the success or
failure of cultural transmission is the number of times one is exposed to the cul-
tural trait. There is no difference if, say, the cultural trait in question is ice cream
or spinach, or if the person posting on Facebook is a celebrity, your boyfriend, or a
semi-​forgotten acquaintance.20
Social learning strategies add some details to similar models. Different social
learning strategies create different population-​level dynamics. Cultural trait passed
32  Wary learners

along by parents will spread slower than cultural traits passed along by peers.
Cultural traits copied because they are popular will spread differently from cul-
tural traits copied because a famous person shows them. Given a social learning
strategy we can predict how a cultural trait will spread in a population. However,
proceeding the other way around is trickier. How confidently can we predict the
behavior of individuals based on social learning strategies?
Earlier accounts considered a small number of social learning strategies. In an
assessment of the development of cultural evolution theory from 2003, Joe Henrich
and Richard McElreath discuss prominently a handful of “context heuristics.”
Model-​based biases result from features of the potential model (the individual
you may, or not, copy from): they are a success bias (copying successful individ-
uals), a prestige bias (copying individuals to whom others show deference), and a
similarity bias (copying individuals who are similar to you). Frequency-​dependent
biases result from an assessment of the frequency of a cultural trait: they are a con-
formity bias (roughly, copying traits that are common) and its converse, a rarity, or
anti-​conformity, bias (copying traits that are uncommon).
Another related question concerns what exactly this typology represents.
Henrich and McElreath write that:

natural selection will favor cognitive mechanisms that allow individuals to ex-
tract adaptive information, strategies, practices, heuristics, and beliefs from other
members of their social group at a lower cost than through alternative individual
mechanisms. Human cognition probably contains numerous heuristics and
learning biases that facilitate the acquisition of useful knowledge, practices, be-
liefs, and behavior (“cultural traits” or “representations”). These mechanisms can
be usefully modeled at the algorithmic level, much as some cognitive scientists
investigate other kinds of information processing.21

Similarly, in Not by genes alone, probably the first popular account of modern cul-
tural evolution theory, Robert Boyd and Pete Richerson discuss how:

selection can favor a psychology that causes people to conform to the majority
behavior even though this mechanism sometimes prevent populations from
adapting to a change in the environment. Evolution also favors a psychology that
makes people more prone to imitate prestigious individuals and individuals who
are like themselves even though this habit can easily result in maladaptive fads.22

From this perspective, transmission biases are considered a suite of psychological

adaptations shaped by natural selection, similar to cognitive adaptations described
by evolutionary psychologists. As we saw, these mechanisms should not be too dis-
criminating, both to provide an optimal cost/​benefit balance and to allow adap-
tive blind copying (remember the wheel experiment). In addition, as evolved
 What are social learning strategies?  33

mechanisms, they should be tuned—​thus be particularly effective—​to the social

and physical environment of small-​scale societies.
The evolutionary logic behind social learning strategies tells us that indiscrim-
inate copying is not expected to evolve, but it leaves, up to a point, the door open to
the maladaptive effects of social influence. As we will see in detail in the next chap-
ters, there are good reasons why strategies such as copying the majority or copying
successful, or prestigious, individuals are efficient, especially when compared to
copying at random. Of course, however, they could also backfire, and the more
the conditions are different from the social environment in which they evolved,
the more undesirable and unpredictable the effect. We have, so the story goes,
both wisdom of crowds and madness of crowds, and influenceable teenagers fol-
lowing the lead of the current pop star, when not the last social media “challenge”
involving dangerous behaviors.
Lately, cultural evolutionists had stressed more the importance of flexibility
when using social learning strategies:  sometimes copying from parents make
sense, sometimes it is the skilled ones, other times the majority, other times not
copying at all. Many experiments show the importance of flexibility. Children,
from an early age, learn preferentially from parents in some domains, from experts
in other domains, and finally they chose to copy their peers for others, such as the
preference for certain toys (as any adult who tries to choose toys for their toddler
knows). In a series of four lab experiments, Tom Morgan and colleagues find that
adult participants were sensitive, among other factors, to the number of demon-
strators (copy the majority), to their own confidence on how good they were in the
task (copy when uncertain), to the cost of errors when using individual learning
(copy when asocial learning is costly), or to the actual performance of the dem-
onstrators (copy successful individuals). Even chimpanzees, when they can access
different rewards, easily adjust their learning strategy. A chimp used to give a white
cup to receive a peanut will not switch its behavior to follow a majority that receive
the same peanut, only by exchanging it for a brown stick. The same chimp, how-
ever, will promptly switch when proposed a new peanut-​exchanging-​machine, one
that gives five, instead of one, peanuts back.23
A recent authoritative review paper highlights that the usage of social learning
strategies is “flexible and changes with ontogeny, experience, state, and context.”
The authors also sketch a broad, if somehow unsystematic, taxonomy of contextual
social learning strategies “for which there is significant theoretical or empirical
support.” They list 21 of them, going from wide and possibly ambiguously defined
ones such as a “copy if uncertain” (how does one know it, in the first place?) to the
clear-​cuts, if of limited utility for humans, such as “size-​based” (copy individuals
bigger than you).24
Late accounts, in sum, emphasize that social learning decisions are, for all spe-
cies, and in particular for humans, complicated by several details. This perspective
provides a better fit with empirical evidence. The trade-​off is that the predictive
34  Wary learners

power of social learning strategies becomes limited (as their usage is extremely
flexible) and it is less clear what they exactly represent, beyond being descriptions
of how individuals seem to behave in experimental situations. But let’s go back to
our main theme. As I will try to convince you that we are not that influenceable, it is
useful to make a short detour to two classic social psychology experiments from a
few decades ago showing that we are ready to submit to the majority opinion, even
when it seems obviously the wrong choice, and to follow authority’s demand to
their extreme, and sinister, consequences.

“Blindly going along with the group”

The great majority of scientific experiments are known only by a small group of
fellow scholars working in the same field. Few of them reach a larger academic
audience and are cited in literature reviews and introductory manuals. Very, very,
few get to be known outside the world of professional science. They have the bur-
densome honor of informing our view of human behavior, usually with spec-
tacular results that tend to satisfy some of our preconceptions on how this behavior
should look. One of these is known as Asch’s conformity experiment. According
to a handbook of social psychology, the experiment provides “one of the most dra-
matic illustrations of conformity, blindly going along with the group, even when
the individual realizes that by doing so he turns his back on reality and truth.”25
In the 1950s, social psychologist Salomon Asch conducted a series of studies
intended to test to what extent social information could influence individual judg-
ment. The participants were introduced into a room where they joined six to eight
other people they thought were also participants, but were in reality confeder-
ates of the experimenters. The experimenter showed them a card with a line—​the
target—​and another card with three lines of different lengths. In the second card
only one line matched the length of the target, while the two others were clearly
different from it and from each other (see Figure 2.2). The experimenter then asked
the group which one of the three lines was the same length as the target. Because of
the way the group was positioned in the room, the participant was always the last
one to answer and listened to what the others said. Each round was composed of
18 trials. In six of them the confederates gave the obvious correct answer, but in the
others they agreed unanimously on one of the “wrong” lines. What was the parti-
cipants’ reaction?
The results, often defined in popular accounts as “chilling,” seem in fact quite rea-
sonable. Twenty-​five percent of the subjects never bowed to the majority opinion,
and kept on, for all trials, to give the correct answer, impermeable to any social in-
fluence, whereas only 5 percent of the subjects always gave the same answer as the
confederates. Over all subjects, and all trials, 36.8 percent of answers, around one
third, were influenced by the majority opinion. One could even interpret Asch’s
 “Blindly going along with the group”  35

A B C

Figure 2.2  Cards used in Asch’s experiment.

Licensed under the Creative Commons Attribution-​Share Alike 4.0 International License. Attributed to
The Author, Fred the Oyster, 2014. https://​commons.wikimedia.org/​wiki/​File:Asch_​experiment.svg.

original results as showing that only a minority of answers were influenced by

others, even when the group was in complete accord. That would be, however, an
unfair interpretation. The stimulus was not ambiguous at all and, in a control con-
dition with no social influence, subjects who were alone made no mistakes whatso-
ever. So, the effect of social influence needs to be explained.
However, imagine what is going on here. You find yourself in an unfamiliar, and
likely intimidating, situation, such of participating in a scientific experiment. All
the people there seem to see something slightly differently from you, or at least they
say so. There is nothing at stake, apparently, so that changing your mind or, anyway,
saying that you agree with everybody else does not look like a big deal. What is sur-
prising, at least to me, is that so few participants adopted the opinion of the group.
Moreover, variations of the set-​up show that by tweaking the experimental details,
social influence might be weaker. When Asch himself replaced a single confederate
with a “true” participant (or instructed confederate confederate to give the correct
answer) so that the wrong answer was still given by a majority of people, but not by
a unanimous one, the participants answered incorrectly “only one fourth as often,”
that is, slightly less than one in ten times.26
Importantly, task difficulty and relevance for the participants interact to de-
termine the degree to which people rely on majority opinion. In a modified ver-
sion of Asch’s experiments, the task was presented to participants as a study of
eyewitness accuracy. The experimenters showed the participants a drawing of a
male figure (the “perpetrator”) followed by a lineup of four male figures, one of
36  Wary learners

whom was the perpetrator, possibly dressed differently in respect to the previous
image. As in Asch’s original set-​up, confederates of the experimenters—​in this case
two students—​were unanimously providing a wrong answer. The experimenters
however varied the importance of the task. In the high-​importance condition
participants were informed they were contributing to upgrading a critical test
used by police departments and courtrooms to assess the quality of eyewitness,
and that the most accurate of them would be rewarded with a monetary prize of
20$. Participants in the low-​importance condition, on the other hand, were told
they were part of a pilot study to test the materials and to find out how to best
present the stimuli, so that their actual answers were not particularly important.
Experimenters also manipulated task difficulty:  in the high-​difficulty condition
the drawings were presented for 0.5 seconds (the perpetrator) and 1 second (the
lineup), while in the low-​difficulty condition they were presented for 5 and ten sec-
onds, respectively.
When the task was easy and the perceived importance was low, the experiment
replicated the results of Asch, with approximately one third of the participants
choosing the wrong answer suggested by the confederates. However, in the high-​
importance condition, the proportion of participants choosing the wrong answer
halved. On the other hand, when the task was difficult the effect was the opposite: in
the high-​importance condition participants were more inclined to copy the answer
of the confederates than in the low-​importance condition. In other words, partici-
pants were not “blindly” going along with the group. When the stakes were high,
participants were less prone to follow social influence when they had good grounds
for thinking they knew the correct answer, and they were more prone to follow so-
cial influence when they had good grounds for thinking they did not know the cor-
rect answer. That the confederates were basically cheating them does not change
the fact that the participants were making a reasonable choice when copying their
answer in the high-​difficulty condition, as they could not know better.27
The results of Asch are robust and have been replicated several times. However,
the interpretation of these results is more nuanced than the way they are some-
times presented. No doubt, social influence is a powerful force and it is fascinating
that, in the appropriate conditions, we may discard our previous knowledge or
even mistrust our own perception. But we do evaluate, up to a point, these condi-
tions, and we hardly follow others’ example blindly and automatically.

The power of authority

Another chilling—​in this case the wording may be more appropriate—​social

psychology research that speaks to our fear of being influenced against our will is
the notorious Milgram’s obedience experiment, carried out a decade after Asch’s.28
Milgram recruited participants for what was presented to them as an experiment
 The power of authority  37

on memory and learning. Participants arrived in pairs to the lab and they were
assigned by a “random” draw in two roles: the “teacher” and the “learner.” In fact,
the learner was always a confederate of Milgram. The “experimenter”—​another
actor—​then strapped the learner into a chair and attached an electrode to his wrist,
with the teacher watching. The teacher was then taken to a room separated by a
transparent glass from the learner. The teacher, instructed by the experimenter,
guided the learner through a word-​pairs association task. Each time the learner
was wrong, the teacher had to administer to them an electric shock. The voltage
started at 15 volts but the teacher had to increase it by steps of 15 volts each time the
learner was wrong, until a maximum of 450 volts was reached. The shocks were, un-
beknown to the participants, simulated. A gradually increasing sound was played
each time the participant administered the electricity to the learner. Moreover, the
learner was instructed to scream louder as the voltage supposedly increased, until
banging in protest on the glass that separated the two rooms. Towards the highest
voltages the learner started to fell silent.
Incredible as it seems, Milgram reported that 65  percent of the participants
stayed to the end of the experiment, administering the 450 volts shock to the learner.
Not surprisingly, Milgram’s results have become well known beyond social psych-
ology. They are often cited to explain why we—​that is, fairly mild individuals—​
can commit atrocities (Milgram himself referred explicitly to Nazi concentration
camps in his first paper on the subject). You may have read the occasional men-
tion of Milgram in media comments about Abu Ghraib and Guantánamo, or even
watched the movie Experimenter about it. Peter Gabriel recorded a song titled We
Do What We’re Told (Milgram’s 37) (the number 37 refers to the participants who
in a particular experimental condition administered the higher shock, on a total
of 40).
Milgram explained his results with what he called agentic theory, which roughly
implies that individuals, in certain situations, assume that others are in control of
their actions and do not feel responsible for the consequences. Not surprisingly,
the results can be readily interpreted as showing the possibly nefarious effects of
social influence. In the case of Asch—​Milgram was his research assistant, and his
PhD dissertation was an extension of his conformity experiments—​participants
obey a majority of their peers, while in Milgram’s case they obey a single individual
who is in a position of authority.
But, again, the message coming from Milgram’s experiment might be more nu-
anced. Differently from Asch’s, the “obedience” experiments have not been pos-
sible to replicate faithfully for obvious ethical reasons. Some participants were
distressed if not traumatized from the experience, and ethical concerns were
raised from the beginning on Milgram’s procedure. So we cannot be sure if this
65 percent of participants willing to inflict extreme pain to unknown others, just
because they were told so, is a robust result. A cautionary note comes from the
fact that Milgram performed 23 different experimental conditions with high
38  Wary learners

variations in the results. While in some conditions the results are similar or even
higher than the figure usually presented (about 60%), in other conditions they are
lower. Pulling together all the results, researchers found that the average of par-
ticipants delivering the highest shock was 43 percent, still an impressive number,
but technically a minority.29 In addition, the variance in the results through dif-
ferent conditions points to the fact that, as for the conformity studies, people are
not “blindly” obeying. For example, in a condition where teacher and learner
knew each other, being friends, neighbors, or even relatives, “only” three out of
20 participants choose to apply the highest voltage. Slight variations, such as the
experimenter not being in the same room, and giving instructions to the teacher
by phone, drastically changed the results, with the proportion of participants
obeying dropping to 20 percent.30
An analysis of the interviews given by participants after the experiment to the
actor who played the role of the experimenter is also particularly interesting. While
clearly, participants may rationalize a behavior that surprised themselves, the an-
swers provided suggest more than unconditional obedience, a mixture of different
motifs used to justify their actions. Importantly, 72 percent of the “obedient” parti-
cipants mentioned that they did not think that the learner was really harmed. Some
may have explicitly suspected the deception, as was suggested by contemporary
critical accounts. Others reasoned more generally that, had the learner really being
in danger, the experimenter would have stopped the experiment. Notice that they
were indeed right.31

Social learning: underused or overhyped?

Our evolutionary perspective, and the presumption of good design that accom-
panies it, should make us at least skeptical of accounts that endow social influence
with too much power. Cultural evolutionists, as we saw, defend a view where we
use social learning strategically. This perspective is, however, not devoid of prob-
lems. There is a tension between an interpretation of social learning strategies as a
suite of few general cognitive adaptations that implement clear rules making us to
copy more in certain situations than in others (“copy common traits,” “copy from
prestigious individuals,” and so on) and another one where there are several pos-
sible strategies, different in different individuals and cultures, used with great vari-
ability depending on the situation. Cultural evolutionists often implicitly endorse
the first interpretation, for example (and for good reasons) in modelling works.
This interpretation leaves open the possibility that our social learning strategies
will be often out-​of-​target in the modern digital world, but it seems, as we will ex-
plore in the next chapters in more detail, not a particularly good description of our
behavior. The second interpretation fits better with what we observe, but at the ex-
pense of not providing, at least yet, a satisfying theory of our behavior.
 Social learning: underused or overhyped?  39

So far we considered social influence, intended as using contextual cues to de-

cide whether to copy or not a cultural trait (How many other people have a red
shirt? Are they high status? Are they successful?), possibly against one’s own evalu-
ation of other pieces of information. Another, more general, aspect is the usage of
social information itself: how much do we copy others at all? As for the discussion
of Rogers’ model, cultural evolution theory clearly predicts that we should not al-
ways use social information. However, many cultural evolutionists—​again, often
implicitly—​think that we should use it often enough: culture is, after all, “the secret
of our success.” A clear illustration of this inclination comes from the interpret-
ation of the results of a computer tournament, which took place few years ago, with
the goal of helping to establish the better strategy to choose between social or indi-
vidual learning.
The tournament involved strategies realized through computer programs com-
peting one against each other in a simulated environment. Imagine a world where
there are 100 possible options—​say, 100 different foods—​and each of them gives
you a different fitness payoff, drawn from an exponential distribution (the conse-
quence is that very few foods are nutritious, while the majority are not particularly
so). In addition, with some probability, the fitness return of a food can change at
each time step (you will have noticed this is, not by chance, a more complicated
version of Rogers’ model). Simulated individuals use the strategies coded in the
competing computer programs. Individuals have three options. The first is to
“innovate,” that is, to choose a food at random and learn its fitness return, which
amounts to individual learning. The second is to “observe,” (i.e. social learning): the
individual acquires information on the food others ate the last time, and on their
fitness returns. The number of observed others is a variable parameter of the simu-
lations. Finally, individuals can “exploit,” that is, choosing one of the foods stored in
their memory, eating it, and acquiring the fitness associated to it.
The model simulated an evolutionary process. Individuals have some fixed
probability to die (their life is 50 time steps long on average) and, when it hap-
pens, a new individual replaces them. The strategy of this new individual is in-
herited from another individual present in the population, chosen proportionally
to its mean fitness: the new entrant is, so to speak, its offspring. So, individuals who
eat better food reproduce more at the expense of individuals who are not good at
discriminating, and the strategies of the former spread in the population. Other
details of the tournament are more complicated (computer programs competed
successively one against each other, or many of them were present in the same
population; various parameters of environmental change or number of observed
others were tested, and so on) but, overall, the tournament had a clear winner, a
strategy called DISCOUNTMACHINE.32
DISCOUNTMACHINE used, among other things, a sophisticated way to dis-
count (hence the name) previous information on fitness payoffs, drawing on its
own estimate of the rate of environmental changes, so that old information was
40  Wary learners

judged less important the more the environment was evaluated as rapidly chan-
ging. The striking feature of DISCOUNTMACHINE, however, was that it almost
never used “innovate”, the move that was deemed equivalent to individual learning.
This approach was found to be a general feature of the successful strategies: “some
of the best performers, including DISCOUNTMACHINE, INTERGENERATION,
WEPREYCLAN, and DYNAMICASPIRATIONLEVEL, ranked 1, 2, 4, and 6, all
played OBSERVE on at least 95% of learning moves,” According to Kevin Laland,
one of the organizers of the tournament, the take-​home message was that “copying
beat asocial learning hands down over virtually all plausible conditions.”33
Is that really the case? It is intriguing that a suggestion going in the opposite
direction comes, among others, from laboratory experiments realized by cultural
evolutionists. Cultural evolution is a young discipline, and the predictions coming
from models have started to be empirically tested only in the last 10–​15  years.
Focusing on experiments in which it is possible to estimate whether and how
much human participants relied on social, as opposed to individual, learning, and
involving only adult participants (as children could be more—​or less—​relying on
social information depending on their age), an interesting pattern emerges. For
many experiments, researchers report a “puzzling anomaly:” participants use so-
cial learning less than what would be expected.
What do these experiments look like? An example is the “virtual arrowhead”
task that Alex Mesoudi used for several studies. Participants need to build on a
computer screen an arrowhead, by providing features like height, width, and thick-
ness (that all matters for the efficiency of their arrowhead), or color and shape (that
does not, which is not known by the participants). When the arrowhead is ready,
participants engage in a virtual hunt, where they are assigned their score. Whereas
there are several variations, the crux of the experiments is similar:  after some
training, the participants are asked before each hunt if they want to copy the arrow-
head of another individual (or a composite coming from the arrowheads of dif-
ferent individuals) or if they want to keep on modifying the characteristics of their
arrowhead, drawing on their previous results. ‘Social learning’ applies when par-
ticipants copy others’ arrowheads; ‘individual learning’ applies when they modify
themselves the attributes of their chosen arrowhead.
In a version of the experiment, participants—​undergraduate students in a
London university—​could choose various alternatives regarding what to copy: the
arrowhead values (how thick, what color, etc.) of the highest-​scoring players, the
most popular values, an average of the values of other participants, or just taking
the arrowheads of another player at random. This information was provided for
free. Finding the optimal features of the arrowhead is a complicated task, and
simulation models, along with the results of the participants that performed better,
indicate that the best strategy is to copy the successful, highest-​scoring, players.
Participants did prefer this strategy (the other social learning alternatives were vir-
tually not used), but they did it only a few times. The great majority of learning
 Social learning: underused or overhyped?  41

events (77.5%) were asocial. In addition, it is not that participants were using stra-
tegically social information, say copying a few times and then performing fine-​
grained adjustments by themselves to optimize their arrowheads. As mentioned,
the few participants who copied extensively got higher scores.34
This is only one experiment, but the same pattern has been found in many
others.35 Besides the very general similarities described above (participants are
human adults and it is possible to estimate the proportion of social and asocial in-
formation usage), the experiments vary widely, making it impossible to perform
a rigorous meta-​analysis of the degree of reliance of social information. However,
since the underuse of social learning has been found in several experiments, with
different set-​ups, it may be reasonable considering it not as an anomaly, but as an
intriguing finding in itself.
How do we reconcile this pattern with the results of the social learning strategies
tournament and, more generally, with the idea that humans are avid copiers? Some
studies have focused on individual variability in the usage of social learning.
It has been suggested that the propensity to be influenced by the behavior of
other peoples might be calibrated, for example, in function of the level of environ-
mental risk they face. Pierre Jacquet and co-​workers showed to the participants of
their experiments unfamiliar faces, and asked them to rate their approachability.
After that, they showed them the approachability ratings of the same faces made
by a fictive group of peers, and asked the participants to rate the faces again. They
also collected information on participants’ levels of childhood harshness and un-
predictability, having them say how much they agreed with statements such as “My
family usually had enough money for things when I was growing up.” or “When
I was younger than ten, things were often chaotic in my house.” Participants who
experienced greater levels of childhood harshness and unpredictability were also
more likely to change their approachability ratings in the same direction of the
peer group. They were relying more on others’ views to make decisions.36
Similarly, others have proposed that our propensity to engage in social leaning is
itself culturally variable. As much as we copy behaviors and techniques, we can also
copy from others from whom to learn, when, and how much to do it: call it the so-
cial learning of social learning rules. Intuitively, if people around you tend to copy
others often, and tend to reward this activity, you will be more motivated to use
social learning than if people around you discourage copying, and believe that it is
important to solve your problem by yourself. This idea makes sense—​I have myself
explored some of its consequences in formal models—​and it is now being tested
empirically, checking whether participants from different societies copy each other
to a different degree.37
Alex Mesoudi and colleagues, using the “virtual arrowhead” task we just de-
scribed, compared the copying frequency in samples of participants with a dif-
ferent background:  British students, Chinese students in the UK, Hong Kong
students, and lastly the “Chinese mainland” sample, that is Chinese students from
42  Wary learners

Chao Zhou, “a relatively small city of 2.6 million inhabitants.” They found that,
while participants from the first three groups replicated the results we already
know, of scarce reliance on social learning, students from Chao Zhou copied more.
Other experiments similarly found difference in social learning usage in partici-
pants coming from groups with different subsistence styles, with “interdependent”
pastoralists and urban dwellers using more social learning than “independent”
horticulturalists.38
Transcultural and ecological variation in reliance on social learning is real and
important, but the next question is then why entire populations—​western stu-
dents, urban Chinese students, and horticulturalists from Ethiopia, from what
we know so far—​do reasonably well underusing social learning. Moreover, the
Chinese mainland students in the experiment of Mesoudi and colleagues copied
others around 30 percent of times, which is higher then what the other groups did,
but not too far from the results of other artificial arrowhead experiments (in the
first we reported above, social learning was used 22.5 percent of times), and still at
odds with the suggestions coming from the social learning strategies tournament.
It may also be that social learning is not underused, but that cultural evolution-
ists may have overestimated our reliance on it. Some reasons are quite technical
and specific to modeling or experimental practices (or to features of particular
models or experiments). For example, it has been noted that, in the social learning
strategies tournament, when individuals used “exploit” (corresponding to “eating”
one of the food stored in their memory and acquiring the associated payoff), the
current payoff value of this food was also updated in the memory of individuals, so
that using “exploit” was effectively another form of individual learning. When “ex-
ploit” and “innovate” are put together, and compared with “observe,” the relation-
ship between frequency of social learning and success of the strategy disappears.39
Or think about environmental change:  for obvious practical reasons, experi-
ments and models cannot last forever. A session in the artificial arrowhead task
lasts 30 hunts, and the life of an individual in the social learning strategies tour-
nament lasted on average, as we saw earlier, 50 time steps. Now, an environment
that is not completely fixed, but changed slowly, could be thought as changing at
least one time in the life of an individual, or in a session of an experiment. In fact,
the minimum probability of change tested in the social learning strategies tour-
nament was 5 percent for each time step (i.e. it happened, on average, only once
for each individual), and for the artificial arrowhead task it was higher, fixed to
three times overall. Imagine that you learn how to hammer nails, but then, every
ten times you do it, something changes, and you would better to do it differently.
Imagine that your favorite breakfast food is likely to become, every two months
or so, poisonous. Life would be hard! In general, when you learn once in your life
that bananas are nutritious or how to put your fingers around a nail, than you can
keep this information, unchanged, for thousands and thousands of breakfasts and
hammerings. When looking at single instances of usages, a reasonable assumption
 Deception and vigilance  43

is that we neither learn individually nor socially, but keep on doing what we did the
previous time in the overwhelming majority of cases.
There are also more general, and possibly more interesting, reasons to explain
the overestimation of social learning. One has been briefly touched on earlier,
when discussing Rogers’ model and its assumption that social and individual
learning are two separate ways to acquire information from the environment. As
we discussed, this assumption can be justified when evaluating in abstract the rela-
tive advantages of individual and social information, but less so when describing
the behavior of human and other animals, and its evolutionary underpinning, with
the result that a reasonable model assumption could have crept into theory and ex-
perimental implementations. To quote philosopher Kim Sterelny:

extensive social learning in human social life is not typically explained by a cost-​
driven flight from direct individual learning. There has been no such flight. Much
human learning is hybrid learning [ . . . ]. Often our adaptations for social learning
do not operate to replace direct trial and error learning about the environment
but to supplement it, by making it more reliable and by reducing its costs.40

Presenting the alternative between using individual and social learning as a dis-
crete, on/​off, choice, renders unreliable the estimates of their relative importance,
when the true answer would be: both. Some experiments have tested set-​ups that
do not require an explicit choice between social and individual learning. In another
study by Maxime Derex and colleagues, involving (again) the construction of an
arrowhead, the participants where, as usual, assigned to two conditions: individual
learning and social learning. In the individual learning condition, not surpris-
ingly, they relied on their past performances to modify their arrowhead, without
accessing information from other participants. In the social learning condition,
participants had access to the arrowhead of the other members of their group, and
they could choose one to copy, but then they had still to build their own arrowhead,
taking inspiration from the copied one. Participants did not have to discard their
own knowledge when copying others, as social and individual information could
both be used, and this was indeed what they did. The arrowheads produced in such
“social” condition were, on average, influenced to a same degree by social and indi-
vidual information.41

Deception and vigilance

Another important factor that explains why our reliance on social learning could
have been overestimated is that cultural evolution models consider the possibility
of errors as the main, if not the only, peril of social learning: the others are wrong,
or we copy them incorrectly. Kate is eating blue fruits, but she did not realize that
44  Wary learners

red fruits are now the best option. We think we saw Kate eating blue fruits, but she
was eating the red ones. The effects of the rate of environmental change-​source of
Kate’s wrong behavior -​and the probability of copying error -​how likely we are to
copy wrongly a correct behavior -​are the main focus of many cultural evolution
models studying the adaptiveness of social learning. There is, of course, another
possibility: Kate is eating the red fruits but, as she does not like us, or because there
are not many of them, or for any other reasons, she tells us that blue fruits are the
best choice. Kate is intentionally deceiving us.
Whereas the majority of cultural evolutionists focus on psychological mechan-
isms that make possible faithful copying, such as imitation, others think that we
should pay more attention to communication. Of course there are forms of social
learning that do not require communication: I suspect many cultural evolutionists
have in mind animal social transmission when they realize their models. Animals,
including humans, learn a lot just by observing other behaviors or even simply by
interacting with an environment that has been modified by others. A well-​known
example of the latter is blue tits learning to pierce the foil tops of milk bottles: they
did not need to observe conspecifics performing the behavior, but only to find
around bottles that had already been opened. On the other hand, communication
is more than social learning: we use communication to coordinate joint actions, to
influence others’ behaviors, to express our feelings, all without necessary intending
to transmit new information to others.42
However, communication, in particular ostensive communication, has a fun-
damental role in cultural transmission, at least among humans. (Ostensive com-
munication is communication that is both voluntary and overt. Voluntary:  we
want to communicate some information. Overt: we also communicate our desire
of communicating.) Many examples of cultural transmission, probably the ma-
jority, involve someone communicating with someone else with the explicit inten-
tion of transmitting some information. Teachers are an obvious example, but any
time you explain something to your friends you are engaging in a form of osten-
sive communication and—​if everything goes well from your perspective—​cultural
transmission.43
Communication, however, comes with the possibility of deception. Occasionally
it is pointed out that language, the prime form of communication for humans, is
the closest thing to mind-​control we have. No need to mention Orwell here: I can
tell my wife to go and buy bread on her way after work and, at least in the majority
of cases, I can expect that she will be home with a fresh sourdough. However, I can
also tell a stranger in the street to give me all they have in their bank account so
I can save the planet from an ecological catastrophe. In this case I would not ex-
pect them to comply. Sometimes, the interests of others overlap with ours (both
my wife and myself want to have bread for dinner), but other times they do not (the
unknown passerby prefers to keep their money in their bank account rather than
giving it to me).
 Deception and vigilance  45

Whereas I would have a net advantage if the stranger gives me their money, in
the long run this situation would not work: strangers would do better being suspi-
cious. How far should we go? In an extreme situation, nobody would trust anyone
else, with the dire consequence that communication would not be useful any
more. We need, in sum, to be able to calibrate our reliance on communicated in-
formation, in a way that, on average, advantages both the sender and the receiver.
Unlike other species—​think about alarm calls that communicate one and only one
thing—​we can talk about practically everything. For a communication system as
complex and domain-​general (to remain advantageous, and hence evolutionary
stable, a parallel system of mechanisms that evaluate communicated information
should have coevolved. Dan Sperber and his colleagues called this system “epi-
stemic vigilance.”44
It is impossible here to provide a full account of the mechanisms for epistemic
vigilance that have been proposed. It suffices to say that epistemic vigilance im-
poses stricter limits on social influence than the transmission biases we described
earlier. Epistemic vigilance concerns a series of more sophisticated cognitive
mechanisms, aimed at plausibility checking (the detection of inconsistencies be-
tween one’s own background beliefs and novel information), trust calibration
(the detection of cues to infer senders’ trustworthiness and commitment), or rea-
soning (the ability to find and evaluate reasons to support or discard a belief). By
the same evolutionary logic, it is unclear what the effect of the radically changed
contemporary conditions is, when communication happens potentially in long
transmission chains, potentially involving billions of unrelated individuals. The
theory of epistemic vigilance, however, suggests that, given potential risky situ-
ations, we would tend to err more on the side of conservatism than on the side of
gullibility.45

The idea that we are easy targets of social influence is a powerful one. In a 2018
survey of US residents, the majority of respondents stated they were generally
skeptical about the information they saw on social media, Facebook in particular.
Only 5 percent of the respondents admitted they believed “all or most of it” to be
true, while the overwhelming majority trusted “a little of it,” or only the informa-
tion shared by friends and family. The interesting part comes after: when asked not
about themselves, but about “most people on Facebook,” 81 percent thought they
(the others) were “too quick to believe all or most of what they see.” This phenom-
enon is so widespread there is a name for it: third-​person effect. The third-​person
effect has been studied for the last thirty years in various settings, and it appears
robust. We tend to believe that social influence, especially when negative, media
propaganda, and manipulation, have a stronger impact on other individuals than
on ourselves.46
We may either overestimate the effect of propaganda on others or underesti-
mate the effect on ourselves and probably we do both. In any case, the third-​person
46  Wary learners

effect is interesting because it opens a window on our anxiety about the possible
negative consequences of social influence and on our scarce ability to properly
evaluate it. How much of it is justified? How gullible are we? Of course, social in-
fluence is a powerful force: we are submerged by social information and cultural
products. At any moment, almost everything that surrounds us is a consequence
of long chains of socially transmitted information. However, exactly because it is
potentially so beneficial—​this is how the story goes—​social influence could be
used to manipulate us, against our judgment and interests. This idea has both a
powerful intuitive appeal and a long reach in psychology and social science. The
advent of digital media, and the internet in particular, amplified this apprehen-
sion. This fear is justified: as seen in the first chapter we have access to a radic-
ally changed network for cultural transmission, which is bigger, faster, and more
opaque than ever before.
This chapter has shown, however, that some caution is needed before crying
wolf. Cultural evolutionists have pointed out that, to be adaptive, social informa-
tion usage needs to be selective, even when all individual interests are aligned,
simply because the environment changes, and the information we acquire from
others may be outdated. Cultural evolutionists talk about social learning strategies
or cultural transmission biases to describe a series of rule-​of-​thumbs rules, such
as “copy the majority” or “copy prestigious individuals,” which serve this role. But
what are these strategies? They can be interpreted as universal, evolved, cognitive
mechanisms that act quasi-​automatically and provide a causal explanation for the
diffusion of a cultural trait (“French bulldogs became popular in the 2010s because
we copied celebrities like Lady Gaga”) or as a large suit of flexible, variable, heur-
istics, which are useful to describe how some traits are transmitted. While these
interpretations are both problematic, the idea of social learning strategies draws a
first line to an unbound influenceability.
Cultural evolutionists think that, even though social learning is subject to pos-
sible shortfalls, it is useful in the majority of situations. A surprising possibility
however is that experiments carried on by the same cultural evolutionists are
showing that our propensity to learn from others may have been overestimated. In
addition, we have seen how cultural evolution models do not consider the possi-
bility of explicit deception—​and for a book focusing on the digital age, that would
be an obvious shortcoming. Consequently, we discussed, shortly, the perspective
from epistemic vigilance theory, which assumes that we apply more sophisticated
cognitive operations when deciding whether to trust information coming from
others and, overall, that we are not gullible as we think we, or perhaps others, are.
This does not mean that learning from others is not important or that we are
impermeable to social influence (these would be strange conclusions indeed) but
points to a different image of humans as “wary learners”, or as others called them,
“flexible imitators”, As we will explore more in the next chapters, wary learners,
while following social cues for cheap and relatively unimportant choices, are
 Deception and vigilance  47

generally cautious and weigh different sources of information when faced with
more consequential decisions, both offline and online.47

Notes

1. Books on cultural evolution:  Sperber (1996), Richerson and Boyd (2008), Mesoudi
(2011a), Henrich (2015), Morin (2016), Laland (2018)
2. Golder and Macy (2011), Baylis et al. (2018)
3. Mayr (1982)
4. See e.g. Richerson and Boyd (2008), Claidière et al. (2014)
5. Sperber (1996), Tehrani (2013)
6. Sperber (1996)
7. Lynch (2016)
8. Gazzaley and Rosen (2016)
9. Henrich (2015)
10. Laland (2004)
11. Rogers (1988)
12. Enquist et al. (2007)
13. Heyes (2012), Danchin et al. (2018)
14. Boyd and Richerson (1988), Kendal et al. (2018)
15. Gergely and Csibra (2006)
16. Carmody and Wrangham (2009)
17. Henrich (2015)
18. Boyd et al. (2011); Henrich (2015)
19. Derex et al. (2019)
20. See e.g. Hodas and Lerman (2014) and references therein
21. Henrich and McElreath (2003)
22. Richerson and Boyd (2008)
23. Wood et al. (2013), Morgan et al. (2012), Van Leeuwen et al. (2013)
24. Kendal et al. (2018)
25. Asch (1955), cited in Hodges and Geyer (2006)
26. Asch (1956)
27. Baron et al. (1996)
28. Milgram (1963)
29. Haslam et a. (2014)
30. Reicher et al. (2012)
31. Hollander and Turowetz (2017), Orne and Holland (1968)
32. Rendell et al. (2010)
33. Laland (2018)
34. Mesoudi (2011b)
35. See e.g. McElreath et al. (2005), Efferson et al. (2007), Mesoudi (2008), Toelch et al.
(2009), Morgan et al. (2011), Toelch et al. (2014), Acerbi et al. (2016), Toyokawa et al.
(2017, 2019), Novaes Tump et al. (2018)
36. Jacquet et al. (2019)
48  Wary learners

37. Acerbi et al. (2014), Mesoudi et al. (2016)

38. Mesoudi et al. (2015), Glowacki and Molleman (2017)
39. Heyes (2016)
40. Sterelny (2006)
41. Derex et al. (2015)
42. Sherry and Galef (1984)
43. Morin (2016)
44. Sperber et al (2010)
45. Mercier (2017)
46. https://​www.cbsnews.com/​news/​americans-​are-​skeptical-​facebook-​can-​protect-​user-​
data-​cbs-​news-​poll/​, Davison (1983), Paul et al. (2000)
47. Morin (2016)
 3
Prestige

Social influence offline and online

On June 29, 2018, Canadian rapper Drake released his double album Scorpion. The
same day, a semi-​obscure Instagram “influencer,” the comedian Shiggy, posted a
video on the same social media site where he danced to one of the tracks of the
album, In my Feelings. The comedian appears to be in a car park, at night, dancing
alone for half a minute to some parts of Drake’s song. A friend with a phone, pos-
sibly sitting on a car, filmed him. He accompanies the lyrics “Kiki Kiki, do you love
me?” by mimicking a heart with his hands, and the subsequent “Are you riding?”
pretending to steer a car. The video is not exactly remarkable, and still, captioned
with the hashtag #DoTheShiggy and #InMyFeelingsChallenge, it was the seed of a
viral internet phenomenon, later on better known as the “Kiki challenge” (from the
name of the woman addressed by Drake in the song).1
Mainstream media reported how celebrities had contributed to the taking off of
the challenge. In particular, New York Giants’ footballer Odell Beckham Jr. posted
his version of the #InMyFeelingsChallenge to celebrate the Fourth of July, and was
quickly followed by various, more-​or-​less known, famous personalities, including
notably Will Smith, who posted, on July12, his version, professionally shot (in-
cluding a cameraman and a drone filming him) from a bridge in Budapest.
Ordinary people like you and me started to post videos where they were dancing
to the same segments of In my feelings, possibly imitating the “love-​heart” and
the “steer” gestures, and in a very short time, it was difficult to be on social media
without seeing a Kiki challenge mention. The hashtag #kiki had, at the beginning
of July 2018, a sudden increase in popularity on Twitter. It peaked shortly after, at
the beginning of August, and it declined as quickly as it rose to success, going back
to pre-​craze levels around September (see Figure 3.1).2
However, the August peak of popularity was caused by something else. Slightly
different versions of the Kiki challenge started to surface online: in these versions,
the protagonists—​now ordinary people—​jump out of a slowly moving car and
dance alongside it while they are filmed from inside the car, while it continues to
move. How these new versions originate is not entirely clear, but they are reminis-
cent of Shiggy’s original version, that looked like it was filmed from a car, and, even
more, of the Odell Beckham Jr.’s version, where the footballer is seen stepping out
of a vehicle in the first seconds of the video and he is clearly filmed from there (the
car is however not moving).

Cultural Evolution in the Digital Age, Alberto Acerbi. Oxford University Press (2020) © Oxford
University Press.
DOI: 10.1093/oso/9780198835943.001.0001
50 Prestige

1.00

Interest (scaled) 0.75

0.50

0.25

0.00
Jan 2018 Apr 2018 Jul 2018 Oct 2018 Jan 2019
Date

Figure 3.1  Relative frequency of the hashtag #kiki on Twitter.

In fact, from the end of July, successful videos were mainly collages of “Kiki chal-
lenge gone wrong” or similar, with a variety of “oblivious dancers crashing into
poles, tripping on potholes or falling out of the cars. One video shows a woman
having her handbag stolen while attempting the challenge and another shows a
man being hit by a car while he dances.” Not surprisingly, newspapers reported that
authorities everywhere began to get worried, issuing warnings about the danger of
the challenge, and reminding that jumping from a moving car was behavior pun-
ishable by a fine. Or, as the Daily Mail reported, with customary understatement,
“Authorities warn people may die if they insist on filming dangerous dance craze.”3
The Kiki challenge highlights several typical themes—​some we will explore
in detail in the following chapters—​of internet micro-​fads. It went viral quickly,
and disappeared in a couple of months, thanks to the immediate reach and avail-
ability provided by digital media. In this short period, it nevertheless changed, with
the addition of the behavior of jumping from a vehicle in movement and dancing
following the car. As we will see in Chapter 7, Transmitting and Sharing, internet
memes are seldom transmitted with total fidelity. Mainly, its spread is often under-
stood as process driven by social influence: the videos of celebrities such as Odell
Beckham Jr. or Will Smith were functional in making the challenge successful and,
once it was diffuse enough, everybody jumped on the bandwagon of the popular
phenomenon.
When giving it more attention, however, how much can we explain with the so-
cial influence narrative? The video that launched the Kiki challenge was posted
by a comedian that hardly fit, before the success, the standard definition of ce-
lebrity. Whereas Will Smith and the other stars have certainly had a role, it is dif-
ficult to establish whether the dance became popular because of their influence
and not because, through their widely followed accounts, they simply made the
 Social influence offline and online  51

video available to a larger amount of people. The same can be said for the effect of
popularity: if something is popular it is also more available, but it is not clear if we
become particularly interested in it because it is popular. In addition, the highest
moments of interest corresponded with the diffusion of “gone wrong” videos, made
by unknown individuals, and often reported by mainstream outlets. In Chapter 6,
Misinformation we will discuss how certain features, such as negative content or
the presence of threat-​related information, even when, as in this case, not relevant
for the viewers, are psychologically appealing and can boost the success of cultural
traits.
As we saw in the previous chapter, the behaviors of our conspecifics, what they
tell each other, and, of course, what they tell us directly, are primary elements of the
environment, in a broad sense, that our cognitive tools evolved to process. Copying
what others do and being influenced by what they say is fundamental for humans.
In this respect, the spread of the Kiki challenge is nothing special. Cultural traits,
be they Shiggy’s dance, Mexican recipes, or how to multiply two numbers, are con-
stantly passed on, retained, and modified, in long transmission chains spanning
many individuals (at least when the traits are successful). Not just that, but some of
the cognitive mechanisms that operate when we decide whether to watch and share
a Kiki challenge video, try a Mexican recipe, or learn a multiplication technique,
are the same. To understand what happens in the digital age, we need to under-
stand how these mechanisms operate in general and whether there are distinctive
features of our digitally mediated interactions that make them operate differently.
We should expect, as we discussed, that these mechanisms would produce, on
average, adaptive results. We replaced the image of easily manipulated, gullible,
individuals, with one of wary and flexible learners (and, if anything, with an inclin-
ation to caution). However, of course, these mechanisms cannot be perfect and,
even more, cultural evolutionists suggest that they need, up to a point, to be open
to some level of un-​scrutinized social influence, so that cultural evolution could be
smarter than us.
In addition, these mechanisms could be more effective when operating in an
environment very different from the one we are facing now. We saw how online
interactions create massive effective cultural populations -​individuals we can po-
tentially share cultural traits with, increasing both the reach of our influence, and
the availability of other cultural traits to us. The difference is not only in size: cul-
tural populations in the digital age are made by countless individuals we do not
know, we do not have reason to trust, and whose competence we are not able to
evaluate (opacity). Possibly to counteract this, explicit and quantified cues of popu-
larity or reputation are widespread on the internet (I call this feature explicitness),
but they are easy to manipulate, and it is conceivable they are also a novelty, in cul-
tural evolutionary terms, to which we still need to adjust.
In this and the next chapter, I  will focus on two such mechanisms:  our ten-
dency to copy prestigious individuals (this chapter) and to copy majority choices
52 Prestige

(Chapter  4, Popularity). Among the social learning strategies discussed earlier,

these are intuitively the most important and they are, indeed, the most studied
in cultural evolution. They both potentially produce smart population-​level out-
comes resulting from actions by dumb individuals: us.(i.e. Prestige-​biased copying
allows, if everything goes well, the copying of adaptive traits without suspecting
they are adaptive, but simply because the source of these traits, the prestigious indi-
vidual, has them. Copying popular traits allows, in theory, to profit from aggregate
individual learning, so you do not need to eat red and blue fruits yourself to figure
out which are poisonous and which are nutritious.
However, the story goes, these mechanisms also bring about the Kiki challenge
(which is, by the way, a decidedly innocuous fad, except for the very few people
who were harmed. I am sure you can think of more sinister examples of online
“persuasion”). How do these mechanisms operate offline and online? Is the fear of
online social influence running amok justified? Let’s start with prestige.

Prestige as a cultural magnet

According to a 2018 survey, the most desired occupation for more than half of
Chinese born after 1995 is “influencer.” Instagram celebrities live the dream
life—​at least according to their profiles. They get paid, often handsomely, to do
what they like, which usually happens to be traveling, wearing lavish clothes, and
participating in fashionable parties. As I write, the latest photo posted on Instagram
by Portuguese footballer Cristiano Ronaldo, with 153 million followers currently
the biggest account on that social media, shows him taking a selfie on his private
jet, wearing a sweater with a big “Fendi” logo on it, with what seems to be a cup of
mate on the table. The photo has been liked more than seven million times.4
Copying prestigious individuals is one of the social learning strategies we dis-
cussed in the previous chapter. While in the age of Instagram it may seem prepos-
terous, there is an evolutionary logic behind it. Let’s go back to our early example of
a learning strategy: copying what your parents do. As we said, it intuitively makes
sense. They care for you, and they have had more time than you to learn useful skills
by themselves or from others. However, this is clearly not foolproof. Sometimes
things change fast and your parents do not know the latest developments. This is a
cliché in contemporary western society, but it is not unheard of in traditional soci-
eties. Anthropologist Barry Hewlett and geneticist Luigi Luca Cavalli-​Sforza ana-
lyzed how Aka pygmies, a group of hunter-​gatherers living in central Africa, learn
various skills, from food processing (“preparing manioc”) to hunting techniques
(“making poison”) to how to dance and sing. They found that more than 80 percent
of those skills were learned through what they call vertical cultural transmission,
that is, from biological parents to offspring. However, in their discussion, they re-
port how the diffusion of some hunting techniques followed a different trajectory.
 Prestige as a cultural magnet  53

When they carried out their study, the crossbow was a relatively recent introduc-
tion, and it was first observed among Aka pygmies by 1958, used by “some [ . . . ]
but many still used bow and arrow.” By the time that Hewlett and Cavalli-​Sforza
write, in 1986, “all Aka use the crossbow, and no Aka use the bow and arrow.” The
crossbow, then, spread homogeneously in the Aka culture in less than two gen-
erations, a time frame that, as recognized, is incompatible with learning from
parents.5
Other skills may be more stable, but specialized and possessed by only a few
individuals—​think playing a musical instrument or doing statistics in our society.
Similarly, some skills exhibit high variation in the population, that is, a few people
are very good at them, but the majority are just average—​think hunting in a trad-
itional society, or mathematics in the contemporary western world (playing foot-
ball like Ronaldo also works well). For these skills, as before, learning from parents
is not a good bet. An alternative strategy is trying to assess directly the expertise of
individuals based on their success, a strategy called pay-​off or success-​biased social
learning: one can look, for example, at the various hunters and learn from the one
that is more successful.6
This strategy presents, in turn, other problems. Success is capricious, and it can
be due to luck instead of being the product of skills. How many hunting sessions
should one apprentice observe before choosing the successful hunter to learn
from? Moreover, the relationship between success and skills may be unclear and
difficult to recognize, especially if one lacks the expertise in that domain (but this is
exactly the situation in which learning is useful!): if I do not know anything about
math, how can I figure out who is a good mathematician? Or think about group
hunting in a traditional hunter-​gatherer society: success depends on many factors,
including difficult-​to-​notice behaviors that facilitate coordination among mem-
bers of the group, and how can a non-​expert decide which, among all hunters, is
the one to learn from?7
A possible alternative is prestige-​biased social learning. Joe Henrich defines
prestige-​biased social learning as a “second-​order” strategy: prestige provides cues
to detect the individuals that the people around us are copying. One can observe to
whom other people show signs of deference or respect, or simply whom they tend
to hang around more. Suppose someone has been already able to determine who is
the best hunter to learn from: they will hang around this individual, and they will
probably admire and respect them. These cues have been defined “second-​order”
cues as they rely on the behavior of other people toward the prestigious individ-
uals. In addition, one can use “first-​order” cues, that is, cues that refer directly to
the appearance of the prestigious individuals (as we will see soon, an ethnographic
example concerns tattoos) or their material possession (another ethnographic ex-
ample: super-​sized yams).8
Deciding from whom to copy based on prestige cues is cheap and fast, but does
it work? Intuitively, there needs to be a reliable correlation between skills and the
54 Prestige

conferred prestige. It is highly unlikely that copying the body ornaments of a pres-
tigious hunter will make you, in turn, a good hunter. Or, what if a hunter is presti-
gious because of their particular ability to build traps, but you are trying to learn
how to shoot arrows? What if people show signs of deference to someone because of
their family, or because of their physical appearance? Formal models from Robert
Boyd and Pete Richerson suggest, however, that, as long as there is any correlation
between skills and prestige, prestige-​biased social learning is more effective at least
than completely indiscriminate copying (assuming also that the costs of detecting
prestige cues are lower than the benefits produced by the copied behavior).9
Experiments from cultural evolutionists give some support to the idea that
people use prestige-​biased social learning. In two different experiments, both chil-
dren and adults preferred to learn from a prestigious demonstrator than from a
non-​prestigious one. In both experiments prestige was operationalized by pro-
viding information on which demonstrator had more attention from other indi-
viduals. In the experiment with children, the 3-​and 4-​year-​old participants saw a
video in which two bystanders stood between two demonstrators, but they looked
at only one of them, for ten seconds. In one condition, for example, the “presti-
gious” demonstrator used a novel toy, a sort of plastic hand attached to a stick, to
play with colored balls. The other demonstrator used the same tool, but to play
with colored bricks. When asked how to play with the plastic-​hand tool, children
said that it was made to play with colored balls. The experiment with adults used
again the “virtual arrowheads” setup we described earlier. Participants could select
different “virtual arrowheads” to copy from, and they had (manipulated) informa-
tion about how long other participants had looked at each arrowhead. Participants
preferentially copied the arrowhead that appeared to be observed more than the
others.10
Ethnographic data tells a more nuanced story. In a study of two villages in a
Fijian island, researchers found that individuals were more likely to copy from
demonstrators who showed cross-​domain success, which would imply they were
following prestige cues. People who were, for example, successful in fishing, were
also copied more in other, unrelated activities, such as how to grow yams effectively
or which medicinal plants to use, and how to use them. The reverse was true for
yams (i.e. people successful in yam growing were copied more in the domains of
fishing and medicinal plants) but not for medicinal plants knowledge. This result
is explained with the supposition that fishing, especially spear-​fishing, and yam
growing both produce conspicuous signals of success—​that is, big fish being caught
and big yams being harvested—​contrary to the more erratic and hidden successes
in the medicinal plants’ domain. It may also be, however, that Fijian people were
correctly inferring that the skills necessary for fishing and yams growing tended to
be correlated, but not the skills linked to the knowledge of medicinal plants. This
would make for a less dumb cultural evolutionary process, as described in the pre-
vious chapter, but would be indeed effective for the villagers.11
Runaway processes, cultural hitchhiking, missing-target problem  55

In another study—​ which happens again to involve ethnomedicinal plant

knowledge—​but with a slightly different perspective, Victoria Reyes-​Garcia and
colleagues tried to estimate whether prestigious individuals were also knowledge-
able in the ethnomedicinal plant domain. They found that this was not the case. The
reason might be that in the society they studied, the Tsimane’ in Bolivian Amazon,
ethnomedicine is not an important skill anymore, as they have easy access to
western medicine. Shamans, the figures “that concentrated Tsimane’ specialised
ethnomedicinal plant knowledge,” and who would supposedly be the targets of
deference and preferential attention, were no longer present in Tsimane’ villages.
The newly prestigious individuals are plausibly those who can speak English or
who went to school, not those who know how to use a medicinal herb.12
Also in the experiment with children just mentioned, toddlers were not blindly
following prestige cues. When demonstrators were observed playing with the
plastic hand, children were following their guidance, but they were not when the
“prestigious” demonstrators were, for example, evaluating the quality of a snack.
The demonstrators made a face that implied disgust after trying a big cracker and
“happily sampled” a small one (or vice versa). In this case, children were less in-
clined to trust the prestigious demonstrator.13
These are just a few studies, but they suggest that prestige cues are followed—​
followed with caution. This is indeed what we would expect if our dispositions to
learn from others function reasonably well. It makes sense to imitate the behavior
of the rich-​and-​powerful, at least in a sense that other people consider them worth
their respect and deference, and still there are many other factors that are weighted
when deciding what to copy. Following prestige cues, in particular, presents a few
risks we have only mentioned briefly. It is time now to explore them in more detail.

Runaway processes, cultural hitchhiking, and

the missing-​target problem

As discussed above, the strength of the social learning strategies proposed by cul-
tural evolutionists stands in a trade-​off between how fast and cheap they are and
their accuracy. In addition, in order to be adaptive, they need to partly override in-
dividual intuitions, so that cultural evolution can be “smarter than us.” The down-
side of this is that they are open to various shortcomings, and different strategies
have different, typical, shortcomings. In the case of prestige bias we can distinguish
three of them:  runaway processes, cultural hitchhiking, and the missing-​target
problem.
In their exposition of prestige-​biased social learning, Robert Boyd and Pete
Richerson use the diffusion of tattoos among Polynesians as a case of dynamic
guided by prestige bias. They argue that prestige bias can generate runaway pro-
cesses similar to what happens with sexual selection—​the proverbial peacock’s
56 Prestige

tail—​and thus explain exaggerated cultural traits. If tattoos provide a cue to the
prestige of an individual, individuals with tattoos will be copied more than individ-
uals without and, in addition, individuals with big tattoos will be copied more than
individuals with small tattoos. Crucially, among the traits that can be copied there
will be also tattoos themselves, so big tattoos will be copied more, with some vari-
ation in size. Repeat this cycle over and over again and it may explain the full-​body
tattoos that were relatively common among Polynesians, especially among boys.14
Another critical point is that tattoos are also a good example of cultural traits
that do not, or very indirectly if we assume one needs to be brave or be able to en-
dure pain for being tattooed, correlate with skills. As we briefly mentioned above,
prestige bias can lead to copying traits that are not useful, a phenomenon at times
dubbed “cultural hitchhiking.” Cultural hitchhiking can be idiosyncratic, but it can
also easily stabilize neutral or, up to a point, detrimental traits. Imagine a presti-
gious individual possesses three traits: the first is a skill that genuinely explains why
the individual is prestigious: say, the archery abilities of a prestigious hunter. The
other two are not: for instance, his habit of singing before hunting and his pebble
necklace. People tend to copy a variable number of traits: the ones who copy only
the necklace and the singing style will not become in turn prestigious and thus will
not be preferentially copied, but the ones who copy any of these, plus the archery
abilities, will. This is cultural hitchhiking. Now, if more people, for random reasons,
copied the necklace and the archery abilities than the singing and the archery abil-
ities, after a few iterations the two may become effectively interlocked, so that all
good archers will have a pebble necklace too.15
Finally, we can copy from prestigious individuals who are prestigious for reasons
that do not interest us. I will call it “missing-​target problem.” Let’s return to the
prestigious hunter: I can be lucky and copy his archery techniques instead of the
necklace, thus avoiding cultural hitchhiking, but my main occupation consists of
collecting herbs, or food processing, or anything else. In this case I will have ac-
quired a skill that is useful in itself, but not in my particular situation.16
In traditional societies, these drawbacks may not be a big deal. Specialization is,
at least if compared to modern societies, limited. Moreover, and more importantly,
effective cultural population size is bounded by the number of individuals you can
have face-​to-​face interactions with, as discussed previously. These people face more
or less the same problems you do, and their skills will generally be useful for you.
Mismatches can happen—​copying prestigious fishermen will not get you good
information on medicinal plants—​and cultural runaway processes are possible—​
from tattoos to, apparently, super-​sized yams—​but what happens when prestige
signals can be detected and produced in a worldwide effective cultural population,
distributed around the globe, and composed of millions of individuals?17
Cultural hitchhiking and the missing-​target problem seem to be especially sensi-
tive to the increased availability provided by digital media. The bigger the network
in which cultural transmission is possible, the harder it is to assess which skills
 Did Einstein really say that?  57

correlate with prestige. The further away to our concrete experience the prestigious
are, the easier it is to pick up skills that are not existent (cultural hitchhiking) or
not relevant in our local environment (missing-​target problem). George Clooney’s
coffee tasting expertise is unlikely to correlate with his acting abilities, still, the
story goes, the success of a brand of coffee depends on the presence of the actor
in the advertisements. Moreover, there are good reasons why Clooney is presti-
gious, but it is unlikely we are copying him in these specific domains. If one copied
Clooney to steal his secrets about how to make it in the acting industry, prestige
bias would indeed be a reasonable strategy.
Of course, while parroting pop stars’ hairstyles or buying the—​alleged—​favorite
coffee brand of celebrities are interesting social phenomena, there could be other
reasons to be worried. Prestige bias, coupled with increased cultural access, could
favor the spreading of misinformation supported by prestigious people, facilitate
extremism and online proselytism or, even, according to some accounts, have a
role in copycat suicides.18
We will now review some of the research on prestige and mass media, and then a
few selected studies on “influencers” in digital and social media to try to assess how
convincing these worries are. First, however, I will describe an experiment I did
together with anthropologist Jamie Tehrani, aimed at exploring these questions in
a very specific, and easy to reproduce in the lab, case: do we prefer quotes that are
attributed to famous individuals??19

Did Einstein really say that?

The phenomenon of online diffusion of misattributed quotes is so widespread

that it earned its own dedicated meme. You may have seen a picture of Abraham
Lincoln, the sixteenth president of the United States, that warns, “Don’t believe
everything you read on the Internet just because there’s a picture with a quote next
to it.” Lincoln, apparently nicknamed “Honest Abe” when young, was assassinated
in 1865, which makes it unlikely he had opinions about the internet, and he is one of
the historical celebrities most quoted (often incorrectly) on the web. Lincoln shares
this questionable honor with the likes of Mark Twain and Albert Einstein. “The
definition of insanity is doing the same thing over and over again and expecting
a different result” is one of the most famous quotes of Albert Einstein. Except
that it is not: the earliest known exact match for the quote appears in a Narcotics
Anonymous information pamphlet in 1981, some 25 years after Einstein’s death.
You may recall a quote I  mentioned earlier, “Everything should be made as
simple as possible, but not simpler,” also credited to Einstein. The quote appeared
for the first time in 1950 in a New  York Times article by the composer Roger
Sessions. Sessions attributed the remark to Einstein, but it cannot be found in any
of Einstein’s previous writings. However, Einstein said, in a 1933 lecture in Oxford:
58 Prestige

It can scarcely be denied that the supreme goal of all theory is to make the irredu-
cible basic elements as simple and as few as possible without having to surrender
the adequate representation of a single datum of experience.

Same meaning, but clearly not as catchy.

Intuitively, the association with a prestigious author should amplify the cred-
ibility and appeal of a quote, and this may explain the success of misattributed
quotes on the web. Or, at least, this is what we thought. To test the hypothesis we
selected 60 mildly famous, but not, according to us, recognizable quotes in three
different domains:  love/​friendship (e.g. “If you love somebody, set them free”),
money/​success (e.g. “There are people who have money and people who are rich”),
and science/​literature (e.g. “Science may set limits to knowledge, but should not set
limits to imagination”). Since we believed that the association with famous authors
would turn out to have an effect, we wanted to understand if the domain modulated
this effect. We reasoned that the influence would have been stronger if the author
was famous because of the activity in the specific domain, thus famous scientists
and writers (Charles Darwin, Edgar Allan Poe) would have been particularly pre-
ferred in the science/​literature domain. We expected less influence for the domain
money/​success—​famous authors may have some general experience of personal
success, more than common people, but that’s not, so to say, their specialty—​and
even less for the domain love/​friendship, where one would not expect famous au-
thors necessarily to know more than any of us.
The participants of the experiment were supposed to help us to choose the “most
inspirational quotes” through a web interface, and they were presented with pairs
of quotes from the same domain, one associated with a famous author, and one
with a randomly generated name (Odis Pennel, Romeo Lyon, and others). We
asked them to choose the one they preferred (notice that, even if we tried to use
wide-​ranging quotes, this randomization may have created in a few cases some
suspects in the participants: a few lucky ones may have seen, for example, the quote
“A kiss is a lovely trick, designed by nature, to stop words when speech becomes
unnecessary” credited to Max Weber or Mahatma Gandhi).
Long story short: we did not find any effect of the association with famous au-
thors. It did not seem to make any difference if a quote was associated with Albert
Einstein or to Romeo Lyon, no matter what the domain was. Of course, the first
thing we thought was that we had made some errors or that participants, who were
recruited online, were simply choosing at random one quote or the other. What we
found, however, was that the success of a quote was correlated with the success of
the same quote in a control condition, where pairs of aphorisms were presented
without any indication of their (supposed) source. In sum, participants were ac-
tually evaluating the content of the quotes, and not who (supposedly) said them.
Quotes such as “It is better to have loved and lost, than never to have loved at all”
had consistently more success than, for example, “love is the big booming beat
 How strong is the influence of celebrities?  59

which covers up the noise of hate,” irrespectively of the name that was associated
with the quote.
To exclude any effect of the content, we ran a follow-​up experiment, in which we
presented the same quote to different participants, and we asked them “how good”
they thought the quote was. Half of the participants saw the quote associated with
a randomly chosen famous author, and the other half with an unknown name. The
evaluation of the quote credited to famous authors was this time slightly higher in
all domains, but the differences were never significant.
Contrary to our expectations, the association with prestigious authors did not
boost the preferences for the quotes. Why, then, the diffusion of misattributed
quotes? We speculated that the reason could be, in fact, the opposite of the intui-
tive one. A good quote will be more likely to succeed independently of whether it
is associated with a famous author or not. However, when reproduced again and
again, the information on the author—​which is seemingly less important than the
content itself—​could get lost. This may favor famous authors in two ways: easy-​
to-​remember authors replace authors who are more difficult to remember. For ex-
ample, the quote “Your assumptions are your windows on the world. Scrub them
off every once in a while, or the light won’t come in,” is usually credited to Isaac
Asimov, but the words were pronounced by Alan Alda, a less famous American
actor and director. Second, anonymous quotes are more likely to be credited to
well-​known individuals than to lesser known ones, such as “Be Yourself, Everyone
else is already taken,” an aphorism for which we have only very recent matches, but
often misattributed to Oscar Wilde.
In sum, while the final result is the same—​famous authors and successful quotes
go hand in hand—​the causal relationship is reversed:  the success of quotations
would not be the result of being misattributed to famous authors. On the contrary,
misattributions would be the result of the wide diffusion of good quotes.18

How strong is the influence of celebrities?

An immediate objection to our results is that no expertise was required to choose

between the alternatives and to provide an opinion on a quote. A basic tenet of cul-
tural evolution theory, the costly information hypothesis, is that social information
is valuable when it would otherwise be hard to obtain by oneself. That was not the
case in our experiment, so this may explain why participants did not use the pres-
tige cues we provided. On the other hand, the stakes were low. We expect individ-
uals, when motivations are not high, to be sensitive to contextual cues, the intuitive
rationale being: why bother to think for myself? Experiments in cultural evolu-
tion show that this is often the case: people follow the example of others for seem-
ingly irrelevant choices such as whether to cover the keyboard of a public computer
they used, whether to write a date analogically (“January 25, 2018”) or numerically
60 Prestige

(1–​25–​2018), or whether to write or draw their comment on a post-​it note in ex-

change for a small prize after a visit to the zoo.19
In fact, our task was similar to one of the favorite examples of prestige bias in
cultural evolution theory, that is, Michael Jordan’s successful involvement in the
advertisements of Hanes’ underwear. The task (choosing your underwear’s brand)
is equally low-​stake, and it does not require expertise. The question is then: do
people really follow Jordan’s advice when buying their underwear? This may
seem a quite surprising question, but think about it: for every George Clooney or
Michael Jordan, how many advertisements with celebrities did not succeed? Ozzy
Osbourne appeared in a television commercial for a butter substitute; Snoop Dog
was involved in another one for car insurances. Stephen Hawking was featured in
the early 2000s in a high-​profile campaign for an online fund platform that closed
shortly after, in 2004. Conversely, how many successful campaigns are out there
without celebrities? Think Apple’s famous “1984” commercial, or, more recently,
Dos Equis’ “Most Interesting Man in the World,” that became a widespread meme.
Examples are countless. There are a few reasons why we may be inclined to over-
estimate the influence of celebrities (more on this later in the chapter) but the avail-
ability heuristic, merely the fact that we remember the successful campaigns with
celebrities and forget the unsuccessful ones, may have a significant role.20
It is impossible here to provide a full assessment of the literature on celebrity
effectiveness in advertisements, but just allow me to discuss a few examples to
convince you that the results are, at best, mixed. The main message of an exten-
sive review of the role of persuasion variables (all of them, not only the prestige
of the person to copy) on attitude changes is that these variables have “complex
effects” and that their interaction is far from obvious. Regarding what are called in
the review source factors, that is, the factors that pertain to specific features of the
demonstrator, “credibility” is the factor which is closer to prestige bias in cultural
evolution. However, credibility depends in turn on “expertise” and “trustworthi-
ness.” Prestige by itself does not beget credibility. Sources are credible in their do-
main of expertise, and only or especially there, and they are credible only if they
seem honest and reliable, similar to what we discussed in relation to epistemic vigi-
lance in Chapter 2.21
An experiment testing whether the advertisements of the same six products was
more effective in a version with or without celebrity involvement concluded that:

Overall, the results from this study do not support the view that using celebrity
advertising is more believable or effective than non-​celebrity advertising for the
brands tested in this study. Consumers generally feel that celebrities are more at-
tractive than non-​celebrities, something that may draw initial attention to the ad-
vertisement. Beyond that, the celebrities do not seem to make the advertising any
more effective or believable. Further, purchase intentions did not vary between
the executions for any of the brands tested.22
 How strong is the influence of celebrities?  61

A meta-​analysis of 32 articles covering around 30  years of celebrity endorse-

ment literature found that the variable with the largest impact was in fact a nega-
tive one:  negative information on a celebrity (think O.J. Simpson and Hertz
Corporation) has the strongest effect on the value of the endorsement. The meta-​
analysis confirmed that celebrities can have a positive impact, but this is mediated
invariably by source expertise and trustworthiness. Michael Jordan suggesting
Nike Air: yes; Michael Jordan suggesting Hanes underwear: probably not. Notice
only the latter is, in cultural evolution terms, an example of prestige bias, while the
former, in which what is copied is in the domain of expertise of the demonstrator,
can be better described as a success-​based bias.23
A more recent meta-​analysis, based on 46 studies, “revealed a zero overall effect
of celebrity endorsements on consumers’ responses.” This does not mean, how-
ever, that celebrities do not have any effect, which would be as surprising as the
opposite extreme idea for which we are ready to figuratively follow our favorite
popstar everywhere, but that negative and positive effects cancel each other, and,
again, they are mediated by factors such as the congruence of the product and the
endorser or (negatively) by the familiarity we have with the product.24
The take-​home message seems to be that celebrity endorsement is an effective
strategy—​until it is not. Britney Spears’ Pepsi commercials of the late 1990s and
early 2000s are considered another case of an extraordinarily efficient campaign.
The advertisements for Toyota, made in the same period, when Britney was equally
famous, did not have great success, and they are long forgotten. Over and over,
the message seems to be that celebrity endorsement can work, especially when the
celebrity is considered knowledgeable in the domain and trustworthy, and the tar-
gets (us) do not consider themselves familiar with the product (I suppose coffee
connoisseurs are not too impressed by Clooney’s approval of Nespresso), but it is far
from being a safe bet.25
A recent experiment, with an explicit cultural evolution perspective, has tested
the effect of prestige on art appreciation, with interesting results. Researchers
showed photographic portraits of faces previously rated as “attractive” or “neu-
tral” to 150 university students. The portraits came from previous studies, so the
orientation of the faces was equal, the light conditions were the same, clothes or
jewelery could not be seen, and so on, but the researchers told the students they
were evaluating artworks. The students were asked to give an evaluation on how
much they appreciated them. The students, consistently with what the authors of
the study call “Evolutionary Aesthetics,” preferred, as artworks, the portraits of
attractive faces.
The relevant part for us comes later. The researchers also recruited a hundred
“expert” participants, by posting the link to the survey on the Facebook pages of a
modern and contemporary art museum, and of an art academy. They told some ex-
perts that the pictures were from the permanent collection of the MoMA (the well-​
known New York Museum of Modern Art) but did not tell others experts, and they
62 Prestige

did the same with the students. For the students, the information about belonging
or not to the MoMA collection was not relevant: they still preferred attractive faces.
In other words, they were not sensitive to a form of prestige bias (notice they were
provided with background information about the museum). On the contrary, the
expert participants preferred the (fake) MoMA artworks to the others, showing
they were sensitive to the prestige of the institution.26
What would have happened if the cues about prestige would have been the
fact that celebrities appreciated the portraits (say, “Katy Perry loves this portrait”)
and not that the portraits were part of a museum collection? What, on the other
hand, would be the reaction of the coffee connoisseurs to ratings from a star coffee
connoisseur—​supposing they exist—​instead of George Clooney? Again, prestige
bias seems to have an effect that is heavily influenced by the context, by expect-
ations, and by previous knowledge.

The Angelina Effect

What happens, then, when the targets of the influence of the prestigious have more
significant practical consequences than coffee brands or photographic portraits
of unknown people? Health-​related ideas and behaviors are a case in point. Elvis
Presley is celebrated by the media for many good reasons, but a curious one is to
have saved millions of lives. It was not for his contribution in spreading rock and
roll, but because in October 1956 he appeared in a picture while having a polio
vaccine shot (the shot was staged) before participating in The Ed Sullivan Show
(see Figure 3.2). The picture was published in all major American newspapers and
its publication coincided with a huge increase in polio vaccinations in the US. The
story is nice—​and it feels good to give credit to Elvis—​but the results of the trial
of the polio vaccine were announced only a year earlier, and massive vaccination
campaigns had only recently started in the country, so that the causal role of Elvis
is likely to be minimal. True, some demographic groups, such as teenagers, were
less sensitive to the vaccination campaign, but several initiatives were launched in
the same period, specifically targeted at them. One, in particular, organized by the
National Foundation for Infantile Paralysis (March of Dimes), which coordinated
thousands of American teens into volunteer divisions to spread the knowledge of
the vaccine, was considered key for the eradication of polio.27
Here is a similar, more recent, story. In May 2013, Angelina Jolie announced in
a New York Times op-​ed that she had had a genetic test revealing a mutation that
increased her risk of developing breast and ovarian cancers, and, as a consequence,
her decision to undergo a double mastectomy. She explicitly claimed to have gone
public with the story because she hoped that other women could benefit from her
experience. Shortly after, a Time magazine cover story titled “The Angelina Effect”
hypothesized that Angelina Jolie’s public confession could have resulted in a sharp
 The Angelina Effect  63

Figure 3.2  Elvis Presley stages a polio vaccine shot before appearing on The Ed
Sullivan Show.

increase in genetic testing, and, in general, in cancer prevention awareness. Did it

happen? An analysis from UK clinics and regional genetic services concludes this
was the case: “referrals were nearly 2.5-​fold in June and July 2013 from 1,981 (2012)
to 4,847 (2013) and remained at around two-​fold to October 2013.” A follow-​up of
the analysis, published in 2015, confirmed that the effect remained consistent two
years after the New York Times editorial.28
It is instructive to reflect a little more about this example of a successful story of
a celebrity endorsement to understand what one can expect to happen in general.
First of all, as in the Elvis case, it is problematic to disentangle the causal relation-
ships precisely. In the same year as Jolie’s editorial, the National Institute of Health
and Care Excellence (NICE) in the UK released the guidelines on familial breast
cancer that created much publicity on the topic. While the analysis mentioned
above controls for this effect, the public interest in cancer genetic screening and
prevention was growing internationally in any case, and Jolie’s op-​ed can be itself
a result of it (Stanley Lieberson, who can be considered the scientist who studied
more in detail how first names are subject to fashion, concludes that this is often
the case for the influence of celebrities on naming: the names are already trending
up when celebrities become famous).29
64 Prestige

Second, it is difficult to distinguish between the effect of prestige per se and the
effect of accessibility, that is, people simply being more exposed to the relevant in-
formation. Following Angelina Jolie’s op-​ed, information on testing and preven-
tion of breast and ovarian cancers become more diffuse. Certainly more people
were talking about it, and certainly medical doctors and practitioners were actively
using the possible increased interest to discuss the topic with patients. However,
could the same effect have been obtained with a successful New York Times edi-
torial without a celebrity, or with another campaign from health institutions?
Together with Stefano Ghirlanda and Hal Herzog we analyzed the effect that
the presence of a dog of a certain breed in a movie has on the popularity of that
breed. The impact of movies has indeed been substantial. We found, for example,
that movies have an influence that can last up to ten years from the initial release.
A striking and well-​known example is the 1959 Disney movie The Shaggy Dog. The
registrations of Old English Sheepdogs were stable at around 100 dogs per year in
the ten years preceding the release of the movie. Ten years later, in 1969 4,226 Old
English Sheepdogs were registered. The Shaggy Dog was an exceptionally profitable
movie and had a lasting impact on popular culture, but it would be a stretch to ex-
plain the diffusion of Old English Sheepdogs with prestige bias. The information
on the breed became accessible, more people got to know about it, therefore more
people who did not already have a preferred breed could easily pick this one.30
The dog breeds example is useful to introduce the third, and last, point. The
choice of a dog breed is, by and large, a neutral choice. Of course different dog
breeds have different features, but the features of different breeds can be reasonably
adapted to one’s own habits (or the other way around) so that the choice does not
have, in many cases, enormous effects on the owners. Many people who are buying
French bulldogs now would have probably bought poodles fifty years ago.
Angelina Jolie’s editorial made a few thousands more women in the UK con-
sult experts, women who were likely to have a family history of breast and ovarian
cancers, but did not generate an epidemic of people getting genetic screenings for
cancer. Celebrity effects, when they can be causally isolated from pure availability,
often concern choices that overlap with individual interests (cancer screenings) or
are relatively neutral for the people who copy the celebrity (dog breeds or coffee
brands and, if that was the case, Michael Jordan’s underwear).
When celebrities publicize choices that do not overlap with individual interests
and are likely to be costly, the effects are often negligible. The public association of
Tom Cruise with the Church of Scientology, with several direct interventions ad-
vertising the church, did not generate a noticeable increase of members. While the
estimates are difficult to assess, it seems that, if anything, the number of members
had decreased in the same period of Cruise’s involvement. When celebrities such
as Gwyneth Paltrow or Jenny McCarthy publicly support questionable practices
or ideas that are proven wrong, such as body stickers that “promote healing” or
more worryingly the existence of a link between vaccines and autism, the majority
 Online celebrities and local influencers  65

of people do not generally think “well, if those celebrities say so . . .” but, rightly
enough, keep on going their own way.31

Online celebrities and local influencers

How does what we said so far relate to what we know about online influencers? Did
the internet change everything and is our recent (apparent) familiarity with celeb-
rities making us copy useless, if not dangerous, cultural traits from them? Using
data from Twitter, Eytan Bakshy and collaborators attempted to analyze the effect
of influential users on 74 million “diffusion events” in that social media in 2009.
There are a few reasons why this study is interesting from our perspective.32
First, researches in digital media often use the number of connections one indi-
vidual has as a measure of their influence. In this way, it is impossible to distinguish
between accessibility and the sheer effect of influence. A message can be successful
not because the individual is influential, but because more people are likely to en-
counter the message (of course, understanding why an individual has many con-
nections is another, interesting, story). A more informative measure is the actual
success of the content that an individual spreads: controlling for the number of
followers, are some individuals better at spreading their cultural traits, or tweets,
in the specific case?
Second, the authors did not use explicit measures of the success, such as the
number of retweets or the number of mentions, but they attempted to follow the ef-
fective spreading of the shared content in which they were interested. For each link
to an external website posted on Twitter by the user X, they tracked, among the fol-
lowers of X, whether the same link was posted later on, and then they did the same
for the followers of the followers posting the link, and so on, recursively, creating a
“diffusion tree” for each link. This method does not control for homophily, that is,
Twitter users that follow each other are likely to share interests, so they could post
the same link without necessarily one being influenced by the other. On the other
hand, it allows tracking forms of content spreading that are not publicly acknow-
ledged by users with explicit retweets or mentions (in addition, in 2009, when
the data were collected, retweeting was not an option automatically provided by
Twitter, but it needed to be manually implemented by users themselves by writing
“RT @username”).33
Third, Bakshy and collaborators’ final goal was predictive: they were interested
not only in describing whether and how some Twitter users were more influential
than others. Rather, they wanted ideally providing to marketers a strategy to target
these users, without knowing beforehand who they were. This is very interesting
for us because one does not assume from the beginning that influencers exist.
Other accounts—​and plenty of anecdotal evidence, as we mentioned above—​just
consider the successful events and describe ex-​post how they unfold, or they look
66 Prestige

at the influence of all members of a population and define a threshold above which
the individuals are, by definition, “influencers”. Bakshy and colleagues instead con-
sidered all the individuals and all the possible spreading events, and fitted a model
to predict individual influence.34
When appraised with these cautions, the role of influential individuals ap-
pears quite modest. There are two features in the model that predict larger diffu-
sion events. Not surprisingly, one is the number of followers one individual has,
though as said above this cannot distinguish between influence and accessibility.
The second is a measure of past performance that the authors call “local influence,”
that is, the number of reposts of the original link by the direct followers of an indi-
vidual, or the first level of the link’s diffusion tree, as described above. Interestingly
however, the authors admit the performance of the model is “pretty poor.” This is
due to the fact that successful events are very rare—​when the large amount of non-​
successful events is considered too—​so that, “while large follower count and past
success are likely necessary features for future success, they are far from sufficient.”
We do not need to enter into the details of the model they sketch to suggest
a strategy to marketers, but the main result is that in the majority of situations,
the most effective strategy consists in targeting a bigger number of relatively or-
dinary users, more than focusing on the few users with many followers and with
good measures of past performance. In general, marketing and advertising studies
of online influencers and celebrities reiterate what we described in the previous
section about the pre-​digital era: their effectiveness is mediated by several vari-
ables, including their credibility, their trustworthiness, or their fit with the adver-
tised product.35
A trend only marginally explored in academic research, but resonating with our
account of the effect of prestige, is the rise of local or micro-​influencers. On the
one hand, the hyper-​connectedness provided by digital media allows traditional
celebrities to reach extremely large followings (as I write, there are twelve individ-
uals with more than 100 million followers on Instagram, and you may remember
that Cristiano Ronaldo, the most followed person on the photo-​sharing platform,
has 153 million followers, which is fifteen times the population of Portugal, where
Ronaldo comes from). On the other hand, digital media allow, as we explored
above, connecting with individuals with similar interests, no matter where they are
physically located: social media users seem to be more keen to interact, and be in-
fluenced by, other individuals who are more similar to them, or who possess more
expertise and are more trustworthy in a specific domain. Hyper-​availability allows
connecting with George Clooney as well as with a local coffee expert, and the latter
can be more persuasive, on average, when coffee is at stake.36
Local influencers are chefs, bloggers of niche subjects, YouTubers specialized
in DIY videos about primitive technology or that keeps up with the latest Apple
product. Their followers are counted in the thousands, and they are highly engaged
around specific topics. Marketing-​oriented research shows, for example, that on
 Prestige in perspective  67

Instagram there is an inverse relationship between the number of followers and the
engagement rate (the number of “like” and comments). In absolute number, the
more followers, the more engagement, but the ratio decreases as the number of fol-
lowers increases. A user with less than 1,000 followers has a “like” rate of 8 percent,
while a user with 100,000 followers has a rate of less than 2 percent. Given the costs
of engaging traditional celebrities, marketers suggest that “the sweet spot for max-
imum impact is an influencer with a following in the 10,000 to 100,000 range.”37
The suggestion about the importance of local influencers may not be restricted
to digital media but could be tested in a broader spectrum of circumstances.
Imagine you want to a start a campaign to sensitize teenagers about problems like
binge-​drinking or unsafe sex. One possibility is to engage big celebrities. Pick the
fashionable footballer or the famous pop star and put in place a traditional bill-
board advertisement campaign—​even though nowadays it will probably be on
YouTube or Snapchat. Another possibility is to mass-​advertise everybody with the
same message. You can go from school to school and lecture all students about the
issue at hand or start a campaign without endorsers. Alternatively, you could aim
at the local celebrities, the trendy kids in the school or the neighbourhood leaders,
with a targeted message, and see whether this helps to spread it broadly.

Prestige in perspective

That “influencers,” pop stars, and prestigious people in general can have less influ-
ence than what is generally thought can be surprising. As we have seen, it is easier
to remember the success-​stories of celebrities’ involvements—​Pepsi and Britney
Spears—​than their failures—​Toyota and Britney Spears—​because the failures are,
exactly because they were failures, forgotten. Even more interesting, we tend to
think celebrities are the cause of events of social diffusion while it is likely the reality
is more complicated: think about Elvis and polio vaccinations, or the “Angelina
Effect” for cancer prevention awareness. One reason for this is that understanding
why something succeeds in spreading is far from obvious, and, as I will repeat sev-
eral times in this book, many different causes, often of small strength when taken
singularly, concur in making cultural traits successful or not. A prime mover, espe-
cially if we can put a (famous) face on it is a more satisfying narrative.
Duncan Watts and Peter Dodds, in a modelling paper showing that, in short,
large cascades of social transmission are more likely to be produced by little events
amplified by many influenceable individuals than by the effect of few influencers,
use an apt metaphor:

Some forest fires, for example, are many times larger than average; yet no one
would claim that the size of a forest fire can be in any way attributed to the ex-
ceptional properties of the spark that ignited it or the size of the tree that was
68 Prestige

the first to burn. Major forest fires require a conspiracy of wind, temperature,
low humidity, and combustible fuel that extends over large tracts of land. Just
as for large cascades in social influence networks, when the right global com-
bination of conditions exists, any spark will do; when it does not, none will
suffice.38

The take-​home message of the research that I presented here is not that there is not
an effect of prestige bias in our online activities, but that the situation is at least not
as bleak as sometimes implied in public discourse. And, of course, if prestige would
work seamlessly and automatically, one could turn it for good, and just ask George
Clooney to have everybody quit smoking and start eating vegetables. Copying pres-
tigious individuals, as we saw, has an evolutionary rationale. Prestigious individ-
uals are individuals from whom others copy, and copying from them, as much as
copying what the majority of people around you do (we will see this in Chapter 4),
allows for a form of population-​level intelligence that goes beyond what you an me,
as isolated individuals, could figure out.
On the other hand, strategies such as preferentially copying prestigious in-
dividuals are prone to shortcomings. Cultural transmission linked to prestige
can generate runaway dynamics where the traits used to signal prestige be-
come exaggerated, like Polynesian tattoos. We can copy traits of prestigious
individuals that are not the reason why they are prestigious, such as Clooney’s
presumed coffee-​t asting abilities(cultural hitchhiking). We can also copy the
traits that are the genuine reason why the prestigious individuals are pres-
tigious, but they are not useful in our local environment (such as the acting
abilities of Clooney—​missing-​t arget problem). The two last shortcomings es-
pecially may be considered a problem in our contemporary society. If digital
connectivity allows for expanding the number of individuals we can copy
from, as it does, we may end up having less knowledge of the associations be-
tween cultural traits and reasons for prestige, and we are exposed to individ-
uals belonging to every kind of different societies or professions, whose traits
may not be useful for us.
However, several studies show that this is not the case: we do not follow prestige
cues blindly, as this would be detrimental. According to the “presumption of good
design” we discussed in Chapter 2, our cognitive abilities are appropriate to a spe-
cies that lives in groups, where social interactions, and communication in general,
are of fundamental importance. The influence of prestigious individuals is medi-
ated by various factors. First, there are factors related to the source of information.
While the terminology varies, we copy prestigious individuals who appear com-
petent and benevolent. Competent individuals look like experts in their domain,
and benevolent individuals are trustworthy. We are rightly suspicious of celebrities
advertising unrelated products, or politicians defending what appear to be their
own interests.39
 Prestige in perspective  69

We may copy traits from celebrities who are not necessarily competent, or ben-
evolent, but when this happens we usually take into account content factors: the
cost of the traits as well as our previous knowledge or preferences. Prestige-​driven
cultural diffusion often involves traits that are by and large neutral, non-​costly, for
the copiers, such as popstars’ hairstyles, or that overlap with individuals’ interests,
like cancer screenings, or preferences, maybe popstars’ lifestyles, if one can afford
it. All these factors interact, unfortunately for simple explanations but providen-
tially for real-​life dynamics, in complex ways: we may copy something that is not
very costly if we have some previous knowledge and if the source appears expert
enough. This is not, of course, a perfect mechanism: errors do happen, but it should
guarantee outcomes that are good on average.
Research on online influence add to this the interesting possibility that “local in-
fluencers” could have an important role. The concept sounds like one of the many
interchangeable marketing buzzwords, but it is indeed exactly what we would ex-
pect from a conception of prestige bias as described above.
Cultural hitchhiking and the missing-​target problem may always have been an
issue, even though only in exceptional cases, even for hunter-​gatherers. We should
expect that a copying strategy based on the characteristics of the sources, such as
prestige bias, would be particularly sensitive to cues that are activated when the
source is not part of a restricted, trusted, circle, as described earlier on in the book.
Digital, online, interactions make such sources available to us: the possibility of
copying inadequate or dangerous cultural traits is real, but it may also be that, un-
expectedly, we become more attuned to signals of benevolence and competence
with respect to what happens in our life offline.

Notes

1. YouTube version of the original Shiggy Instagram video:  https://​www.youtube.com/​

watch?v=yPdI26ghCvc
2. https://​www.billboard.com/​articles/​columns/​hip-​hop/​8464402/​drake-​in-​my-​feelings-​
dance-​challenge-​videos, Odell Beckham Jr.: https://​www.youtube.com/​watch?v=CKfF_​
UjO6pg, Will Smith:  https://​www.instagram.com/​p/​BlI86RBnfSA/​?hl=en, Data from
https://​osome.iuni.iu.edu/​tools/​trends/​, Davis et al.  (2016)
3. https://​www.theguardian.com/​music/​2018/​jul/​30/​kiki-​keke-​challenge-​drake-​police-​
warn-​dangerous-​viral-​dance, https://​w ww.dailymail.co.uk/​news/​article-​5982419/​
Spanish-​police-​urge-​stop-​feelings-​viral-​challenge-​MOVING-​cars-​stunt-​goes-​wrong.html
4. https://​j ingdaily.com/​c hina-​i nfluencer-​f atigue/​, https://​w ww.instagram.com/​p /​
Bs8waFIgyxt/​
5. Hewlett and Cavalli-​Sforza (1986)
6. Mesoudi (2011)
7. Stibbard-​Hawkes et al. (2018)
8. Jimenéz and Mesoudi (2019)
70 Prestige

9. Henrich and Gil-​White (2001), Henrich (2015), Boyd and Richerson (1988). Notice that
Boyd and Richerson consider in their model the relation between individual fitness and
cultural traits for the effectiveness of a more general “indirect bias.” Indirect bias includes
any form of assessment based on the features of the individual to copy from (as opposed
to direct bias, in which the assessment is based on the features of the cultural trait).
10. Atkisson et al. (2012), Chudek et al. (2012)
11. Henrich and Broesch (2011)
12. Reyes-​Garcia et al. (2008). For a general assessment of social learning biases in ethno-
graphic records, also finding a relative small role of prestige, see Garfield et al. (2016).
13. Chudek et al. (2012)
14. Boyd and Richerson (1988)
15. Henrich and Gil-​White (2001), Mesoudi (2014)
16. Boyd and Richerson (1988)
17. Boyd and Richerson (1988)
18. Acerbi and Tehrani (2018)
19. Boyd and Richerson (1988), Coultas (2004), Claidière et al. (2014)
20. Boyd et  al. (2011), Ozzy Osbourne:  https://​ www.youtube.com/​ watch?v=uMs7-​
MQb5as, Snoop Dog:  https://​www.youtube.com/​watch?v=5pPAOy_​gMNQ, Stephen
Hawking: https://​www.youtube.com/​watch?v=dUpkGjHC_​tE,
Dos Equis: https://​www.youtube.com/​watch?v=U18VkI0uDxE, Apple’s 1984: https://​
www.youtube.com/​watch?v=axSnW-​ygU5g
21. Petty et al. (1997)
22. Menon et al. (2001)
23. Amos et al. (2008)
24. Knoll and Matthes (2017)
25. See also Erdogan (1999), Bush et al. (2004), McCormick (2016)
26. Verpooten and Dewitte (2017)
27. Mawdsley (2016)
28. Evans et al. (2014, 2015)
29. Lieberson (2000)
30. Ghirlanda et al. (2014)
31. For Scientology membership estimates see the 1990, 2001, and 2008 surveys in http://​
commons.trincoll.edu/​aris/​
32. Bakshy et al. (2011)
33. E.g. Goldenberg et al. (2009). For a review of various strategies used to identify influen-
cers in online social networks see: Probst et al. (2013).
34. See e.g. Cha et al. (2010).
35. Spry et al. (2011), Jin and Phua (2014), Van Norel et al. (2014)
36. Ferris (2010), Djafarova and Rushworth (2017)
37. https://​digiday.com/​marketing/​micro-​influencers/​, https://​www.nytimes.com/​2018/​
11/​11/​business/​media/​nanoinfluencers-​instagram-​influencers.html
38. Watts and Dodds (2007)
39. For these distinctions see again Sperber et al. (2010).
 4
Popularity

Saint Matthew and beyond

This chapter tackles two broad, and related, questions. The first concerns whether
popular things have a special advantage on the web. Both critics and advocates of
the digital age emphasize that online success generates a few big winners—​today
the like of Google, Amazon, Katy Perry, and so on—​and a long tail of relatively un-
known players—​me and probably, you, and the thousands of search engines that
you never heard of, because they failed or because few people use them. Critics
generally see this as a substantial drawback, while advocates notice that, if nothing
else, you and me are part of a long tail, whereas until a few years ago our possible
influence would have been even more limited (this is what I defined as reach at the
beginning of this book), and, if nothing else, the other cultural traits present in the
long tail become available to us (as explored in Chapter 1). Is this situation a pe-
culiar feature of web popularity? And, more generally, what do cultural evolution
theory and cognitive science tell us about how popularity influences our choices?
The second question concerns the effects of the proliferation of explicit cues of
popularity in digital media. Put simply, until a few years ago the way to assess if an
idea was popular was to see how many people around us were talking about it. One
could notice if everybody at school started to skate, or wore their backpack on two
shoulders instead of on one, and react accordingly. Today we still do this but the
internet, and social media in particular, provide explicit, immediate, and quanti-
fied, cues of popularity. We can still pay attention to how many of our friends are
talking about a certain topic, but we can also see that the post about that topic got
a certain number of likes or has been shared a certain amount of times and so on.
We see what is trending in real time, or the top-​N most popular posts of the day.
These features, as mentioned, are not uniquely related to the web, but they are, as
we will see, enhanced and partly transformed by it. We can see how many people
gave five stars to a book, or a movie, or a hotel, or, well, everything. Today I bought
a ring pillow, that is, a small pillow on which to place wedding rings and, of course,
I checked which ones were the most popular, and looked at their reviews.
Let’s begin to see what popularity looks like on the web. How things are dis-
tributed can give us a lot of information about their characteristics. Think about
height: the average height for males in Europe is between 170 and 180 centimeters.
This means that if you meet a random European male it is quite likely he will be that

Cultural Evolution in the Digital Age, Alberto Acerbi. Oxford University Press (2020) © Oxford
University Press.
DOI: 10.1093/oso/9780198835943.001.0001
72 Popularity

tall. According to the Guinness World Record, the tallest living person is a Turkish
man named Sultan Kosen, and he measures 251 centimeters. If Sultan and myself
were stood side-​by-​side the difference would be impressive (I am a good represen-
tative of the average European, if anything slightly on the shorter side). He would
be almost one third taller than me.1
Now think about wealth. The average net monthly salary in the European Union
in 2017 is between 1,500 and 2,000 euros. However, the average net monthly salary
for Moldavians is around 200 euros, and the average net monthly salary for Swiss
people is more than 4,000 euros. So, the average Swiss earns 20 times more than
the average Moldavian. I  am quite lucky, and my salary is slightly higher than
the average for the Netherlands, where I work, but if I was to meet the economic
equivalent of Sultan Kosen, their salary would be probably higher than mine in the
order of thousands of times, not one third.2
Moreover, wealth accumulates. If you have a high salary today, it is likely you
have had a high one in the past too. More money makes for more possibilities
of investment, which makes, in turn, for even more money. Finally, wealth goes
through families: as disheartening as it is, the best predictor of your wealth in the
United States, is how wealthy your parents were, and this holds for the majority of
other countries. Italian economist Vilfredo Pareto was the first to realize that the
inequality in the distribution of wealth followed some regularities: he calculated
that at the end of the 19th century 80 percent of the Italian land was owned by only
20 percent of the Italian population (with the remaining 80 percent owning the
20 percent of the land).3
Height, as many other things, going from the amount of hours we sleep at night
to the temperature in Rome in July, roughly approximates normal distributions.
For our argument, what matters here is that their averages are quite representative
of the overall samples, and the extremes are relatively not too far from the aver-
ages. Some people sleep more and some sleep less, but the majority of people sleep
around seven or eight hours, and big sleepers do not sleep hundreds of hours, as
they would do if sleeping time was distributed like salaries. Wealth, conversely, as
many other things, going from the magnitude of earthquakes to -​you expected it
-​the popularity of social media services and websites, are skewed, or long-​tailed,
distributions. In these situations, averages are not very representative of the overall
samples, and the extremes are very far from the average. There are very few, very
big, earthquakes and very few, very popular, websites and many small earthquakes
and many unknown websites. The average is not representative. On the one hand,
big earthquakes and extra-​popular websites are many times bigger or popular than
the average. On the other, in contrast to picking random European adult males, if
one takes a random website, its traffic will be lower than the average traffic, because
Google and Facebook have skewed the average toward high values.
While some skewed distributions result from purely physical processes, such
as earthquakes, many of them have a characteristic behavioral signature. They
 Saint Matthew and beyond  73

are sometimes called winner-​take-​all distributions. In the words of Matthew’s

gospel: “For to every one who has will more be given, and he will have abundance;
but from him who has not, even what he has will be taken away.” YouTube videos
are a good example of these skewed distributions. Let’s take, as an illustration, the
popularity of Johann Sebastian Bach’s works on YouTube. Bach was an exception-
ally prolific composer: he authored more than a thousand compositions, some of
them still widely known even beyond the classical music enthusiasts. Figure 4.1
shows the number of views for the most popular 1,000 compositions. The views are
averaged across various videos as the same composition is often present in more
than one YouTube video. Finally, the compositions are ranked in order of popu-
larity, with the most popular on the left side of the figure.4
The most popular composition is the Orchestral Suite No. 3 (BWV 1068) with
more than 20 million views (the title may not tell you anything, but if you listen to
its second movement, the Air, you will recognize it at once) closely followed by the
Toccata and Fugue (BWV 565), as heard in Walt Disney’s Fantasia. The third ranked
piece is already at around six million views, less than one third of the first two. The
majority of compositions, however, have far fewer views than the hits: they look, in
the plot, like a straight line—​the long-​tail of the distribution. The dotted straight
line close to the zero of the y-​axis is the average, at 168,597 views. This information
does not tell us much about the Orchestral Suite No. 3, which has been viewed 130
times more than this figure. At the same time, the majority of compositions have
been viewed less than what the average tells us: the median is 23,824 views.

20
Number of views (millions)

15

10

1 250 500 1000

Rank

Figure 4.1  Count of YouTube views of the most popular 1,000 compositions of Johann
Sebastian Bach.
74 Popularity

These skewed distributions are caused by the amplification of the success of al-
ready popular items, but it would be inappropriate to consider that some features
of digital media are specifically responsible for this. Alex Bentley and colleagues
have shown that the distribution of traits in many cultural domains is a long-​tailed
one. Parents often spend a lot of time thinking about the name to give to their new-
borns. Very personal considerations, family histories, are all weighted in a choice
that supposedly reflects unique concerns and desires. However, taking a bird’s eye
view, the popularity of names displays the same regularity of earthquakes, salaries,
and YouTube videos: a few are very popular and many are not. This works on dif-
ferent scales. If one takes, for example, the last hundred years of data we have about
US male names, the most popular four—​James, John, Robert, and Michael—​
largely account for 10percent of all the names given to boys. Around the hundredth
ranking position —​Logan, Harry, and Bobby are here—​the summed popularity of
ten names is less than the popularity of one of the first four. You would be around
15 times more likely to find in the archive a James than a Harry (see Figure 4.2).
With some caveats, it also works considering single years. The difference here is
that the distributions recently became less skewed, suggesting that parents are now
more inclined to choose what they perceive as relatively original or rare names.
When I was born, a few decades ago, Michael, the name ranked first, accounted for
4 percent of all male names. In 2017, only slightly less than 1 percent of baby boys
were called Liam—​the most popular name.5
The same happens in many cultural domains: some dog breeds are very, very,
popular, such as Labradors or German Shepherds, and many are not; few pop

5
James
John
4 Robert
Michael
Total count (millions)

Logan, Harry, Bobby...

1

1 250 500 1000

Rank

Figure 4.2  Count of given male baby names in the US in the past 100 years.
 Saint Matthew and beyond  75

musicians reach quasi-​universal success and many have just a niche following;
few scientific papers are widely cited and many remain relatively obscure. With
this perspective, what we observe online is not that special: it is just another in-
stance of a more general phenomenon that tells us that cultural success is hard to
achieve and self-​reinforcing. Bentley and archaeologist Stephen Shennan found
the same kind of distribution for decoration styles, such as incised lines, strokes,
and their combinations in pottery from Neolithic farming settlements along the
river Merzbach, near Bonn in western Germany. Again, few decorations were very
popular and many were by comparison rare. One can safely exclude the influence
of digital media here.6
There is another interesting aspect for us. What kind of individual behavior do
we need to postulate to explain these distributions? At first glance, they look like
the result of a tendency to follow the crowd, to prefer popular options against un-
popular ones. As with the Asch experiment we encountered earlier in the book,
people are supposedly avoiding making decisions by themselves under the pres-
sure of conformity. However, that is not the case. The models by Bentley and col-
leagues show that these long-​tailed distributions are the result of pure availability. In
fact, these models are usually called “random copying” or “unbiased transmission”
models. In the models, at each time step, the modeled individual simply picks an-
other individual at random and copies their cultural trait (or, with a low probability,
introduces a new invention). Imagine you want to choose a name for your daughter.
Go out in the street and ask the name of the first woman you encounter. Done. This
process suffices to create a situation in which some names will become very popular,
independently of their intrinsic characteristics but, also independently of any bias
toward successful or popular items, or any tendency to conform. It is just that small,
random, initial differences will tend to accumulate: if you are slightly more likely
to encounter a woman called Emma, you will be slightly more likely to call your
daughter Emma which, in turn (with some simplifications) will make other people
more likely to meet an Emma in the street, and so on. Of course, this does not mean
that we choose our children’s names at random. This would go against a common
intuition, and in this case, the common intuition is perfectly right. It simply means
that at a low-​resolution scale, the various individual decisions cancel each other out,
producing the general pattern we observe at population-​level.
As we discussed in the previous chapter when considering the role of celebri-
ties in spreading cultural traits, availability is an important determinant of cultural
evolution. Simply, you cannot call your daughter Emma—​or it is very unlikely—​if
you have never heard of anyone called Emma. Moreover, the more Emmas you
encounter, the more the cultural trait Emma will be, as it were, at your mind’s
disposal: you will remember it more than other names, you will consider its sub-
jective advantages and disadvantages more than for other names, you will chew
over it more than over other names. Availability works like in the physics-​or com-
puter science-​inspired models we quickly mentioned when introducing cultural
76 Popularity

transmission biases: nothing more than the number of times of exposures deter-

mines the success of a trait.
Online success is strongly unequal. However the distributions we observe do
not tell us that this is a specificity of the digital age: neolithic pottery decorators
may have lamented the same! Neither this inequality is due to a conformist ten-
dency or to a general tendency to prefer successful traits. We will explore now what
the signatures of a tendency to actually prefer popular traits are, and whether we
can detect this tendency in online data.

Conformity and popularity

Cultural evolutionists use the term conformity in a technical sense, and for good
reasons. The first is that “following the majority” does not require any special
tendency. Random choices, as we just discussed, result in self-​reinforcing popu-
larity. Conformity, on the other hand, requires that individuals actually prefer the
popular choices to the unpopular ones. Imagine you are in a café in an unknown
town, and you do not know what to have. There are 50 others patrons: 30 of them
are sipping a red-​colored drink, and 20 sit in front of a glass with something black
inside. What do you order? The red drink is the majority choice, so you may go for
that one. However, as we just saw, if you choose randomly—​remember how the
name Emma was picked—​you will also be more likely to choose the red drink, pre-
cisely with a 60 percent probability. To be conformist, in the sense used by cultural
evolutionists, your probability to order the red drink should be more than 60 per-
cent. Technically, a conformist bias is defined as a disproportionate tendency to
copy the majority or, similarly, as the tendency to copy the majority’s cultural trait
with a probability higher than its frequency.
Notice this is also different—​very different—​from how the term is used to de-
scribe the results of Asch’s “conformity” experiment that we discussed earlier,
where participants gave a wrong, but popular, answer to the question of the ex-
perimenter. In that setting, all other individuals are confederates and provide the
wrong answer, so that, in fact, simple random copying would predict that all par-
ticipants would also answer wrongly, while only (so to speak) around one third
of the answers were wrong. In a variation of the experiment, interesting from our
perspective, Asch replaced a confederate with a true participant, or instructed a
confederate to give the correct answer. When this happened, he found that subjects
answered incorrectly slightly less than one in ten times. In this case, a majority
of around 80–​90 percent (six confederates out of seven) produces a “probability
to copy” of around 10 percent, far from the strict requirements of conformity as
defined above. (Of course, the fascinating part of Asch’s experiment is that parti-
cipants had the right answer under their eyes, so that random copying, differently
from the case of the red and black drinks, is an unrealistic baseline.)
 Conformity and popularity  77

A characteristic signature identifies conformity in cultural evolution (see

Figure 4.3). When plotting the probability to copy a trait (think again the red
drink) versus its frequency, the function linking the two is s-​shaped. It starts
slowly, with unpopular traits being copied with a probability lower than their fre-
quency, and then increases: when a trait is in the majority, the probability becomes
higher than its frequency. It is instructive to compare this with random copying
(the dotted line in Figure 4.3). The function is now linear, with the probability of
copying following the frequency of the trait. Popular traits are still advantaged,
but just because they are more likely to be encountered.
Second, one may ask, why is this important? Joe Henrich and Rob Boyd have
discussed how conformity may reinforce the strength of individual learning ten-
dencies at the level of population. Consider that, as we generally have assumed,
individual learning is adaptive on average, but it is also error-​prone. Imagine
you need to choose whether to buy a new energy-​saving oven for a higher initial
price or a standard model for a lower initial price. In the first case you spend more
money at the beginning, but you expect to save it on the long run, and the op-
posite is true for the latter. You can decide that the energy-​saving oven is the better
choice. Estimates of energy spending and usage are not obvious, so your individual
learning process—​deciding by yourself which type of oven to buy—​will be correct
on average, perhaps with a probability of 60 percent. This means that, in a popu-
lation of individual learners, 60 percent of households would have made the same

1.00

0.75
Probability to copy

0.50

0.25

0.00

0.00 0.25 0.50 0.75 1.00

Frequency of the trait

Figure 4.3  The logic of conformity and random copying.

78 Popularity

correct choice. Now, if you copy at random, you have exactly the same probability
of being correct, but if you are conformist, the probability is, by definition, higher
than 60 percent, so it is better than deciding by yourself. Of course, in real life, the
situation is complicated, among other things, by the fact that other people may
also have copied the choice, so that the 60 percent of the population choosing the
energy-​saving oven does not reflect the proportion of individual learners, thus
making conformity a risky strategy. We will come back to that later.
In addition, it has been proposed that conformity may be one of the forces
necessary to maintain stable cultural differences between groups, together with
others, such as third-​party punishment of norms’ violators, different environ-
ments requiring different adaptive solutions, and so on. Everything else being
equal, the proportion of patrons sipping black and red drinks in our imaginary
café will remain constant with unbiased random copying. Every time a new
customer enters, there is a 60 percent probability they choose red, and another
patron goes out. There is nothing in principle that will change the proportion
between the colors (not considering that in such a small population drift will be
powerful, but it does not matter for our example). However, imagine that there
is now a competitor around the corner, where patrons prefer black drinks. If
there is even a very small probability that, from time to time, a customer of the
“black” café will go in the “red” one and vice versa, random copying will result,
slowly but surely, in the blending of the traits, and in the loss of the identity of
the two cafés. With conformity, on the other hand, the tendency, in the absence
of other forces, is for the majority trait to increase in frequency, so that small
perturbations, such as the customers who change their favorite café, will be
counterweighted. The two cafés are, in real life, human groups, and the moving
patrons represent migration, intermarriage, commerce, or more gloomily, wars
and invasions:  in sum, any possible intergroup contact. Since some level of
intergroup contact is always present among humans, conformity provides, in
theory, a powerful explanation of the maintenance of cultural differences, espe-
cially not directly functional ones -​one does not need to invoke conformity to
explain why people living at the tropics dress differently from people living at
higher latitudes, for example.7
If random copying produces skewed, long-​tailed, distributions with few take-​all
winners, an actual preference for popular cultural traits, like the one implied by
conformity, should produce popularity distributions that are even more skewed.
This is indeed what Alex Mesoudi and Stephen Lycett found by analyzing the re-
sults of computer simulations where they modified the “random copying” model
of Bentley and colleagues we described above. In their modified model, a variable
proportion of individuals adopted by default only the most common trait in the
previous generation, instead of picking one randomly. Even when only 20  per-
cent of individuals were conformist, a single trait ended up dominating the cul-
tural marketplace, with an increased distance in popularity in respect of the traits
 Conformity and popularity  79

ranked below it. To illustrate, in the case of names, this would resemble a situation
in which not only the majority of baby boys are called John, but also the ones called
Henry (the name ranked second) are sensibly less than what we had with random
copying.8
To substantiate the claim that digital media amplifies popularity-​based social
influence, online popularity distributions should look like the even-​more-​skewed
distributions produced by conformity. As mentioned above, there is a broad con-
sensus that the popularity of websites, measured as the number of visitors or as the
number of external links received, is indeed a long-​tailed one. The same goes for
the number of social media users. However, these distributions resemble the dis-
tributions of baby names, or dog breed popularity, neolithic pottery decorations,
and so on, more than the distribution found by Mesoudi and Lycett for conformist
social influence.9
Some other distributions do look more skewed: Google dominates the search
engine market similarly to what is predicted by Mesoudi and Lycett’s model,
being a single “winner” that indeed “takes-​all.” In December 2018, the worldwide
market share of Google was more than 90 percent. Those that rank next—​bing,
Yahoo!, Baidu, etc.—​get only a tiny piece of the cake, with 2 percent or 1 percent
of share.10
However, it is also possible to argue that, for many domains, online popularity
distributions are less skewed than the ones predicted by the random copying
model. Using the number of external links received as a metric of popularity, when
pooling together all and only universities’ homepages, or newspapers’ homepages,
or scientists’ homepages, it looks as if, in these cases, winners are more magnani-
mous. Whereas the highest ranked entries still have a considerable advantage in
terms of inbound links, low and middle-​ranked entries do better than what would
be predicted by random copying: while a few universities, or newspapers or sci-
entists, receive many external links and a few receive very few external links, the
majority of them cluster around an intermediate, average, number between 100
and 1,000. The popularity of university homepages looks more like the distribution
of height. There are few Sultan Kosens (the 251 centimeters-​tall man), but middle-​
range universities are also relatively popular. If you live in Knoxville, the University
of Tennessee is as much, or more, important than Harvard and the MIT.11
The data for the analysis just described were collected in 2001, eons ago in
digital-​age time, and the situation may be different today. It is conceivable, for ex-
ample, that social media, by filtering increasing amounts of web traffic, would pro-
duce different, more skewed, distributions. Perhaps the homepage of the University
of Tennessee is indeed losing ground with respect to Harvard. We do not know.
What we do know, however, is that we do not have robust support for the idea that
digital media always boosts the popularity of the big winners at the expense of all
other players. Big winners win, but as they do in the majority of cultural domains
and, if anything, the other players can at least participate in the game.
80 Popularity

The strength of conformity

As in the case of prestige, which we examined earlier, cultural evolutionists have

started to perform experiments to test whether we are conformist, in the technical
sense defined above (of course, several other researches have been carried on in
social psychology following Asch’s steps but, as just discussed, the definitions of
conformity do not always correspond). And, as in the case of prestige, the answer
to the question “are we conformist?” is quite nuanced: it depends.
I mentioned earlier experiments where participants were influenced by others
in choices such as covering or not the keyboard of a public computer they used,
or writing a date analogically (January 25, 2018)  or numerically (1–​25-​–​2018).
Julie Coultas, a psychologist at the University of Essex, ran the former experiment,
with almost 400 students as participants in a naturalistic setting over three years.
She cleverly took advantages of an everyday situation. Students in psychology
had to take practical computer classes, and she had noticed that none of the stu-
dents covered the computer keyboard at the end of the classes. The participants in
the experiments were thus simply these students (they were told afterwards that
they were participating in an experiment). For the experiment, she placed on the
screen of some computers a note that read: “IMPORTANT—​Please place keyboard
covers on top of the computer.” The students sitting at these computers followed
the guidance of the note, unknowingly becoming confederates. By manipulating
the number of computers with the note, it was possible to modify the “frequency of
the trait” and, in this way, to check whether the reaction of the students (the “prob-
ability to copy”) followed the s-​shaped curve depicted in Figure 4.3.
Students were influenced by the unaware confederates, and also by their
number, but the relationship between the proportion of confederates and the pro-
portion of students covering their keyboards was hardly following the s-​shaped
curve that indicates conformity. With minor exceptions, students covered the key-
boards only when the frequency of confederates started to be higher than 60 per-
cent (the 14 percent of students at this proportion) and, after that, the “probability
to copy” increased more or less linearly until reaching a maximum of 48 percent
when the frequency of confederates was higher than 90 percent. According to the
conformity logic, after the majority is reached, the probability to copy should be
higher than the frequency of the behavior, so, for example, when the frequency
of confederates was higher than 90 percent, more than 90 percent of the students
should have covered their keyboard.
The catch here is that covering the keyboard was a rare behavior: no student
did it if they were not prompted by seeing others or by the notes specifically
placed by the experimenter. This is quite a different situation from the ideal, neu-
tral, scenario we described—​the café that serves two drinks, of which you have
no previous knowledge, or preference. Thus, in the second experiment, about
the writing a numeric or an analogic date, Coultas controlled for pre-​existing
 The strength of conformity  81

preferences. The students had a strong propensity toward one of the two behav-
iors: when no other indication was present, 80 percent of them wrote the date nu-
merically (“1–​25–​2018”), and only 20 percent analogically (“January 25, 2018”).
As before, unaware students had to fulfill an everyday task, in this case signing a
consent form and then a confirmation that they had received payment for taking
part in an unrelated experiment. The first signature was to determine if the stu-
dents, with no influence, were writing the date numerically (the “social norm” in
Coultas’ paper) or analogically (the “rare behavior”). The second signature was
the actual experiment, with a manipulation of the number of previous dates the
students could see.
For students who wrote the date numerically, and who were exposed to a
various proportion of analogic dates (thus switching from “social norm” to “rare
behavior”) the results were, as expected, in line with the previous experiment: a
circa-​linear increase in the proportion of copiers, never reaching an higher-​than-​
the-​observed-​frequency proportion, the mark of conformity. In the other case,
however—​students going from analogic to numeric date, or from “rare behavior”
to “social norm”—​the results supported the conformity hypothesis. When they ob-
served a majority of dates written numerically, students tended disproportionately
to switch: after 70 percent of frequency had been reached, all students switched to
writing the date numerically.12
Other experiments in naturalistic settings also uncovered similar tendencies.
Nicolas Claidière and colleagues conducted a public engagement activity in a re-
search center at Edinburgh Zoo. The visitors were invited to share their thoughts
and ideas about the center, with the possibility of winning a small prize. They had
at their disposal cards with some general questions on them (“What do scientists
do?,” “Do you know something interesting about monkeys?,” etc.) and colored
pencils. The visitors could see the display panel with the previous answers: some
of them were written and others were drawn. Claidière and colleagues manipu-
lated the proportion between answers with drawings and answers with text, and
recorded the visitors’ choices.
In this case as well, however, visitors brought their own previous preferences
into the experiment: writing was the most common choice. Even when the vis-
itors saw only cards with drawn answers in the panel, around 40 percent of them
wrote on their cards, thus going against an absolute majority. Again, the influence
of the perceived proportion was not absent: when the proportion of written an-
swers displayed increased, the proportion of visitors writing their answer also in-
creased, reaching between 90 percent and 100 percent when all the answers on the
panel were written. In addition, this increase was linear (quite confusingly, in my
opinion, Claidière and colleagues calls it “linear conformity,” and they use the term
“hyper-​conformity” for what we—​as does everybody else in cultural evolution—​
call “conformity”), suggesting that the effect could be explained by a combination
of a general preference for writing, instead of drawing, the answers, plus random,
82 Popularity

unbiased, copying, that reflects simple availability of more examples of one kind or
another.13
In another experiment, Julie Coultas and Kimmo Eriksson tried to take into ac-
count the fact that different behaviors come with a different propensity, and to
explicitly include this propensity as a baseline in the analysis. They presented to par-
ticipants, recruited in a shopping mall, two-​column forms with an initial question,
such as “Eating garlic protects you from catching a cold,” or “Volvic bottled water
is better than Evian bottled water.” Participants could see the previous number of
choices, manipulated by the experimenters who again, were checking how the pro-
portion of previous answers influences the participants’ answers. In addition, the
experimenters used as a baseline the proportion of answers given when the two col-
umns were presented as empty, which exemplifies a condition free of any social influ-
ence (if you want to know, for the cases above, 65 percent of participants thought that
garlic protects from colds, and 68 percent of participants preferred Volvic to Evian).
The results are not surprising from what we have seen so far: yes, more examples
of one answer led to more answers in the same direction, and, no, participants were
not conformist, so that the effect is likely to have arisen simply on the basis of more
availability. In fact, the analysis of Coultas and Eriksson suggests the existence of
“a general tendency of decreasing marginal impact of the numerosity of sources
of social influence:” in other words, a big majority is proportionally less likely to
“convince outsiders” than a small group, which is, effectively, the opposite of what
conformity predicts.14
A few other studies have investigated conformity among adults, in a cultural evo-
lution perspective, in more traditional laboratory settings. It would be tedious going
through all of them. Participants had to play computer games with questions and
could see (manipulated or not) previous participants’ answers. Participants played
games where they had to decide which “technology” they wanted to use (presented as
an abstract choice: the “red” or the “green”) or which “crop” they wanted to cultivate
(“wheat” or “potatoes”) with yield depending on variable environmental conditions.
The results tell, overall, the same consistent story: with rare exceptions, participants
do use frequency cues but not disproportionately, as predicted by conformity, and,
mostly, their choices are more complex to model, and they depend on various factors.
Here are just some of the results. Participants copied the majority only when
they perceived they were similar to them, the logic being that if someone is similar
to you, or in a similar situation, they will perform a behavior tuned to that specific
situation and thus useful to copy. Participants were primarily evaluating pay-​off
cues (how good is my behavior?) and only after that, if the evaluation was negative,
considering what the majority was doing. Participants had a different idiosyncratic
behavior, with some of them resorting consistently to frequency information and
others (“mavericks”) almost always discarding social cues, as we saw early in the
book. Participants did not follow majority cues when they were thinking the task
was easy or that they were able to solve it by themselves. You get the gist.15
 Popularity made explicit  83

In general, to sum up, in line with the discussion on wary learners earlier in the
book, and with what we said about prestige cues, worries about our online herd be-
havior may have been overestimated. There are good reasons, as we saw, to follow
the majority, but it makes sense not to do this all the time, and this is indeed what
usually happens.
Conversely, there are also good reasons to not always follow the majority.
We mentioned that, assuming that what we learn individually is on average
more correct that incorrect, conformity boosts individual learning, giving a higher
probability of acquiring the correct behavior by copying it. However, we also noted
that there is another assumption here: that the other individuals did not themselves
copy others, and each of them learned independently. If this assumption is violated
(as it often is) the risk is that the “wisdom of the crowd” becomes the “madness
of the crowd” and behaviors untested through individual learning, spread uncon-
trollably. These cases—​from Tulip mania in seventeenth-​century Netherlands to
modern day suicide “epidemics”—​are often used to suggest the dangers inherent
in social learning, but the evidence that they are indeed caused by blind copying is
scant, in line with our readings of the experimental literature.16
Another reason why we should not always follow the majority, especially in the
case of relatively small majorities, is that the majority can simply be wrong, or that
the majority can be actively misleading us. This is why it makes sense, as parti-
cipants in the experiments reported above were doing, to consider whether the
majority’s behavior is in line with what we would do ourselves, or if it seems to pro-
vide advantages in respect to alternative behaviors.
Finally, an important factor that determines our reliance to majority that we can
only briefly touch on here, is whether we “do as the Romans” because they do the
correct thing or simply because we want to be like them. In academic jargon these
are known as “informational” and “normative” conformity. Cultural evolutionists
are in general more interested in the former but, since both can act simultaneously,
some experimental results can be interpreted in both ways, including some of the
experiments we described above. Think again of Asch’s experiments: the partici-
pants did not need to be convinced by the confederates, but they could simply say
they were, so as not to look different in this, unusual for them, group. In line with
this, in Asch’s experiments extensions, when participants are aware that their an-
swer is important, they give the correct one, switching from the normative to the
informational aspects of the situation.17

Popularity made explicit

From the beginning of its operation, in 1995, Amazon gave its customers the op-
tion of leaving reviews of the products they had bought. “Many people thought
the Internet retailer had lost its marbles. Letting consumers rant about products
84 Popularity

in public was a recipe for retail suicide”, critics thought. As we now know, the story
has been different. Whereas Amazon does not provide the exact number of reviews
on its products, recent estimates point to a figure of around 250 million. Various
surveys suggest that the great majority of customers give great importance to how
a product is reviewed, and often decide what to buy after reading various reviews.18
In 2009, three years after its public opening, Facebook introduced a feature
inspired by a competitor, FriendFeed. “We’ve just introduced an easy way to
tell friends that you like what they’re sharing on Facebook with one easy click.
Wherever you can add a comment on your friends’ content, you’ll also have the
option to click ‘Like’ to tell your friends exactly that: ‘I like this.’ ” Apparently, the
introduction of the like button was preceded by heated discussions. If the Amazon’s
review system was received with scepticism, as it gave customers the chance to ex-
press publicly their opinions about products, a Facebook like effectively allows an
immediate way to judge the content shared by a friend, which is often a personal
episode of their life. In 2011, Facebook added another public feedback metric: the
number of times a post was shared by users, with, in addition, the possibility to see
which users had shared it.19
A general effect of counters of shares and likes is that they provide an explicit,
immediate, and precise quantification of popularity. If, as cultural evolutionists
predict, we are sensitive to popularity cues, this innovation is interesting for two
reasons. First, social media would piggyback onto a natural tendency we have to
look at these cues, and to find them attractive. We would spontaneously be at-
tracted by these quantifications, and their widespread use would reflect this ten-
dency. Despite occasional complaints about the possible negative effects of the
social pressure generated by the public availability of this information and some
unsuccessful attempts by parts of the social media to ignore these features, explicit
popularity cues are a part of all major social media sites. (As I write it is the singer/​
producer Kanye West, followed by 30 millions on Twitter, warning that “we should
be able to participate in social media without having to show how many followers
or likes we have.”)
Second, as mentioned above, popularity before the digital-​age, had to be esti-
mated from various, often indirect cues (the ubiquitous diffusion of top-​lists pre-​
internet is an interesting exception, and we will get back to it in a moment). Now
this information is in front of us all the time:  we know exactly which “cultural
trait” is more popular and exactly how popular it is. What are the effects of these
changes? On the one hand, we could expect that given that we have precise infor-
mation at virtually no cost, we may be more inclined to follow majority cues. On
the other hand, however, one could speculate that, as these precise quantifications
are a very recent (in evolutionary time) cultural invention, we may be less prone to
use them to decide whether to copy, or like, something.20
Do you remember our experiment on famous quotes, where participants did not
prefer quotes randomly associated to well-​known personalities? We ran the same
 Popularity made explicit  85

experiment with a slightly different set-​up. This time, quotes were not followed by
the supposed author, but by the number of participants who had already chosen
them. As before, we were manipulating this information. Participants, as before,
had to choose the “most inspirational” between two quotes, one that had already
been chosen, say 100 times, and one that had already been chosen, say 300 times.
The numbers were randomly generated, but with the constraint of being approxi-
mately one quarter and three quarters, respectively, of the total (I’ll explain why in a
second). Did participants choose the popular quotes? In this case, contrary to what
happened for prestigious authors, they did: popular quotes were chosen on average
around 62 percent of times, against 38 percent of unpopular ones. Remember the
quotes were randomly assigned each time to the two categories, so that in practice
there were no other differences than our manipulated cues of popularity.
There are a couple of clarifications to be made. Participants were sensitive to
popularity cues, but and this should not be too surprising now, not conformists.
If they had been conformists, as explained above, they should have preferred the
popular quote “disproportionately,” that is, more than 75 percent of them should
have gone for it, and less than 25 percent should have chosen the unpopular one
(this is why we chose a fixed proportion between the previous preferences). But
this did not happen. In addition, consistently with the other experimental results,
popularity cues were not overriding an individual evaluation of the content, but
rather they were working together. As much as in the previous condition, with
famous and unknown authors, the best predictor of a quote’s success was its suc-
cess in the control condition, where the quotes were presented without popularity
cues.21
Luckily, you do not have to trust me based only on my own work. Our results
fit with previous research that found that when popularity is explicitly quantified,
we do use it as a cue to decide whether we like something or not, but we are not
blinded by it. In a rightly renowned Science paper, Matthew Salganik, Peter Dodds,
and Duncan Watts describe their “artificial music market,” a series of large-​scale
experiments they realized with the exact goal of finding how the perceived popu-
larity of a cultural item influences its success. The set-​up reproduced an ordinary
activity in recent years: listening to unknown songs on a website, rating them from
one star (“I hate it”) to five stars (“I love it”) and finally deciding whether to down-
load them or not. In a first series of experiments, participants were randomly as-
signed to two conditions: in the “independent” condition, they saw on the screen
a grid with the names of the bands and their songs, and no further information.
In the “social” condition, the songs were accompanied by the number of previous
downloads. In a variant of the “social” condition, let’s call it “social/​top-​list,” the
songs were accompanied by the number of previous downloads and presented in
an ordered list, from the most to the least downloaded (in the other conditions,
both individual and social, the order of presentation of the songs was randomized).
Participants in both social conditions were further divided into eight “worlds”
86 Popularity

where downloading counts were scored independently. In this way, it was possible
to observe whether songs becoming popular in a world were also popular in the
other worlds: in sum, to isolate the effects of social influence.
The researchers emphasize that the social conditions increased both the in-
equality and the unpredictability of songs’ success. Inequality: the social worlds
were more unequal. There was more difference in number of downloads between
the successful and unsuccessful songs in the social than in the independent con-
dition. In social worlds, popular songs were more popular, and unpopular songs
were more unpopular (remember Matthew’s gospel). Unpredictability: the so-
cial worlds were more unpredictable. The average difference in total downloads
for the same song between different social worlds was higher than it was when
measured between randomly drawn subpopulations in the independent condi-
tion. Imagine you want to predict which songs are popular in the social world
B, with the only information being the popularity of songs in the social world
A. This would be more difficult than predicting which songs are popular in a
group of participants in the independent condition, knowing which songs are
popular in another group of independent participants. in social than in indi-
vidual worlds.
In addition, these effects were amplified in the “social/​top-​list” condition, with
both inequality and unpredictability being around one third more than what was
observed in the “social” condition: top-​lists do have an effect on our choices. The
spreading of top-​lists predates digital media, and it is almost a hallmark of the
broadcasting era (in the United Kingdom the first introduction of a top-​chart pro-
gram on BBC radio dates back to 1957 and the US Billboard Hot 100 went on air
the year after), but it has grown enormously in recent years, with on-​line top-​lists
of virtually everything. From a cultural evolution perspective, top-​lists are not only
sources of cheap estimates of popularity, but they also supply an explicit way to im-
plement a conformist bias, by effectively trimming off unpopular items from the
set of possible choices.22
The social conditions in the experiments of Salganik and colleagues look much
similar to the real world. Artists’ pages in the Spotify app, in January 2019, show
prominently the first five most popular songs, with the number of previous plays
close to their titles. Are we living in the “social worlds” of the “artificial music
market” experiment? Is popularity driving our choices at the expense of the quality
of the songs (or anything else)? Not so fast. The independent condition in the ex-
periment provides a measure of songs’ quality, at least in that pool of participants,
and it can be used to check how much the social worlds differ from the non-​social
ones. Success in the independent worlds is correlated with success in the social
worlds, as happened in our quotes experiment. As the authors put it: “the ‘best’
songs never do very badly, and the ‘worst’ songs never do extremely well.” In sum,
perceived popularity can mix-​up things among the “good” songs (we will not
know beforehand which one will become the top-​hit) and it can occasionally push
 Popularity made explicit  87

an “average” song toward the high zones of the chart. This, however, seems a less
worrying reading of the results.23
There are few other considerations that should be made before applying these
results to real-​life dynamics. First, the artificial “worlds” were completely separ-
ated. Whereas this makes much sense from the experimental point of view, it is
an unlikely condition, especially in our digital world (in the next chapter we will
explore research on echo chambers, seeing how this phenomenon may have been
overestimated). Information spillovers from one world to another (“Here is the top
hit in world C!”), would be likely to reduce the differences among different worlds
and to give more opportunity for quality to emerge as a good predictor of songs’
success.
Second, step for a second into a participant’s shoes: you are faced with a choice
for which you do not have any prior information and for which the costs are neg-
ligible. As we discussed earlier, these variables are important, and this looks like a
situation in which going for the popular choice is the obvious thing to do. Imagine
you see unknown names of bands (“The Thrift Syndicate,” “52metro,” “Moral
Hazard”), the only information you can use is indeed, when present, whether
others liked them or not. Add to this that there is some support for the popular
idea that the main factor explaining music appreciation is familiarity, so that the
more you listen to a song, the more you like it, independently from all other char-
acteristics of the song, what is unexpected to me, is that social influence had such a
limited role in the experiment.24
Salganik and Watts, later on, conducted a twisted extension of the same
experiment. Let’s go back to the “social/​ top-​list” condition, where partici-
pants see the band names and their songs ordered from the most to the least
downloaded, in addition to the actual number of downloads. Now, as we
know from the previous results, in these “worlds” there is some level of un-
certainty, but also a good correlation between the quality and the popularity of
songs. The experimenters let the world run until the songs’ ranking reached
a steady state, that is when the songs broadly did not change ranking with the
introduction of new participants. At this point, they cheated the new parti-
cipants, presenting to them an artificially flipped ranking. The song ranked
first in the “real” social/​top-​list condition (“She Said” from “Parker Theory”)
was presented as the last in the chart, while the last ranked (“Florence” from
“Post Break Tragedy”) was shown as the first. The number of downloads was also
swapped. The same was done for all other songs, with the second ranked becoming
the penultimate and vice versa, the third becoming the third from bottom and vice
versa, and so on.
If popularity cues would dominate our choices, the new participants should go
on increasing the (fake) success of “Florence,” and forget about “She Said” (and, if
familiarity is the main ingredient of our music appreciation, they should also start
to like “Florence”). The “cheated” participants indeed listened more to the “fake”
88 Popularity

hits—​as just mentioned, popularity is the only cues they have to decide what to
do—​but, interestingly, the rate of downloads decreased. When “Florence” replaced
“She Said” people started to like less the chart number one. Even more interest-
ingly, “She Said” was not forgotten at the bottom of the list: the rate of download
for the last position of the chart started to increase. Participants generally, when
noticing that the top hit was not that great, and that they could find good songs in
the last positions, discarded popularity cues.
There is another remarkable effect of the artificial inversion of the ranking. As
said before, participants tend to listen to highest ranked songs more but, as they
generally did not like them in the “flipped” world, they downloaded them less. This
decrease is not fully recovered by the increase in downloads of the lower ranking
songs, because they are, in general, listened to less. The outcome is an overall de-
crease of downloads in the flipped world: in respect to the “real” social/​top-​list
condition, participants downloaded a quarter fewer songs.25
The general message of these experiments is that, while popularity cues do have
some importance, they do not act in a vacuum: they interact in a complex way
with personal considerations and with the actual quality of the cultural traits. Since
this happens for the download of unknown songs, it makes sense that it will also
happen for more costly choices. An interesting aspect—​we will come back to it
in the chapter Misinformation—​is that social cues, like popularity, can be easily
manipulated and can have stronger effects than the intrinsic quality of cultural
traits. One can flip a top-​list and this will give a formidable boost to the low-​quality
songs. However, the effects of intrinsic qualities is consistently in the same direc-
tion as, in many cases, it depends on quite general and stable features such as in the
case of songs, harmonic and rhythmic properties. While there is plenty of histor-
ical and transcultural variation, in simplified experimental scenarios it is relatively
easy to find “good” and “bad” songs (as well as quotes, for example). To counteract
the effect of intrinsic qualities, one would need to keep on artificially re-​adjusting
the top-​list. Extrapolating, in a conservative way, from their results, Salganik and
Watts show indeed that the artificial, flipped, ranking would be disrupted on the
long term.26
There is another important take-​home message for our everyday digital life: in
the flipped world, participants were downloading fewer songs than in the world in
which popularity cues were not manipulated. Now, the ultimate goal of platforms
like Spotify is to hook their users and have them to listen to more songs. Similarly,
Amazon wants you to buy more products, and Tripadvisor wants you to go to the
hotels and restaurants that are prominently featured in the website. It would be
against their interest, to actively tweak their rankings too much. They should also,
according to this logic, try to avoid the interested parties—​bands, amazon sellers,
restaurants, etc.—​having the possibility of cheating. At the risk of sounding cyn-
ical, platforms like TripAdvisor do not need to have an “ethical” motivation to, say,
fight against fake reviews, but they simply need to act in their own interest.
 Fake restaurants and bad drugs  89

Fake restaurants and bad drugs

In 2017, a London-​based journalist/​video-​maker/​artist named Oobah Butler de-

cided to pull a prank on TripAdvisor. Having worked as a writer of fake reviews,
for which he was paid £10 each, Oobah decided that the next logical step should
be to fake an entire restaurant, from scratch. He used his shed in south London as
inspiration and, after a quick phone number verification on a cheap mobile bought
for the occasion, “The Shed at Dulwich”, from the name of the neighborhood,
was online, ranked 18,149th out of 18,149 restaurants in the capital of the United
Kingdom. So far nothing too remarkable. However, from these humble beginnings
“The Shed at Dulwich” climbed position after position until reaching, six months
after the initial listing on TripAdvisor, the first place in the ranking, becoming ef-
fectively the top-​ranked restaurant in London. Without existing.
The story was a hit in the press. The take-​home message—​up to a point,
justified—​was how easy gaming the system was in TripAdvisor and, more gener-
ally, how online reviews could not be trusted. However, the story became a sensa-
tion exactly because the “The Shed at Dulwich” top ranking was a rather surprising
outcome. If fake restaurants (or anything else) were common, we would not read
about them in newspapers. In addition, Oobah did everything properly and, from
his own account, it looks like he was involved in the “restaurant” full-​time for all
the months the prank went on. “The Shed at Dulwich” had a professional-​looking
website, with professional-​looking pictures of dishes (they were, in any case, also
fake) and with a professional-​looking menu. The menu ably parodied fashion-
able trends in restoration: “Instead of meals, our menu is comprised of moods.
You choose which fits your day, and our Chef interprets that.” Moods included
“Contemplation” (“A deconstructed Aberdeen stew; all elements of the dish are
served to the table as they would be in the process of cooking. Served with warm
beef tea”) or “Lust” (“Rabbit kidneys on toast seasoned with saffron and an oyster
bisque. Served with a side of pomegranate soufflé”). For obvious reasons, the web-
site did not list opening times or the precise address. Only the street was reported,
and the restaurant was defined as “appointment-​only.” Whereas this was functional
to the hoax, it also fitted with the up-​and-​coming fashion of micro-​or pop-​up res-
taurants. Finally, and probably most importantly, Oobah had hundreds of friends
writing credible five-​stars fake reviews for him over six months.27
The story of “The Shed at Dulwich,” in sum, tells us that full-​blown forgery is
the spectacular exception. As we mentioned before, websites like TripAdvisor
have a “selfish” interest to keep under control the amount of fake reviews, be-
cause rankings that are too far from reality are likely to frustrate users. “The Shed
at Dulwich” opened for a single night, serving outdoors, in November and in
London, microwaved frozen food to a handful of “lucky” visitors. The Vice article
concludes by reporting about a couple who were willing to book again, but—​and
you will not be surprised at this point in the book—​the others guests did not ignore
90 Popularity

their personal experience in favor of the reviews’ social influence. They just left
during the dinner.28
Of course, fake reviews do exist, and we would all be better off without them.
Researchers took advantage of the fact that Yelp, a website similar to TripAdvisor,
signals “suspect” reviews. An algorithm filters, and makes accessible to users, re-
views that are likely to be fraudulent. Sampling restaurants in the Boston area,
16 percent of reviews were filtered by Yelp. This figure needs to be taken with a
grain of salt, but it is still a big number: one or two reviews out of ten are considered
suspect by Yelp’s algorithms. Filtered reviews have several intuitive features. First,
their distribution is bimodal: fake reviews tend to be 1-​or 5-​star with few of them
in the middle, as it does not make much sense to fake an average review. Second,
restaurants that went down in the ranking tend to respond with more fake reviews
in respect to restaurants that went up, quite obviously. In addition, restaurants that
are part of well-​known chains have less fake reviews, since their reputation de-
pends mainly on the chain’s reputation and less on the website’s reviews.29
The openness of Yelp, however, is an isolated case. Virtually all the major review
sites do not allow explicit access to their suspicious reviews nor provide figures
on their proportion. Nevertheless, one can circumvent this limitation, comparing
how the same products are reviewed in websites with different rules. For example,
in TripAdvisor anyone can leave a review. in Expedia, however, which is more of a
meta-​travel agency than a review site as it allows comparing, and booking, offers
from different companies, one needs to have actually travelled, or stayed at an hotel
to qualify as a reviewer. Thus, everything else being equal, we would expect more
fake reviews in TripAdvisor than in Expedia, and this is what happens. A special
case in which hotels could benefit more from fraudulent reviews is when they have
an establishment of similar category in the same neighborhood, as they are likely
to compete for the exact same customers. Using data from hundreds of thousands
of hotel reviews in medium-​sized US towns, researchers found indeed that being
an hotel “with a neighbor” entails having more negative (1-​or 2-​star) reviews in
TripAdvisor than in Expedia, which suggests the possibility of fake reviews. How
strong is this effect? The increase is 1.9 percentage points. This is not necessarily
negligible, since negative reviews are a small proportion of overall reviews—​we
will be back to this in a moment—​so that a 1.9 percentage points increase may still
be meaningful, but it puts in perspective the bigger figure assessed by Yelp (1.9%
is not, however, an estimate of “fake reviews” per se, but of a particular category of
them, i.e. the negative ones).30
Cultural evolution-​inspired research on online reviews is virtually absent. An
interesting exception is a work from Mícheál de Barra about reviews of medical
treatments. As we discussed at length, cultural evolution is expected to produce,
on average, adaptive behaviors. A central question is thus how maladaptive cul-
tural traits spread and, sometimes, thrive (in the chapter Misinformation I  will
elaborate on this point). Medical treatments are a spectacular case of cultural
 Fake restaurants and bad drugs  91

maladaptations, first because of their major repercussions and second because of

their relative success: from bloodletting to contemporary anti-​vaxx parents, it does
not take too much trouble to find pertinent examples.
De Barra focuses on a quite specific aspect of the problem, namely that there is a
“reporting bias” on medical treatments for which we tend to report more positive
than negative results. Benecol, for example, is a brand of food products, especially
yogurts, butter, and soft cheeses, advertised as “proven to lower cholesterol.” And,
according to medical trials, these products do lower cholesterol. De Barra com-
pared however the positive effects reported by the medical trials with the values
reported in the reviews of the same products on Amazon, and he found that the ef-
fects in the reviews were three times greater, that is, more positive, than the average
effect found in the scientific literature. With some differences, the same pattern
held for other cholesterol-​reducing and weight-​loss products.31
In this case, the reviews are neither fake nor fraudulent (de Barra checked this,
and estimated that over 90% of reviews were “high-​quality”), but they are not likely
to consist of a representative sample of the users. In sum, if you used Benecol and
your cholesterol level got better, you are more likely to review the products on
Amazon than if you used Benecol and you did not have any effect. The reasons be-
hind the preference to share more positive than negative outcomes are interesting
(also because, as we will see in the Misinformation chapter, there seems to be a
strong preference to share negative stories online), but what is striking is that this
is just one of several studies that discovered a perhaps surprising general tendency.
At odds with the “everybody is mean on internet” common narrative, online re-
views are consistently skewed toward positivity.
Airbnb, the well-​known online marketplace where members, the “hosts,” offer
to “guests” lodging opportunities in their properties, is an extreme example: the
average rating of a property is 4.7 stars (the median coincides with the maximum,
i.e. 5 stars), and almost 95 percent of properties claim an average rating of either
4.5-​or 5-​star. Virtually no Airbnb properties have less than a 3.5 stars rating.
Airbnb is a particular case, and there are few reasons why ratings are positively
skewed: the most important one is possibly the mechanism of “mutual reviewing,”
for which reviews are posted only if both the host and guest complete a review. In
the worst case, a host who suspects a negative review from a guest can simply avoid
reviewing them. Distributions are, however, also positively skewed in websites
were this mechanism is not in place: the same study calculates a 3.8 stars average
for TripAdvisor, and according to the “neighbor hotel” study we mentioned before,
in TripAdvisor almost 40 percent of reviews are 5-​star (the maximum), and the
average is 3.52. In Expedia, the 5-​star figure goes to 50 percent with an average of
3.95.32
The situation is similar in Amazon, the biggest online marketplace: a 2009 study
(but with data collected in 2005) reports that “78%, 73%, and 72% of the product
ratings for books, DVDs, and videos are greater or equal to four stars.” The same
92 Popularity

study, however, highlights the peculiar distribution of ratings: in Amazon, as well

as in many other websites that offer to the users the possibility of leaving reviews,
ratings are J-​shaped. The same is true today: in Figure 4.4 I plotted more than 4 mil-
lion ratings of movies and TV shows, collected from the beginning of Amazon’s ac-
tivity to 2014. The overwhelming majority is positive, but 1-​star ratings are more
numerous than 2-​stars ratings—​the left tip of the J.33
Why does this distribution emerge when putting together many, and up to a
point independent, reviews? The same we said before in relation to possibly fraudu-
lent reviews also works for authentic ones: there are not incentives to write a review
if you had an ordinary, average, experience with the product you bought. It is, after
all, a costly, if minimally, activity. This reporting bias would explain why middle-​
range votes are underrepresented. Second, if I hurry to watch the last season of
Black Mirror, or to buy again another pair of hiking boots, it is because I already
enjoy science fiction or walking in the hills. People not interested in science fic-
tion, who will be more likely to give low rates to Black Mirror (exactly because they
do not like science fiction) will be also neither looking nor reviewing it. This pur-
chasing bias may contribute to explain the fact that the two tips of the J-​curve are
asymmetric, with more positive than negative reviews.34
Consumer reviews, as well as algorithmically personalized recommendation
systems, have complex effects on users’ preferences and cultural evolutionary
studies specifically dedicated to these effects are lacking. Details may vary, but
what is remarkable is that these studies revel a robust and counter intuitive pattern.
The general picture, with the possible exception of the Yelp study discussed at the
Number of reviews (millions)

0
1 2 3 4 5
Number of stars

Figure 4.4  Distribution of ratings of movies and TV shows on Amazon.

 Fake restaurants and bad drugs  93

beginning of the section, seems to be, on one hand, that fake or fraudulent reviews,
both in big players like Amazon or TripAdvisor, and also often in niche domains
(a study of ratings of beers posted by Finns on the internet concludes that “online
beer ratings made by Finns represent a relatively unbiased source of social infor-
mation for other Finns”), are a minority, that they are often recognized, and that, in
general, ratings are reliable. On the other side, reviews are consistently skewed to-
ward positivity. The main challenge with online reviews is thus not that “trolls ruin
everything” or to recognize fraudulent positive or negative reviews, but to evaluate
among too many upbeat reviews the ones that are only “lukewarm.”35

In this chapter I tried to cast some doubt on the widespread idea that the internet
is the ultimate rich-​get-​richer machine, where a few big players win at the expense
of everybody else taking advantage of our alleged herding behavior. Yes, online
popularity is heavily skewed, but these long-​tailed distributions are a trademark
of many cultural domains, from first names to dog breeds. The Neolithic farmers
we mentioned earlier could similarly complain of how everybody seems to like the
same decoration on their pots. In addition, I showed how these distributions do
not necessarily imply the existence of conformity, i.e. an individual-​level tendency
to prefer popular things. They can be the result of bare availability: the more ex-
amples of an item, the more likely we will encounter it, and the more likely we will
become interested in it.
In fact, we did not discuss at all another aspect, namely that long-​tailed distribu-
tions do not even presuppose necessarily any form of copying: there is no need to
bring up cultural transmission to explain, for example, why wheels are everywhere
circular. The popularity distribution of wheels’ shapes is very skewed (non-​circular
wheels do exist, but they are very rare) but this is due to the fact that the circular
shape is superior to other shapes. Distinguishing the respective role of intrinsic
quality and social influence with only observational data is a social scientist’s
nightmare, and, according to many, the effects of social influence have often been
overestimated.36
When considering conformity, experiments from cultural evolutionists give
a quite nuanced image, which is consistent with what we discussed in the pre-
vious chapter for prestige. We are, by and large, careful learners, and the popu-
larity of something is just one of the many cues we consider when deciding
whether to copy something. Popularity interacts in complicated ways with other
factors, such as the importance of the task at hand, our previous knowledge,
whether or not we have reasons to trust others or not, and so on. Popularity, as
much as prestige, is a good cue, and there are good reasons why we should be
equipped to pay attention to it, but this does not imply that it will override other
important cues.
The fact that online popularity is quantified and made explicit is an interesting
one, and we do not know yet what the outcomes of this novelty will be. What we
94 Popularity

know, however, is again by and large reassuring. Numbers of likes, top hits, and
similar of course have an influence on our choices -​especially top hits, as we have
seen but, as said, they are just some of the many factors that we take into account.
Cues of popularity coming from hundreds of reviews made by strangers are even
odder from an evolutionary point of view, still the initial research shows that re-
views are generally reliable and, if anything, too positive. The take-​home mes-
sage is, in sum, that the availability provided by digital media does not necessarily
bring to popularity-​based cascades of harmful or dangerous cultural traits. In fact,
it turns out to be an unlikely outcome. When this happens, we may predict that
they will be non-​costly behaviors: checking what is Despacito on YouTube, yes; epi-
demics like the alleged “Blue Whale” suicide game, probably not.

Notes

1. http://​www.guinnessworldrecords.com/​world-​records/​tallest-​man-​living
2. https://​en.wikipedia.org/​wiki/​List_​of_​European_​countries_​by_​average_​wage
3. Chetty et al. (2014), Pareto (2014)
4. https://​www.reddit.com/​r/​bach/​comments/​a7cb4v/​bachs_​most_​and_​least_​popular_​
pieces/​
5. https://​www.ssa.gov/​oact/​babynames/​
6. Neiman (1995), Bentley and Shennan (2003), Bentley et al. (2004), Herzog et al. (2004)
7. Henrich and Boyd (1998), Mesoudi (2018)
8. Mesoudi and Lycett (2009)
9. Huberman and Adamic (1999), http://​gs.statcounter.com/​social-​media-​stats
10. http://​gs.statcounter.com/​search-​engine-​market-​share
11. Pennock et al. (2002)
12. Coultas (2004)
13. Claidière et al. (2012)
14. Eriksson and Coultas (2009)
15. Efferson et  al. (2016), McElreath et  al. (2008), Efferson et  al. (2008), Morgan et  al.
(2011). Other cultural-​evolutionary inspired experiments with adult participants rele-
vant to conformity are: McElreath et al. (2005), Efferson et al. (2007), Muthukrishna
et al. (2016), Glowacki and Molleman (2017), Molleman and Gächter (2018)
16. Morin (2016)
17. Toelch and Dolan (2015)
18. https://​www.bloomberg.com/​news/​articles/​2009-​10-​15/​amazon-​turning-​consumer-​
opinions-​into-​gold, https://​www.buzzfeednews.com/​article/​nicolenguyen/​amazon-
​fake-​review-​problem
19. https://​www.cnet.com/​news/​facebook-​turns-​on-​its-​like-​button/​, https://​www.quora.
com/​ F acebook- ​ c ompany/ ​ W hats- ​ t he- ​ h istory-​ o f-​ t he-​ Awesome-​ B utton-​ t hat-​
eventually-​became-​the-​Like-​button-​on-​Facebook, https://​www.adweek.com/​digital/​
view-​shares/​
20. Morin and Mercier –​Majority Rules -​submitted
 Fake restaurants and bad drugs  95

21. Acerbi and Tehrani (2018)

22. https://​www.theatlantic.com/​magazine/​archive/​2014/​12/​the-​shazam-​effect/​382237/​
23. Salganik et al. (2006)
24. Madison and Schiölde (2017)
25. Salganik and Watts (2008)
26. Salganik and Watts use a linear extrapolation, but, as songs change their position in the
chart, their download rates are also expected to change.
27. https://​www.theshedatdulwich.com/​menu/​
28. https:// ​ w ww.vice.com/​ e n_​ u k/​ a rticle/​ 4 34gqw/ ​ i - ​ m ade- ​ m y- ​ s hed- ​ t he- ​ t op-
​rated-​restaurant-​on-​tripadvisor
29. Luca and Zervas (2016)
30. Mayzlin et al. (2014)
31. de Barra (2017)
32. Mayzlin et al. (2014), Zervas et al. (2015)
33. Data from: http://​jmcauley.ucsd.edu/​data/​amazon/​
34. Hu et al. (2009)
35. Chua and Banerjee (2013), Niemelä et al. (2017), Bridges and Vásquez (2018).
36. See discussion in Morin (2016).
 5
Echo chambers

The biggest threat to democracy

In 2011, an opinion article of the New York Times on the Arab Spring—​the wave
of civil unrest that had just hit several countries in North Africa and the Middle
East—​included the following lyrical description:

NATO planes streaked a sky so blue it seemed it to speak of other Septembers.

Faces, too, were lit by a new blue light. It was the era of the revolution down
through the wires: time was collapsed and geography shrunk by the use of social
networking. The whole world was indeed watching, listening, e-​mailing.1

In the public consciousness, the Arab Spring was the social media revolution.
Mobile phones, Facebook, and Twitter seemed more important than everything
else to bring democracy and freedom. Only five years later, however, the World
Economic Forum warned us that our social media feed could be “the biggest threat
to democracy.”2 What did happen in the meantime?
One of the usual suspects is again related to the hyper-​availability provided
by digital media. We prefer to interact with like-​minded others and, if we can
choose between a virtually unlimited number of potential others, we will end
up grouping with people of similar interests, as well as with similar political and
ethical views. In such a situation, our pre-​existent opinions are shouted back at
us—​the “echo” in echo chambers—​and, of course, we do not have access to con-
trasting opinions and critical voices. If you are part of the Facebook groups “meat
is bad” and “meat is murder” (I am making them up, but I am sure something
like that exist) you will have information about the disastrous ecological im-
pact of mass meat production, but probably less about the possibility of making
it more sustainable with reduced consumption and local production. You will
have information about the serious health risks associated with a meat-​based,
especially processed meat, diet, but probably less about the possible benefits of
animal proteins.
Lacking access to contrary information is however only one, and perhaps not
the most worrying, side of the problem. In a classic account, Cass Sunstein ex-
plored how groups of like-​minded people, after deliberating, do not converge on
an opinion that represents the average opinion of the members. Instead, they con-
verge on an opinion that represents the extreme. In an experiment, students were

Cultural Evolution in the Digital Age, Alberto Acerbi. Oxford University Press (2020) © Oxford
University Press.
DOI: 10.1093/oso/9780198835943.001.0001
 The biggest threat to democracy  97

asked to judge guilt or innocence in simulated legal cases. Some groups of students
were presented with pieces of evidence pointing to the innocence of the defendant,
while other groups were presented with pieces of evidences pointing to his guilt.
Students were asked their opinion before discussing with the other members of
their mock-​jury, but after having seen the evidence. Students’ opinions were, in
general, more moderate before the group deliberation, in both directions:  after
discussing, students who had been presented with strong evidence against the de-
fendant became more inclined to think that the defendant was guilty, and students
who had been presented with weak evidence of guilt became more convinced that
the defendant was innocent. In sum, group discussion made the initial individual
decisions more extreme.3
So-​called group polarization seems to be a pervasive phenomenon in real-​life
settings. White people who already show racial prejudice are more convinced,
after discussing, that racism is not a cause of the more difficult conditions fa-
cing African-​Americans in the US. Conversely, people who did not show racial
prejudice become more convinced that racism is part of the problem. Moderate
pro-​feminist women become more extreme pro-​feminists after discussion; inter-
viewers with mild preferences for one applicant prefer the same applicant even
more after group discussion.
What are the causes of group polarization? Sunstein points to two culprits: so-
cial comparison and persuasive arguments. First, each of us wants to be perceived
favorably, especially within our group. A good way to obtain this is to have opin-
ions that are in line with the majority, but slightly more extreme. If you want to
be well accepted, or perhaps you are trying to be the leader, in a group in which
everybody agrees that intensive animal farming is generally bad, it may be a good
bet to declare that the impact of animal farming is “one of the biggest threats to
the environment.” Second, the arguments discussed in a group already inclined
to support a certain opinion will all tend toward the same direction, and they will
reinforce each other. One person can disapprove of animal farming especially be-
cause of the health risks to human consumers, another because of animal suffering,
and another one for the deleterious impact on the environment. When they discuss
together, it is likely that each one will provide new arguments to reinforce the same
opinion, and push it toward a more extreme direction.4
Is this what happens online? In this chapter, we will see how the cognitive pre-
dispositions that underpin echo chambers are what we can expect in a social spe-
cies like us, which heavily relies on communication and cultural transmission, and
we will see how these mechanisms have an evolutionary rationale. As we saw for
prestige and popularity, these mechanisms are also adaptive on average, and could
perform suboptimally when their ideal conditions of functioning are not met. We
will review some of the growing literature on online echo chambers and polariza-
tion, suggesting that these phenomena may have been overestimated as we tend to
be relatively flexible to social influence. This should not be a surprise now. While
98  Echo chambers

we do prefer people who agree with us and to have our own thoughts confirmed,
we are not blind to contrary evidences and opinions, offline and online.

Self-​similarity

In the previous chapters we have reviewed some of the strategies that cultural evo-
lutionists propose are used to optimize the outcomes of social learning, such as
paying preferential attention to parents, successful persons, or prestigious ones.
Another strategy has been dubbed self-​similarity bias. As when preferentially
copying from prestigious individuals, copying from people who are similar to our-
selves is a reasonable way to sift through the many possible models we can learn
from. When we have the option, we can choose to copy individuals of our own
sex, age, and ethnicity. The reasons should be quite intuitive. The more a person is
similar to us, the more likely it is that they live in similar situations, and that they
possibly face similar challenges.
Joe Henrich uses gender as an example. In many traditional societies (and non-
traditional too), men and women are expected to master different skills. Until a few
years ago, for example, cooking in Italy was almost exclusively a women-​related
activity. Men who were able to cook were generally doing it as a profession, or they
were an exception to the norm. Observing same-​sex adults models would have pro-
vided to young girls information on an activity that they were supposed to master,
while a toddler boy observing women cooking would have learned something that
would probably not be useful in his future.5
The same logic, in an evolutionary framework, can be applied to ethnic
groups: to use the same example, why should one learn cooking from people who
may hunt different animals, or grow different crops and vegetables? The skills ac-
quired will not, on average, be useful. In addition, cultural skills such as cooking
rely on complex systems of affinities between ingredients, preparation procedures,
or processing techniques that require all to be integrated, so that isolated observa-
tions may not be sufficient to acquire the necessary skills. Imagine, for instance,
observing a sushi chef only once or twice and then trying to prepare sushi by your-
self, starting from the bare ingredients. Choosing individuals similar to us, day
after day, allows us to continue to copy the same cultural traits, thus scaffolding our
knowledge when the required skills are complicated to acquire.
A subtler, but even more important, reason is that persons similar to us, in par-
ticular of the same ethnic group, share the same norm systems. Even if your neigh-
boring ethnic group happens to cook the same food, and with similar procedures,
there will be social rules that govern their communicative and cultural interactions.
If one violates these rules, which often are arbitrary-​looking and opaque, social
transmission of information and communication in general can fail. The sushi chef
of your neighboring group has skills that you could use, but you do not know the
 Myside bias  99

rules that govern the interactions necessary to learn them. Can you look directly at
her while she is preparing the food? Are you allowed to ask questions?
As we have already seen, as much as all the other social learning strategies pro-
posed by cultural evolutionists, self-​similarity is also a rule-​of-​thumb and we may
expect many occasions in which a preference to copy individuals similar to us is
not the best thing to do or is not what we do in practice. By and large, one can often
assess whether or not a skill shown by a person of a different sex, or age, or ethnic
group, can be useful: if a male toddler does not learn to cook from adult females,
he can certainly learn words, or plenty of other cultural traits. Prestigious people
are often different from us—​which is why we may want to copy them—​and one
of the problems of the prestige-​based learning strategy we discussed in the pre-
vious chapter was that we may indeed copy individuals who are too different from
us and end up acquiring useless skills. While the normative aspect is trickier, one
can guess at least some of the norms that regulate communications in different
cultures, using broadly common principles (be respectful with the sushi chef) or
using, in turn, other cultural cues (are other people talking with her while she is
working?).

Myside bias

Another cognitive predisposition that could make us particularly prone to become

stuck in informational echo chambers has been recently explored by cognitive sci-
entists Hugo Mercier and Dan Sperber under the name of “myside bias.”
Psychologists generally refer to the tendency of being more inclined to believe,
or search for, opinions and information that validate what we already believe as
“confirmation bias.” The existence of a confirmation bias is one of the most robust
findings in psychology, even if the interpretations of what causes the subtle dif-
ferences in behavior found in the experiments are various. Imagine you are pre-
sented with a triplet of numbers, say 2–​4–​6, and you are asked to figure out the
rule that was used to generate it. Done? Now you can test your hypothesis by sug-
gesting additional triplets and checking whether they are consistent with the rule.
Which triplets do you want to test? We tend to test triplets that already confirm
our hypothesis. If, for example, we hypothesized that the rule was “a series of con-
tiguous even numbers,” we will suggest checking triplets like 10–​12–​14 or 20–​22–​
24, which will indeed confirm our hypothesis. This, however, is not useful: what we
would need to do is to suggest triplets that can disconfirm it, and still be consistent
with the first triplet we were presented with, such as 3–​5–​7 (a series of number
increasing by 2), or 12–​45–​777 (a series of number increasing in size), 46–​22–​8
(any three even numbers), and so on.6
Mercier and Sperber, however, prefer another interpretation. The strategy
adopted in the triplet experiment is a reasonable one: we use an intuitive “positive
100  Echo chambers

test strategy” that works in many real-​life situations, and it happens to fail only be-
cause of the artificiality of the set-​up (which, of course, is designed on purpose to
trick us). They are more interested in cases where we are explicitly looking for ar-
guments and evidence to confirm our beliefs, for example, when they are disputed
by others. What if someone asks us why we tested the triplet 10–​12–​14 and we need
to argue whether this is a good choice or not? In this case, we are more prone to
look for information that we can use as arguments and justifications for our choice,
and to reject contrary arguments. Mercier and Sperber notice that the definition
“confirmation bias” can be misleading: we are not prone to believe and look for
confirmatory arguments as opposed to disconfirmatory ones. We are very ready to
produce counterarguments, as long as they are counterarguments to beliefs with
which we already disagree. Understood as such, a myside bias is not a bug of our
cognitive system: it is a feature. It is useful to argue with others—​and to convince
them. If we were to evaluate critically all our beliefs and, especially if we were to
provide others with easy ways to evaluate them, we would have a hard time in our
discussions.7
The downside of this phenomenon is that, in an information-​rich world, we end
up selecting the information and the viewpoints that reinforce what we already
believe and discarding the others. It is an easy exercise: you can google “why meat
is good for you” and “why meat is bad for you” and see for yourself. If you believe
that people living in Scandinavian countries are happier than people living in other
countries, you can refer to the World Happiness Report, and cite several studies
showing how generous social support generates happiness and life satisfaction. If
you believe the opposite, just google “seasonal affective disorder,” or cite the several
studies showing how unreliable self-​reports of happiness are, and how transcul-
tural variations make the same concept of “happiness” very difficult to compare in
different countries.
The tendencies to learn from people similar to us and to search for, and prefer-
entially accept, arguments that confirm our pre-​existent positions seem to conjure
up ways to make our digital interactions the perfect ground for echo chambers
and polarization. Given thousands of news outlets available, I can choose to read
only the ones with which I agree. My social media contacts are filtered by my own
choices, so that, when I disagree with someone or I am just not interested in their
posts, getting rid of them is a matter of a click. Availability goes together with opa-
city: there is no need to tell our contacts that we do not want to interact any more
with them. Facebook is careful to notify you, after unfollowing a friend, that they
will not know it. On the other hand, these cognitive predispositions do not make
us act in autopilot mode. Self-​similarity is a cue, among many others, useful some-
times to select from whom to copy, but can be easily overridden. The myside bias
makes us partisan when finding arguments that support our own beliefs and over-​
critical when evaluating views that challenge ours, but we are sensitive to good ar-
guments, and we do change our minds, especially when discussing in groups.8
 Science and conspiracy theories on Facebook  101

Science and conspiracy theories on Facebook

If you look carefully at pictures of NASA photos documenting successive moon

landings, you will notice a striking similarity: astronauts, landing modules, rocky
landscapes, and the occasional US flag stand in front of dark, starless, skies. Where
have all the stars gone? In the dark, cloudless, moon sky, outside of earth’s atmos-
phere, stars should be clearly visible. In fact, one should see them better than from
the earth. Or shouldn’t they? The absence of stars in the NASA pictures is often
presented as one of the evidences showing that the moon landings never happened
(it strikes me that, while orchestrating such a complicated ploy, the moon landing
inventors forgot a silly, and easily to put right, detail).
The broad success of conspiracy theories, such as the claim that the moon land-
ings were, in reality, staged hoaxes, is often associated with the diffusion of social
media. In the next chapter we will examine in detail the claim that misinforma-
tion is a digital-​age phenomenon—​even though I  guess you may, at this point,
imagine what my take will be: just consider that the book We Never Went to the
Moon: America’s Thirty Billion Dollar Swindle was published in 1976, exactly thirty
years before Facebook’s public opening.
In any case, conspiracy theories, and the people believing them, have been clev-
erly used to detect polarization and echo chambers in social media like YouTube
and Facebook in a series of studies conducted by Walter Quattrociocchi and col-
leagues. One can go through Facebook, and manually isolate pages that refer to
conspiracy theories, such as CancerTruth, active with almost 300,000 followers
when I write (its main claim seems to be that alternative and more efficacious treat-
ments for cancer exist, but “Big Pharma” does not want us to know about them),
and scientific sources, such as News from Science, the official Facebook page of the
Science magazine.9
Quattrociocchi and colleagues registered the posts appearing in these two
groups of pages—​the “conspiracy” and the “science” pages—​and the users that
interacted with them, with comments, likes, and shares. They defined users as po-
larized when, considering their likes in either science or conspiracy pages, more
than 95 percent of them was falling under the same group. With this system, they
found more than 200,000 users who were polarized on the scientific pages, and
more than 700,000 polarized on conspiracy pages. Similarly, they showed that only
1 percent of the comments on science pages were coming from users categorized
as polarized on conspiracy, and around 10 percent of the comments on conspiracy
pages were coming from the science fans.
One can also check how the information related to science or conspiracy the-
ories actually spread in social media. Does news about a natural “treatment” for
cancer, ostracized by Big Pharma, reach the people mainly interested in science? Is
the last editorial of Science read—​or at least commented on, perhaps negatively—​
by the conspiracy theories crusaders? Quattrociocchi and colleagues calculated
102  Echo chambers

the “edge homogeneity” of the “information cascades” of science and conspiracy

posts: leaving aside the jargon, the concept is quite simple. When considering an
information transmitted from individual A to individual B, the edge homogen-
eity is positive whenever the two individuals are categorized as polarized in the
same domain, and negative if they are not. With this in mind, the homogeneity
of a cascade—​the full history of a diffusion of a post—​will be close to 1 when the
post has been mainly transmitted among individuals categorized as polarized in
the same domain, and close to −1 when this has not happened. Both science-​and
conspiracy-​related cascades ended up to be close to 1, meaning that, not too sur-
prisingly, very few posts were transmitted from one community to the others.
Finally, another analysis revealed a peculiar mood signature of the communi-
ties. Comments were categorized as emotionally positive, negative, or neutral, with
the help of a supervised learning algorithm. The longer the discussion on a par-
ticular post, the more negative the content of the comments (a possible empirical
confirmation of Godwin’s law for internet discussions: “As an online discussion
grows longer, the probability of a comparison involving Hitler approaches one.”).
In addition, the more the engagement of a user, measured as the total number of
her comments, the higher the probability that the comments will be negative. Both
findings were interpreted as supporting the idea that users become more polarized
and extreme in their views.10
Studies like these are an example of what newly available digital data can do for
social science, and they represent what ultimately makes this book possible. The
results need however to be interpreted with a grain of salt. An opinion piece in
the Washington Post commenting on the Facebook study reported above was ti-
tled “Confirmed: Echo chambers exist on social media” and, in general, the results
were almost unanimously interpreted as a confirmation that social media produce
polarization, if not, as we mentioned above, that social media are a threat to dem-
ocracy. Are these conclusions justified?11
A first reason to be cautious concerns the selection of the data to analyze. If one
chooses deliberately two communities that are obviously opposed to one another
in many respects—​politics, interests, religion, education, etc.—​it is not very sur-
prising that there will not be much dialogue between the individuals belonging to
them. One can choose supporters of science and conspiracy theories, but it would
have worked the same with extremist far-​left and far-​right political groups, or, as
in the example I have made above, with the groups “meat is murder” and “Texas
BBQ Posse” (I swear the former these exists). Quattrociocchi and coauthors note
the problem, and mention, quite en passant, that “other kinds of data sets may not
show the particular patterns we observe here.” Their results could also be inter-
preted more reassuringly as demonstrating that data from social media confirms
the not-​too-​surprising fact that people with very different opinions do not tend
to hang out together, more than social media and polarization are linked in any
meaningful way.
 Echo chambers and deliberating enclaves  103

Related to this, to identify polarized individuals, as explained a few paragraphs

above, the rule was to consider, for each individual and among all the likes either in
the “conspiracy” or the “science” pages, if the percentage for one group was higher
than 95 percent. Is this a “stringent test” as affirmed in the study? Imagine I like
mainly pictures of my cousins’ kids and mum’s roasts, plus the occasional meme
and local news, but I also liked once an article on aromatherapy or on the benefit
of yoga that happened to be in one of the pages flagged in the “conspiracy” group.
From my profile, I would count as polarized, with a 100 percent likes for conspiracy
and 0 percent likes for science pages. While the methodology is formally correct,
I doubt that brings out what the majority of us imagine when thinking about polar-
ization and echo chambers.
In addition, it is not clear why the fact that comments become more negative
in prolonged discussions should be put in relation to polarization. If anything, in
an echo chamber everybody should agree with each other, and one would expect
mainly positive or neutral comments. Contrary to what many commentators seem
to think, there is not an obvious way in which the two sentences “there is too much
hatred and aggression in social media” and “social media create echo chambers
where everybody has the same opinion” are contemporarily true.

Echo chambers and deliberating enclaves

Another aspect that has broader relevance for the concept of echo chamber is that
not only interesting discussions may happen among like-​minded people but, some-
times, discussions need to happen among like-​minded people to go forward. Think
about science, one of the two domains in which echo chambers were identified in
the study of Quattrociocchi and colleagues. In the popular image, scientists and
academics are specialized to the point of not being able to talk with each other out-
side of their small community, and this is not too far from true. Interdisciplinary
studies are often tricky to carry on exactly for this reason. Besides the widespread
advice of cultivating multiple interests and remaining curious outside one’s own
specialization, the truth is that everyday science is made of small advancements,
obtained and discussed in relatively isolated communities. Evolutionary anthro-
pologists do not usually take the time to discuss creationists’ arguments at their
conferences, and in fact they assume that all other attendants will share a vast
amount of opinions and knowledge, and they go straight to matters of minimal—​
from an external point of view—​disagreement.
The specialized communities we described in the first chapter of the book—​
remember historical costuming and esoteric programming languages—​are all ex-
amples of relatively closed groups, where information is likely to spread mainly
within boundaries, exchanges with other groups are limited, and so on. Still, these
groups represent the possibility, often the only possibility, for people to interact
104  Echo chambers

with others that share the same interests, and to keep certain cultural traditions
alive. How does this fit with the echo chamber narrative?
Prominent proponents of the digital polarization danger, such as Sunstein, have
of course recognized this possible ambivalence. Sunstein talks about “deliberating
enclaves,” allowing for a “form of deliberation that occurs within more or less in-
sulated groups, in which like-​minded people speak mostly to one another.” Not
only do these enclaves make it possible for people to interact with others with the
same interests, but again according to Sunstein, they crucially allow the voices of
minorities to be heard. If, say, proponents of same-​sex marriage could not have
expressed their once (at least in the majority of western countries) radical ideas in
semi-​isolated, like-​minded-​people-​filled, communities, these ideas would possibly
not have emerged at all.
Sunstein continues:

But there is also a serious danger in such enclaves. The danger is that members
will move to positions that lack merit but are predictable consequences of enclave
deliberation [ . . . ] It is impossible to say in the abstract, that those who sort them-
selves into enclaves will generally move in a direction that is desirable for society
at large or even its own members.

Predictable mentions of Nazis, terrorism, and conspiracy theories (“Barack Obama

was not born in the United States”) follow in the next paragraph. What then does
exactly distinguish echo chambers and deliberating enclaves? Unfortunately it
is difficult to pin the answer down. A reasonable conjecture would be that like-​
minded groups that do not drift towards extreme opinions count as deliberating
enclaves. This, however, does not seem to be the case:  “group polarization has
helped fuel many movements of great value—​including, for example, the civil
rights movement, the movement for equality between men and women, and the
movement for same-​sex marriage. All these movements were extreme in their
time, but extremism need not to be a word of opprobrium.” In sum, it looks as
though some groups of polarized, like-​minded, people function as deliberating en-
claves, while others function as echo-​chambers, depending on how we evaluate
their outcomes.12
Coming back to our main concern: what is the role of social media in all this?
We saw that, giving the hyper-​access that digital tools provide, there is the pro-
spect that we will choose to interact only, or mainly, with other individuals with
whom we already share the same opinions and interests. While this may not be
necessarily bad—​as the case of scientific activity and the possibility of constructive
“deliberating enclaves” suggest—​it may hinder contrary opinions, and favor polar-
ization and extremism. Intuitively, social media facilitate this tendency, even when
we do not consider extreme cases such as groups of people interested in science
and groups of people interested in conspiracy theories. Other studies made clear
 Echo chambers online and offline  105

that, even without assuming a pre-​existent division among groups, political dis-
cussions tend to generate clusters of individuals who scarcely communicate with
one another. Twitter data collected before the 2010 US mid-​term elections showed
that retweets exhibited a “highly segregated partisan structure,” that is, people who
retweet politicians, or political-​oriented material, tend to retweet only content
that reflects their views. Interestingly, the same study also found that segregated
clusters couldn’t be isolated when considering mentions, that is tagging someone
else’s username in a tweet. In this case, a single, politically heterogeneous cluster
emerges, composed by the whole twitter-​sphere interested in politics. In other
words, Republicans retweet Republicans, but also mention Democrats, in all like-
lihood to criticize them. When calculated using retweeting behavior, polarization
was still present, and indeed possibly higher, before the 2016 general election in
the US.13
Again, this is not that surprising. The critical issue is whether digital inter-
actions, mainly through social media, cause echo chambers and polarization or at
least, make them more prominent than what happens with interactions in real life,
or with old media, such as press, radio, and television.

Echo chambers online and offline

Let’s assume that our online interactions are often limited to like-​minded others,
a series of repetitions of similar opinions that reinforce our pre-​existent attitudes.
What about offline life? Are workplace discussions and pub conversations full of
unexpected point of views? How many times do family meetings between the con-
servative parents and the liberal kids after a festive lunch make either change their
opinions? Are newspapers and TV channels, now and before the internet, a para-
digm of plurality? Again according to Sunstein, this should roughly be the case:

The diverse people who walk the streets and use the parks are likely to hear
speakers’ arguments about taxes or the police; they might also learn about the na-
ture and intensity of views held by their fellow citizens . . . When you go to work or
visit a park . . . it is possible that you will have a range of unexpected encounters.14

In my personal experience, online diversity seems to be higher than offline diver-

sity. My Twitter feed is politically schizophrenic: I happen to follow evolutionary-​
minded types that oppose the post-​modern social sciences and tend to gravitate
politically on the right/​conservative side, as well as the more customarily liberal
academics and friends. I imagine a dinner with a few people taken from the two
extremes would unlikely to be a success. I do not ask you to base your opinion on
my anecdote however, as a good number of recent researches is exploring precisely
this topic.
106  Echo chambers

One of the first works to compare explicitly polarization online and offline, which
appeared in 2011, concluded that “ideological segregation of online news consump-
tion is low in absolute terms, higher than the segregation of most offline news con-
sumption, and significantly lower than the segregation of face-​to-​face interactions
with neighbors, co-​workers, or family members.” Gentzkow and Shapiro, the two
authors of the study, quantified ideological segregation using an “isolation index,”
that gives a score to each of the domains considered:  internet, television news,
newspapers, actual networks of families, co-​workers, etc. For example, in the case
of websites, Gentzkow and Shapiro pooled together data from surveys on political
attitudes with traffic data from more than 100 political and news sites. They first
calculated the share of conservatives and liberals visiting that website, giving a score
to each of them: foxnews.com will have a high conservative share, while nytimes.
com will have a low one. Then, they calculated the exposure of each individual,
which is the average share of the websites they visit. If one individual takes all their
news from nytimes.com, their score will be the score of nytimes.com; if one reads
news from nytimes.com and foxnews.com, their score will be the average of the two.
Finally, they used these measures to calculate the isolation index, which tells us, ba-
sically, how much conservative exposure a liberal has, and how much liberal ex-
posure a conservative has: “If conservatives only visit foxnews.com and liberals only
visit nytimes.com, the isolation index will be equal to 100 percentage points. If both
conservatives and liberals get all their news from cnn.com, the two groups will have
the same conservative exposure, and the isolation index will be equal to 0.”
The isolation index they calculated for the internet was equal to 7.5 percentage
points. To understand this result it is necessary to compare it with the indexes cal-
culated in other domains: cable television news, magazines, and local newspapers
all had a lower index—​meaning they were less polarized—​but national news-
papers had an index of 10.4 percentage points, higher than internet websites. The
results for face-​to-​face interactions are possibly more surprising. They were all
higher: real-​life networks such as work colleagues (16.8), neighborhoods (18.7),
family (24.3), trusted friends (30.3) and political discussants (39.4) resulted in
being clearly more polarized than internet websites.
Gentzkow and Shapiro propose an explanation for these results based on two
facts we will encounter again when discussing studies of polarization and misin-
formation in the digital age. First, non-​extremist websites have a remarkably larger
traffic than extremist, polarized websites. For every click on breitbart.com, there
are several clicks on cnn.com or the like. Focusing on the extreme positions—​such
as in the research on conspiracy and science on Facebook—​may give the impres-
sion of a polarized world, while it is likely the majority of people are fairly mod-
erate. Second, a big share of individuals that read their news on websites with a
polarized political position tend to be interested in politics in general. They are, in
fact, more likely to also read non-​polarized news websites, or even websites from
the opposite side of the political spectrum.15
 Polarization in social media  107

Of course, there are a few caveats with this study. The estimates of individual
political affiliations were taken from surveys, and they were squeezed to fit a binary
liberal/​conservative position. We know results from surveys need to be interpreted
with caution. Some groups of individuals are more likely to answer than others;
individuals may lie about their political position, and it is not for everybody ob-
vious to place oneself in a certain position in a scale “very conservative/​somewhat
conservative/​middle of the road/​somewhat liberal/​very liberal.” Also, to calculate
the isolation index of face-​to-​face interactions, which show the interesting results
of being higher than online data, Gentzkow and Shapiro had to rely on assessments
of the political orientation of people’s acquaintances, which may be also problem-
atic: respondents of the survey were asked the number of, for example, co-​workers
they were “pretty certain are strongly liberal” and “pretty certain are strongly
conservative.”
Another aspect to consider is that the study, as mentioned above, appeared in
2011, and the data regarding online interactions are from 2009. This is an eternity
for the fast changing pace of the digital world: social media are not even considered
in the study. One could legitimately say that the research documents a previous
age of internet, and that polarization and echo chambers are a problem now.
Notwithstanding these limitations, the article of Gentzkow and Shapiro shows that
we should, at least, not take for granted that the internet in itself causes polariza-
tion. Interestingly, however, more recent analyses tell a similar story.

Polarization in social media

Data from more than 10  million Facebook users, collected in 2014, show that,
among users who self-​report their ideological affiliation (liberal or conservative),
around 20 percent of their friends declare the opposite ideology. While not sur-
prisingly, individuals of the same political ideology cluster together, this picture
is far from a system of segregated echo chambers. One out of every five friends of
a conservative is liberal, and one out every five friends of a liberal is conservative.
Thinking about it, Facebook friends are often a heterogeneous mix of familiars, co-​
workers, old contacts from school, people sharing your hobbies, etc. and it would
be quite surprising if they all have the same political ideology. But this is exactly the
claim of the advocates of the echo chambers hypothesis.
Another question, however, is whether this relative diversity begets an actual
exposure to content of the opposite ideology. The authors of the study considered
seven million distinct web links shared by these users, and they estimated that
13 percent were concerning political matters. They found that around 30 percent
of the political content shared by one’s friends is of the opposite ideology. Here
the exact percentages differ for liberals and conservatives: the content shared by
friends of liberals is of conservative orientation 24  percent of times, while the
108  Echo chambers

content shared by friends of conservatives is liberal 35 percent of times. This means

that conservatives could potentially see opposite content more than liberals, and,
in fact, more than one third of their available content was coming from the other
end of the political spectrum.
The story does not finish here: while these figures tell us what friends share, what
do we actually see in the news feed? After passing through the algorithmic filter,
diversity was indeed reduced, but not that much: “after ranking, there is on average
slightly less cross-​cutting content: conservatives see approximately 5 percent less
cross-​cutting content compared to what friends share, while liberals see about
8 percent less ideologically diverse content.”
Diversity goes again down when considering what people actually click: even
if exposed to cross-​cutting content, individuals prefer, up to a certain limit, to
click on posts coming from the same political side as their own. Both liberals and
conservatives end up clicking on 13–​14 percent cross-​cutting content, or around
one post out of eight. As conservatives were exposed to more liberal content, this
means they also click less than liberals, in proportion, on this content. Not only are
these figures relatively high but, as the authors of the study comment, they show
that what mainly limits the fruition of content from different ideology is not ex-
posure, but individual choices.16
Let’s move to Twitter. We mentioned that data from the 2010 US mid-​term
elections and until the recent 2016 election pointed to a fairly segregated retweet
network, where conservatives tend to retweet conservatives and liberals tend to re-
tweet liberals. Other studies give, however, a more nuanced view. A research from
Pablo Barberá and colleagues is interesting for a number of reasons. First, the two
analyses we discussed in detail so far—​the Gentzkow and Shapiro study and the
Facebook one—​both rely on self-​reported political identifications, in surveys and
in the personal Facebook profile, respectively. Barberá and colleagues tried to find
an objective measure of political affiliation. They estimated what they call “ideo-
logical latent space” by checking who a Twitter user is following. If one follows con-
servative politics, the Twitter account of Fox News, and so on, they will be placed
on the conservative side of this space, and vice versa. Notice this procedure can
be applied iteratively: once many users are placed in the ideological space, even if
I do not follow any major political figure or politically-​oriented news source, my
ideology can be estimated using the estimates made for the people I follow.
The researchers were able to perform this analysis for almost 4 million Twitter
users, and they measured their retweet activity for 12 “interesting” topics, six of
them considered politics (the 2012 elections, marriage equality, etc.), and six of
them explicitly not (the 2014 Super Bowl, the 2014 Winter Olympics, but also
the Newtown shooting and the Boston Marathon bombing—​I will be back on
this soon). They found that, for political topics, retweets were indeed happening
mainly, even though far from exclusively, among users with similar estimated pol-
itical orientation, confirming with a different methodology the results on the 2010
 Polarization in social media  109

elections. However, for nonpolitical topics they found this was not the case, com-
menting that “ideological homophily in the propagation of content related to non-
political events is low; in this sense, discussions of current events do not strictly
conform to the image of an echo chamber.” If we add, as quickly noted before, that
nonpolitical events included the 2012 Newtown shooting, the Boston Marathon
bombing and even the Syrian civil war, one can push the conclusion a little further
and consider that echo chambers were not found for some topics that were indeed
political.
By assigning a cumulative measure of polarization to each topic, Barberá and col-
leagues found, more than a dichotomy between political and non-​political topics,
rather a continuum of polarization. At the two extremes there were, quite predict-
ably, the 2014 Oscars and the 2012 Election, but, for example, the Boston Marathon
bombing was among the less polarized topics. Finally, they found that polarization
is a highly dynamic process. Some topics, such as the Newton shooting—​one of the
deadliest school mass shootings, where 26 individuals, plus the perpetrator and his
mother lost their lives, went from not being polarized to being polarized, which
is roughly what would be predicted by the group polarization hypothesis. Others,
however, such as the Syrian civil war, showed the opposite trend, losing their polit-
ically polarized structure of retweet as the events unfolded.17
A different approach to assess the effect of social media on polarization is to
compare how the material an account receives contrasts with the material an ac-
count posts. As we discussed above, the idea of echo chambers implies that in-
dividuals are not exposed to contrary opinions. On top of this, being in an echo
chamber would lead to group polarization, making individuals more extreme in
their views. If this is what is going on, one would predict that the content posted by
an individual should be, on average, more extreme than the content they receive.
Apparently, this is not what happens: the “political slant” of the websites linked
from Twitter is, on average, more moderate than the links one is exposed in the
timeline.
There is a fascinating addendum to this result: one can go beyond the average
behavior and characterize the Twitter social network in terms of core/​periphery
structure. The core is the set of highly followed users, that may follow each other,
but do not follow the “periphery” of users. The latter, in turn, follow the accounts
of the core, but are less likely to follow each other. A slightly different, but com-
parable, way to consider it is the ratio following/​followers. If you have a low ratio,
it means that you follow few people but you have many followers—​you are in the
core—​while if you have a higher one it is the opposite. Alternatively, another way to
go beyond the average behavior is to consider the number of links to news that one
account posts: few accounts are very active and post an abundant number of links
to news, whereas others post only few, if any, of them.
When restricting the analysis to accounts that are in the core and that post a
lot of news, they do tend to post material that is more extreme than what they
110  Echo chambers

receive, and the stricter the criteria to select the core and the “news-​centricity”
of the accounts, the higher the polarization. On the contrary, the majority of
accounts, that are on the periphery and that post fewer links to the news, are
not tweeting links to extreme sources. Have you ever heard about the friend-
ship paradox? Your friends are likely to have, on average, more friends than you.
This happens because the number of friends one has—​like many other quan-
tities, as we saw in the Popularity chapter—​is distributed in a peculiar way, with
few persons that have many friends, and the majority of us that have only a few.
Since each of us is likely to be a friend with one of the super-​popular (exactly
because they are super-​popular), the average number of friends of our friends
is increased. This is the same idea we discussed for the average number of views
of Bach compositions of YouTube, skewed by the Orchestral Suite No. 3 and the
“Toccata and Fugue,” or average visitors of websites, skewed by Google and the
like. Here something similar happens: highly followed and very active accounts
could, according to the authors of the study, give the misleading impression of
polarization on Twitter, while, in fact, the great majority of accounts is not or is
scarcely polarized.18

Old and new media

What about the good old analogic media, like newspapers and television? As just
discussed, a few years ago their audience seemed, in general, less polarized than
the audience of news outlets on the internet, with the exception of US national
newspapers. What is the situation now? A recent research with data from 2016 pro-
posed, similarly, that offline audiences are more polarized than online audiences.
The researchers used surveys in six countries—​in addition to the US, they con-
sidered the UK, Spain, Denmark, Germany, and France—​and they built networks
in which the nodes are all the possible media outlets. Each pair of outlets is con-
nected if their audience is “duplicated.” The audience of two media outlets, say the
New York Times and the Washington Post, is duplicated if it is overlapping above a
certain threshold: if the number of participants in the survey who said they read
both the New York Times and the Washington Post is higher than what we would
expect by chance, the two will be connected in this media network. At this point,
one can measure fairly standard properties of the network, such as density (the
proportion of possible links between nodes that are actually present) or diameter
(the shortest path between the farthest links in the network) to describe the media
landscape in a country.
Density, in particular, gives an indication of the polarization of the media: the
less links that are present, the less people consume news from different sources.
Differently from the prevailing expectation, the network density of offline media
was lower than the density of online media, for all six countries considered in the
 Old and new media  111

analysis. Individuals are more likely to gather information from different news
websites than from different newspapers or television channels.19
This result, like many of the others we considered above, points to the coun-
terintuitive possibility that not only digital media do not create polarization but
that, in fact, they could even reduce it. More diversity could lead to more exposure,
and more exposure does not necessarily lead to echo chambers. Sure, polarization
did seem to increase, and the use of digital, especially social, media also increased,
but the two do not need to be related. Digital media, and their social effects, have
been repeatedly indicated as one, if not the main, cause of the fragmented and
conflictual political situation. Many commentators have considered the election
of Donald Trump, or the results of the 2016 Brexit referendum, as an outcome of
our over-​reliance on digital platforms. One hunch that this may not necessarily be
the case comes from demography. Both in the Trump election and in the Brexit
referendum, older people tended to be pro-​Trump, or pro-​leaving the EU. This is
exactly the opposite of what we would expect if pro-​Trump and pro-​leavers were
mainly influenced by social media, as older people use less social media than
younger ones.
This hunch was properly tested and extended more generally to polarization
tendencies. Levi Boxell and colleagues analyzed eight measures of political polar-
ization, taken from surveys covering the period from 1972 to now. These measures
take into consideration aspects such as “Partisan affect polarization,” that is, the
difference between how much a person views favorably other individuals with the
same political orientation versus individuals with the opposite political orienta-
tion, or “Straight-​ticket voting,” that is, the frequency with which individuals vote
“straight-​ticket” (meaning they vote for all candidates proposed by one party, a sign
of strong political identification) at successive elections. A composite index of po-
larization showed indeed an increase in the last four decades. The researchers also
considered data concerning internet and social media usage, from 1996, and they
then divided the population into three age groups. As for the hunch, the older the
people the bigger the growth in polarization and, not surprisingly, the less the esti-
mated usage of internet and social media. People most likely to use the internet—​
in the age group between 18 and 39—​are the ones for which polarization increased
less. The opposite was true for people aged 65 or older. Putting together the data in
a statistical model, the authors of the paper concluded that, out of four measures
of internet and social media usage, representing different surveys, the effect on the
increase of polarization was positive for only one, and it was relatively small, with
internet usage accounting for 6 percent of the increase in polarization through the
years. For the other three measures the effect was instead negative.20
It is not too difficult to find similar suggestions. A  recent survey from the
Cultural Cognition Project at Yale University showed that Millenials, defined there
as individuals born between 1982 and 1999, are less polarized than older groups on
issues such as human-​caused climate change or the importance of evolution for the
112  Echo chambers

human species. This does not simply mean that more young people tend to agree,
for example, on the fact that “there is solid evidence of recent global warming due
mostly to human activity such as burning fossil fuels” which is comforting, but
does not say much about polarization. This means that the difference in opinions
between strongly conservative and strongly liberal, or very religious and not reli-
gious, individuals is smaller than the same difference for older people. This may
not be old news but, again, young people are the ones who use social media more.21

Depolarization

I hope you are starting, at this point, to be at least less sure that digital media are
producing echo chambers and causing polarization. We saw that polarization may
be stronger offline, in media and real-​life networks such as friends and co-​workers,
than online. We saw that, on average, a Facebook user has one out of five friends
from the opposite end of the political spectrum, and they may be exposed to as
much as one third of cross-​cutting content (if they are the average conservatives.
A little bit less, one fourth, if they are the average liberals). We saw that discussions
on Twitter are not polarized for all topics, and that polarization does not neces-
sarily increase with time. We saw that the majority of Twitter accounts post content
that is less extreme than the content to which they have access to, and that the im-
pression of polarization may be the product of few, but very active and highly fol-
lowed, polarized accounts. We saw that polarization may have indeed increased in
the last years, but the groups of people most polarized are the ones who use social
media to a smaller degree.
This brings us to the even more unpopular idea that social media could reduce
polarization. How plausible is this outcome? There are good arguments, as we dis-
cussed at the beginning of the chapter, to believe that when people are surrounded
by like-​minded others—​and not exposed to opposite ideas—​they tend to become
more extreme in their points of view. However, the question is whether the hyper-​
access provided by digital media produces, or at least favors, this ideological seg-
regation. This is more dubious. Could it be that the availability of more sources
produces less informational segregation?
The fact that young people tend to be less polarized than older ones does not
provide any support for this conjecture. It is true that young people use more so-
cial media, but there may be many other reasons why they are less polarized, in the
same way as the temporal correlation between increase in social media usage and
increase in polarization does not prove that the former causes the latter. There are
some indications that individuals who consume a wide variety of media tend to be
less polarized, and that social media users, in particular, use more different sources
of news than non-​social media users.22 Once more, however, this may not be due
 Depolarization  113

to the usage of social media itself, but to the fact that social media users are, on
average, younger, more educated, and more interested in news generally.
A way in which social media could reduce polarization is by allowing weak ties
to be kept, as I mooted could be the situation in my own Twitter feed. Weak ties are
contacts with whom we share some specific interests, whether these are historical
costuming, evolutionary social science, or speleology, old friends we do not meet
anymore, or work colleagues we are not seeing much. These contacts are in fact
more likely than our strong ties, such as the friends we meet daily, close members of
the family, or people at work we are in close contact with, to have different opinions
from us, in various domains. In addition, we are unconcerned about their political
views, as much as they are good speleologists or reliable experts of costumes from
the Victorian period. Weak ties are more politically heterogeneous than strong ties.
Pablo Barberá used the method we described a few paragraphs above—​
estimating the position in the “ideological space” of individuals based on whom
they follow—​to measure their exposure to diversity on Twitter. According to the
estimate, he divided users into two groups, conservatives and liberals, and then
checked their diversity exposure, that is, what was the proportion of individuals in
each user’s network who did not share the same ideology. If homophily was perfect,
we would expect diversity exposure to be equal to zero, while, on the contrary, if
Twitter users followed other people independently of their ideology, or at random,
the diversity exposure should be around 50 percent. The results were closer to the
latter than to the former. Barberá collected data in three countries, and found that
in Germany and Spain, more than 40 percent of individuals in one’s own network
are from a different ideology. This calculation is lower in the third country he con-
sidered, the US, but the proportion of individuals with the opposite ideology is still
33 percent, that is, one third of the accounts one follows on Twitter are from the
opposite political side. This result is in line with what we saw for Facebook, where
the proportion was around one fourth. The fact that the diversity is even bigger on
Twitter than on Facebook is in line with what we considered in the first chapter,
when we discussed how Twitter can be regarded mainly as a network for exchange
of information, while Facebook is a “circle of friends,” or it is only currently transi-
tioning from a “circle of friends” to a network for exchange of information.
So far, nothing about depolarization. Barberá measured it in two ways. The first
is again through surveys. Measuring political extremism before and after elections
in the three countries, the respondents who reported using social media in Spain
and the US become more politically moderate after the campaign in respect to the
respondents who reported they did not, controlling for other variables such as age,
gender, education, usage of other offline news, and political ideology. The result for
Germany was not significant. Notice this means, in any case, that social media did
not make people more polarized even in Germany, while in the US and Spain social
media users became less polarized.
114  Echo chambers

The second measure is more interesting, as it does not involve external, and
possibly biased, estimations such as the ones coming from surveys. Barberá
calculated the position in the ideological space at the beginning of 2013 and
then again in 2014, and evaluated what the effect of the exposure to political
diversity on changes of ideological position. The results are interesting for two
reasons. First, they confirm that individuals who are mainly exposed to only
one side of the ideological spectrum will tend indeed to become more polar-
ized. However, when the diversity exposure had values around 30 percent, as
was the case in Spain and Germany, and around 20 percent, as was the case in
the US, the effect starts to become negative: the average individual becomes
less polarized. Since, as we saw before, the average diversity exposure in the
three countries is higher than these values, one should expected that social
media will indeed foster political moderation by exposing users to a relatively
large array of diverse views.23

This short review provides some reassurance on the role of digital, especially so-
cial, media in creating echo chambers and polarization. Social media activities are
certainly not immune to propensities that exist in the offline world, such as a so-
cial learning strategy based on self-​similarity (the preference to copy from people
who are similar to us) or a myside bias (the tendency to search for evidences that
reinforce our previous opinions). At the same time, these propensities are not all-​
or-​nothing. As we repeatedly saw for other cognitive tendencies that regulate our
social interactions, they tend to be flexible enough not to create systematic, and
too costly, drawbacks. Again, the question is whether the current situation is too
unusual for these tendencies to act properly, but the research analyzing polariza-
tion and echo chambers in social media provide results that are, at best, mixed. In
particular, when we compare our digital interactions to other benchmarks, they
do not seem to be more polarized in respect of other media or, in fact, day-​to-​day
interactions with our friends and co-​workers.
Of course, this assessment is far from definitive. The study of polarization in so-
cial media is a recent undertaking and, as such, there is no clear agreement on what
are the defining properties one should look for. In the few researches we considered
we saw a varieties of methodologies to detect polarization, ranging from surveys to
the indirect detection of the political affiliations of the individuals in one’s own
social network. This is a reflection of the fact that, social media or not, measuring
polarization is not straightforward. Whereas there is some general agreement that
polarization increased in the last years, different measures can give different re-
sults, and weaken or strengthen the supposed link between polarization and the
diffusion of social media.
Some cultural evolution-​inspired analyses, for example, as the one proposed by
Peter Turchin, see recent events in a long-​term perspective, where political polar-
ization has started to increase at least from the 1970s. This is obviously not down
 Depolarization  115

to social media, as they were not even imagined then, but to structural socio-​
economic causes, in particular the increase of inequality. Even more, the same
structural forces where acting in the same way in the years just before the US Civil
War and, according to Turchin, generated dynamics comparable to the ones we can
observe today. We can acknowledge an increase in polarization and, at the same
time, not assume that social media need be the cause of it.24
As a further note of caution, assessing polarization in online activities is as
much, or more, difficult than assessing polarization in general. Polarization, for ex-
ample, can be checked inside a single social media, such as Twitter or Facebook, or
in the overall online activities. Social media change fast, and existing social media
may change their policies to favor, or hinder, the informational and ideological
segregation of users. Mainly, new social media or new forms of digital interactions
could deliver, in a few years, a very different picture from the one described here.
This does not mean that we should just accept the situation as it is. Social media
have many features that make interactions different from offline life and that can
have an effect on polarization. We cannot just get rid of our conservative dad at
Christmas dinner (or, in any case, there are strong disincentives to do that), while
muting a contrary voice on Twitter is a matter of one click. Communications on
social media can be effectively anonymous. The effect of anonymity, however, is
complex, as we saw for the Wikipedia editors earlier and as we will discuss again in
the Misinformation chapter. Not having to reveal one’s own real identity can boost
the troll inside you, but anonymity can also favor the exposition of legitimate ad-
versarial opinions, which one would avoid in a non-​anonymous interaction.
Algorithms—​as we briefly saw for Facebook, and as we will explore again—​can
do the job of excluding contrary opinions for us, without us being even conscious
of it. Even if I never visit it, I am aware of the existence of the gardening section of
my favorite bookstore or, even if I do not check with them, I can suspect some of
my neighbors have different political opinions. We can be tricked into thinking
that our Facebook timeline represents the world like it is, where everybody op-
poses Trump, and plenty of people get excited about the last article on evolution
and culture (that would be my case).
Is this opacity a real risk? If the recent scare about social media could have had
the effect of making people aware of this problem, then there may be some good
in it. There are websites, such as allsides.com, that explicitly “provide[s]‌multiple
angles on the same story” to counteract polarization and echo chambers. A typ-
ical page presents a short summary of a piece of news and three links to exter-
nals sources rated according to their political ideology (“from the left,” “from
the centre,” “from the right”) with no further comments. Newspapers have now
sections such as The Guardian’s “Burst your bubble,” advertised as a “weekly guide
to conservative articles worth reading to expand your thinking” (unnervingly,
however, in this case they are articles about conservative matters, but often critical,
so that the point is missed).25
116  Echo chambers

Of course, if the tendency to strengthen echo chambers online is real, individ-

uals will actively avoid this contrary information, as the study on Facebook we dis-
cussed before suggested (we click on around half of the “cross-​cutting” content we
are shown in the Newsfeed). But, as we saw, the consequences of online echo cham-
bers and polarization could have been overestimated. Extremist views exist, and
extremist people are the ones who generally tend to accumulate more evidence
supporting their views, social media or not. If the views of more moderate people
become extremist, political instability follows, but I would focus, as Turchin sug-
gests, on reducing inequality, more than on regulating Twitter. This, naturally, is
material for a book very different from the one you are reading.

Notes

1. https://​www.nytimes.com/​2011/​12/​25/​opinion/​sunday/​arab-​spring.html
2. https://​www.weforum.org/​agenda/​2016/​08/​the-​biggest-​threat-​to-​democracy-​your-​
social-​media-​feed
3. Myers and Kaplan (1976)
4. Sunstein (2002)
5. Henrich (2015)
6. Nickerson (1988)
7. Mercier and Sperber (2011, 2017)
8. Trouche et al. (2014)
9. https://​w ww.facebook.com/​C ancerTruth.FanPage/​, https://​w ww.facebook.com/​
ScienceNOW/​
10. I  am referring here mainly to Quattrociocchi et  al. (2016), but see also:  Bessi et  al.
(2016); Del Vicario et al. (2016)
11. ht t p s : / / ​ w w w. w a s h i n g t o np o s t . c o m / ​ n e w s / ​ i n - ​ t h e o r y / ​ w p / ​ 2 0 1 6 / ​ 0 7 / ​ 1 4 /​
confirmed-​echo-​chambers-​exist-​on-​social-​media-​but-​what-​can-​we-​do-​about-​them/​
12. Sunstein (2018)
13. Conover et al. (2011), Garimella and Weber (2017)
14. Cited in Gentzkow and Shapiro (2011)
15. Gentzkow and Shapiro (2011)
16. Bakshy et al. (2015)
17. Barberá et al. (2015)
18. Shore et al. (2018), Feld (1991)
19. Fletcher and Nielsen (2017)
20. Boxell et al. (2017)
21. http:// ​ w ww.culturalcognition.net/​ blog/ ​2 018/ ​ 1 / ​ 1 7/ ​ meet- ​ t he- ​ m illennials- ​ p art- ​ 3 -​
climate-​change-​evolution-​and-​gen.html
22. Dubois and Blank (2018); Fletcher et al. (2015)
23. Barberá (2014)
24. Turchin (2016)
25. https://​www.allsides.com/​, https://​www.theguardian.com/​us-​news/​series/​burst-​your-
​bubble
 6
Misinformation

Everybody can spread misinformation

That lies are better spreaders than truths is not only a recent concern. More than
three centuries ago, Jonathan Swift lamented that “if a lie be believed only for
an hour, it hath done its work, and there is no further occasion for it. Falsehood
flies, and truth comes limping after it.” And yet, “post-​truth” became the Oxford
Dictionaries’ Word of the Year in 2016. Did something change? The Oxford
Dictionaries’ motivations do not mention technology explicitly, but the rise of
digital communications and in particular social media is considered one of the
main—​if not the only—​cause of the unfortunate current situation. Of course, it is
easy to find countless examples of spreading misinformation that precede the ad-
vent of the internet, or that were disseminated by traditional mass media such as
newspapers, journals, and television in recent years, so in these cases the role of the
new digital technologies seems scarcely critical. To be in a post-​truth era, we need
to have been before in a truth era, and this is a non-​trivial assessment.1
Few plausible reasons have been put forward to elucidate the specific role of
digital technologies in the diffusion of misinformation. One is that the cost of
producing and diffusing information has plummeted. In 1835, the newspaper
New York Sun ran a six-​part special dedicated to “the publication of a series of ex-
tracts from the new Supplement to the Edinburgh Journal of Science.” The special
detailed the discoveries allegedly made by Sir John Herschel, a real British astron-
omer who was unaware of the newspaper’s report, “by means of a telescope of vast
dimensions and an entirely new principle.” Among the discoveries, you may have
imagined, there was that of the existence of life in space: on the moon, to be precise.
After describing the technical features of the telescope, and even indulging in some
scientific methodology, the twist arrived in a climax only in the fourth part of the
series, with the description of creatures that were “covered, except on the face, with
short and glossy copper-​colored hair, and had wings composed of a thin mem-
brane, without hair, lying snugly upon their backs,” dubbed “Vespertilio-​homo, or
man-​bat.” Crazy as it may seem today, the publication of the series was widely cov-
ered by other media outlets and set off a debate between sceptics (including Edgar
Allan Poe) and believers.
The story is often reported as showing one emblematic step in the road to suc-
cess of a new, sensationalistic, advertised-​based, model of publishing, but it seems
that the causal chain goes in the opposite direction. It was not the moon hoax that

Cultural Evolution in the Digital Age, Alberto Acerbi. Oxford University Press (2020) © Oxford
University Press.
DOI: 10.1093/oso/9780198835943.001.0001
118 Misinformation

boosted the success of the New York Sun, but it was its already high circulation
and reach that made the moon hoax a success. Newspapers such as the New York
Sun had a large distribution because of their prices (revenues from advertise-
ments covered the costs and the cover price was sensibly lower than the competi-
tors, hence the name “one-​penny” journals), their distribution (the New York Sun
was the first to use titles-​shouting “newsboys” selling the paper on the street) and
the usage of peculiar technological advancements, such as the newly-​introduced
steam-​powered printing press. Edgar Allan Poe, while criticizing the Sun, had tried
himself to publish a similar hoax a few months earlier, in another newspaper, The
Southern Literary Messenger. The paper, however, had limited circulation and Poe’s
attempt hardly made any impact.2
Until a few years ago, only dominant media outlets with widespread reach,
powerful governments, and big corporations had the power to spread misinfor-
mation. In 2016, reports of Macedonian teenagers who were making cash by pro-
ducing and publishing so-​called “fake news” especially targeted to the US market
made the media rounds.

The first article about Donald Trump that Boris ever published described how,
during a campaign rally in North Carolina, the candidate slapped a man in the
audience for disagreeing with him. This never happened, of course. Boris had
found the article somewhere online, and he needed to feed his website, Daily
Interesting Things, so he appropriated the text, down to its last misbegotten
comma. He posted the link on Facebook, seeding it within various groups de-
voted to American politics; to his astonishment, it was shared around 800 times.
That month—​February 2016—​Boris made more than $150 off the Google ads on
his website. Considering this to be the best possible use of his time, he stopped
going to high school.3

According to the report, Boris went on to assemble five to ten articles about Trump
and Clinton every day until November’s 2016 election. He registered websites with
names such as PoliticsHall.com, USAPolitics.co, and even NewYorkTimesPolitics.
com.
It is difficult to assess if these stories are downright true, if, for example,
“Macedonian teenagers” were acting independently and they were as politic-
ally naïve as they are made to appear, but one thing is clear. Forget printing press
and newsboys: anybody with a laptop, an internet connection, and a basic know-
ledge of programming can nowadays share pieces of information hoping they
will spread far and wide. From a purely technical point of view, this is true. As
we saw in the chapter Popularity, however, reach cannot be taken for granted.
Things climb towards success in the cultural arena in a haphazard manner and
the vast majority of them never succeed. A possibility is to fabricate information
serially—​as Boris did. Technology allows this a Wordpress website can be set up
 Everybody can spread misinformation  119

in minutes, articles can be copied and pasted from somewhere else in seconds and
straightaway shared on social media. Twitter bots are supposedly flooding social
media with coordinated and wide-​scale attacks. However, this only increases your
probability infinitesimally in a market in which there are billions and billions of
other competitors. For every Boris out there, there are hundreds of thousands of
websites with no traffic, abandoned blogs, and YouTube videos with a handful
of plays.
In addition—​this is an obvious point but it is worth spending some time on it—​
the fact that publishing and spreading information is fast and cheap, and can poten-
tially be affordable by individuals, also makes it easier at least in principle to spread
true information. The majority of scientists would probably agree that the diffusion
of digital media represented a net positive change for academic practices. A para-
digmatic case is the publication, in 2010, of a hi-​profile paper published in Science
claiming the discovery of a bacterium, found in a meteorite from Mars, that could
grow using arsenic instead of phosphorus, an element believed essential to sustain
life. The implication was nothing less than the possibility of radically diverse life
forms. Two days after the paper was made public, Rosie Redfield, a microbiologist,
published a blog post in which, while not excluding that such bacteria could exist,
vigorously criticized the methodology of the paper, casting serious doubts on the
findings. Redfield did not expect “anyone other than a few researchers to ever read”
the blog post, but in fact it generated thousands of reactions from fellow scien-
tists, who also found problems in the paper and then teamed up with Redfield. Two
years later, after performing several follow-​up analyses—​some of which needed
the expertise of other scientists who volunteered through the blog—​a manuscript
was posted on the arXiv preprint server, and then also published in Science. This
closed the case, for the time being, of the possibility of arsenic-​based life, although
as I write the original paper has still not been retracted, and it has been cited almost
500 times according to Google Scholar.4
The storyline may be familiar to many scientists. Papers are published and their
claims can be discussed in real-​time in social media. In different institutions, in
different countries, scientists working on the same topic can exchange their im-
pressions, re-​do the analyses and share their results. If you are interested in social
and human sciences, and if you have not been living under a rock for the last ten
years, you will be aware of the “replication crisis,” or “replication revolution” if you
see the glass half-​full, that hit various disciplines, in particular psychology. In a
nutshell, a surprisingly high number of published studies failed to replicate, casting
doubts on the original findings and, more generally, on certain aspects of scientific
practice itself, such as the traditional peer-​review process, or the preference of high
impact journals to publish “sexy” results. Although many other factors may have
contributed to these developments, it is undeniable the central role played by social
media, blogs, and preprint servers—​all technologies that allow individuals to share
information independently, cheaply, and quickly. Andrew Gelman, statistician and
120 Misinformation

a pivotal figure in the movement, as a critic of poor methodological practices, was

explicit on this subject when he wrote in his blog:

When it comes to pointing out errors in published work, social media have been
necessary. There just has been no reasonable alternative. Yes, it’s sometimes pos-
sible to publish peer-​reviewed letters in journals criticizing published work, but
it can be a huge amount of effort. Journals and authors often apply massive resist-
ance to bury criticisms.5

Not surprisingly, the scientific establishment reacted against these changes.

Criticism coming from “the new media” has been denigrated as originating from
“un-​referred,” “un-​curated,” or “unfiltered” sources, as opposed to the gold standard
of the traditional peer-​review system, confirming indirectly Gelman’s statement.
Of course, things are not black and white: there are some good reasons to be cau-
tious about social media and blogs as channels for informal post-​publication peer-​
review. For example, concerns about the tone of criticisms are sometimes justified;
it is likely that few published articles generate sufficient interest to be properly
scrutinized voluntary by fellow scientists; or, it has been noted, almost all cases of
post publication peer-​review concerns negative considerations (but this may be
due to the fact they target articles that have already passed peer-​review, so are sup-
posed to be accurate).6
It is not just technology-​savvy individuals who exploited the falling cost of pro-
ducing and diffusing information through the digital channel, mainstream media
have also done so. In 2013, I took the time to collect data on the spread of a story
about Oreo cookies. According to the news, scientists had demonstrated that Oreo
cookies were as addictive than cocaine, or possibly even more so. The story spread
and faded very quickly, in a couple of days in October 2013, but was reported by
hundreds of English language media outlets, including prominent ones such as the
Huffington Post and The Guardian.7 The story was based on a laboratory experi-
ment with rats. Some of the rats were placed in a maze with Oreo cookies at one
end, and a control substance—​a rice cake—​at the other end. Other rats were placed
in an analogous maze but, this time, at one end they were given an injection of
cocaine, and at the other they were given an injection of a saline substance (the
control, equivalent to the rice cake in the other maze). According to the official
press release, the researchers found that rats in the two conditions were spending
the same amount of time in the Oreo end and the cocaine end. Apparently the
researchers also “used immunohistochemistry to measure the expression of a pro-
tein called c-​Fos, a marker of neuronal activation, in the nucleus accumbens, or the
brain’s pleasure centre” and found that “Oreos activated significantly more neurons
than cocaine.”8
The majority of subsequent articles that I collected are interchangeable. Media
reported the study in a standard format, with various successful additions. Rats
 Everybody can spread misinformation  121

were, like most of us, breaking the cookies in the middle and eating the cream first.
A senior researcher commented that, after having seen the results of the experi-
ment, he was sadly not able to enjoy eating Oreos any more. A minority of articles
were more cautious. Besides the dubious extension of results from rats to humans,
the fact that rats were spending the same amount of time on the Oreo and on the
cocaine end of the maze obviously does not say anything about the relationship be-
tween the two, but only that, in both cases, Oreo and cocaine were preferred to the
alternatives.
Moreover, it is sufficient to look to the official press release to discover that
the “scientific study” was, in reality, a student’s project, not published, and thus
had never gone through peer-​review, and had not even been presented pub-
licly. There is a reference to a future presentation to a “Society for Neuroscience
conference,” but I am not able to find any information on it. Strangely, an ac-
tual scientific paper on the topic had appeared in a legitimate scientific journal,
Addiction Biology, more than a year earlier, from unrelated authors in a different
institution, and it is not mentioned in any of the media, including the critical
ones. This study had all the same eye-​catching features—​Oreos, cocaine, rice
cakes—​but it concluded, more modestly, that “greater sensitivity to the motiv-
ational properties of palatable foods may be associated with individual differ-
ences in vulnerability to the reinforcing effects of cocaine,” that is, the individual
rats that were more sensitive to the food rewards were also the ones more sensi-
tive to the drug.9
This dynamic resembles many other internet phenomena. First, it spread and
faded very quickly (remember the “Kiki challenge”). Its diffusion timeline showed
a rapid spreading, with all mentions appearing on the same day, or in the days im-
mediately after the Connecticut College press release. The few critical articles and
blog posts followed in the next days, as a reaction to the first peak of popularity.
Interestingly, periodical appearances of the same “discovery” keep on popping up,
with low frequency, including a 2017 article on Fox News, which, under the pre-
sumably catchy title “6 things you do not know about Oreos,” dug out the same old
addiction narrative. In fact, the same successful event (the Connecticut College
press release) was itself a reappearance of a story already circulating, but that never
reached popularity.10
In addition, the content of the story is culturally palatable (pun intended). It
mixes a strong intuitive appeal (everybody likes Oreos and we all know it is very
difficult to stop eating them when you have started—​don’t we?) with a surprising,
but not-​too-​unexpected, feature (can they really be like cocaine?). It appeals to
a topic, such as food, and in particular, overeating, which has both a day-​to-​day
relevance for all of us, and is commonly recognized as a contemporary, urgent,
problem of the industrialized world. On top of this, the story is presented as “sci-
ence:” researchers “discovered” or “proved” a surprising effect of Oreos, through
“experiments,” and they even “used immunohistochemistry”, whatever this might
122 Misinformation

be. Not many accounts reported the fact that there was not a proper published
study, a piece of information probably considered not relevant for the majority of
the intended readers.

The illusion of consensus

I spent time on these examples because they illustrate a few points that, although
seemingly self-​evident, are important to keep in mind and are often overlooked in
the discussions about the spread of online misinformation. Digital technologies
allow almost everybody to spread information in a way that just was not possible
before. Until a few years ago, besides isolated and lucky cases, one needed to sit on
the editorial board of the New York Sun to propagate a hoax, but today this can be
done by anybody with internet access, a digital device, and some basic technical
knowledge. Still, reach, as we called it, does not by itself explain the diffusion of
misinformation. The vast majority of attempts are ineffective, single individuals
succeed in spreading useful and reliable information too, and mainstream media
outlets are themselves not faultless when disseminating false or, at least, untrust-
worthy information.
There are more substantial reasons, however, to link widespread connectedness
and misinformation. Anthropologist and psychologist Pascal Boyer had speculated
that the diffusion of digital technologies can favor the spread of falsehoods because
it makes it more likely to find someone else that believes the same thing, no matter
how implausible it is. Imagine you start to ponder on whether the moon exists or
not, the longevity of Queen Elizabeth II is due to her habit of eating human flesh,
or Donald Trump is in fact an alien. As you are a sensible person, you are hesitant
to check with your friends and family, but by googling it you quickly discover that
there are some other people that have had the same views (I searched for all three
topics and, yes, there is material on all of them). As we saw earlier in the book—​on
the positive side—​the availability made possible by digital technologies, coupled
with our willingness to share information with others for apparently no gain, gen-
erates enormous effective cultural populations, where it is likely to find almost an-
ything. This is good if you have a niche interest or if you search for a solution for a
very specific problem, but it also allows finding at least some people who suggest
that Queen Elizabeth II is nothing less than a cannibal, which would be extremely
unlikely to find in your circle of friends.11
In addition, even a few examples of confirmatory information may persuade you
that your belief is widespread, and potentially convince you that is worth contrib-
uting to spreading it further. Psychologists have called it false consensus effect: we
believe our opinions are widely shared by others. False consensus effect has also
been studied in extremist online communities. American neo-​Nazis tend to over-
estimate the proportion of the US population that thinks “that we have gone too far
 The illusion of consensus  123

in pushing equal rights in this country” and radical environmentalists do the same
for the proportion of people judging that “globalization is a bad thing.”12
As we saw in the previous chapter, there are grounds to think that a myside bias
exists, and that it influences how we collect information online. The idea that an
overestimation of consensus for niche, or just weird, beliefs, coupled with extreme
availability, boosts the spread of misinformation is a plausible one. How convincing
though? There are a few reasons to be cautious. The false consensus effect is a ro-
bust finding, but its size is rather limited. Neo-​Nazis thought that 50 percent of the
population was skeptical of the equal rights agenda, though the real proportion, ac-
cording to the survey used by the researchers, was 44 percent. Somewhat stronger,
radical environmentalists on average estimated that 44 percent of Americans saw
globalization negatively, as opposed to the 31 percent that was found in the survey.
What would be the effect for less common beliefs? What would be the overesti-
mation for a belief such as, to remain in topic, that Hitler is still alive? This is not an
idle question. Example of relatively common beliefs can be found easily offline, so
online connectedness does not make a difference. For digital technologies to boost
the spread of misinformation via the false consensus effect, this should work on
rare beliefs, beliefs that are unlikely to be encountered offline.
Perhaps a more relevant example is vaccine hesitancy. Although vaccination
rates are increasing worldwide and remain high in industrialized areas, a growing
number of parents in western countries request alternative vaccination schedules
or even wish to decline to vaccinate offspring altogether. Mistrust of vaccines is an
old phenomenon that accompanied the introduction of vaccines themselves: the
question is whether, as often claimed, digital technologies and, in particular, social
media has increased vaccine hesitancy. Availability and illusion of consensus could
be important factors here. It may be unlikely that your offline contacts share your
same preoccupations about vaccinations as, after all, the great majority of people,
when they can, vaccinate their children, and for good reasons., but online you will
certainly find information confirming your doubts, and probably making them
worst. The same reasoning, however, can be made for any opinion that is locally in
the minority. If one is surrounded by people who think that human activity does
not cause climate change, they will find plenty of available information online to
confirm their (correct) opinion that human activity does cause climate change. As
before, availability by itself, does not explain the spread of false information.13
Allow me a short digression. To put this in perspective—​we will return later
on why negative sentiments on vaccinations are good spreaders—​do we have any
suggestion that the diffusion of digital media and vaccine hesitancy are related?
As the World Health Organisation suggests, measles immunization coverage is a
good indicator of immunization programs in general, and of the sentiments to-
wards them. What is the situation if we look to the percentage of children aged
between 12 and 23 months who have been vaccinated against measles in the 28
states of the European Union from 2010 (an arbitrary point, when social media
124 Misinformation

starts to have a substantial penetration in European Union countries) to 2016 (the

last year for which data are available)? In almost half of the states, 12 out of 28, there
has been indeed a decrease in this period of time. Countries such as Romania and
Italy have a proportion of immunization abundantly below 90 percent (86 percent
and 85 percent, respectively), and they both had in 2017 outbreaks of measles with
more than 5,000 cases reported. This is certainly an alarming trend. The 12 coun-
tries in alphabetical order are: Bulgaria, Croatia, Greece, Finland, Estonia, Italy,
Latvia, Lithuania, Netherlands, Romania, Slovak Republic, and Slovenia.14
It would be far-​fetched to associate the presence of a negative trend in measles im-
munization with digital media penetration. In fact, it looks like the opposite: coun-
tries with less estimated internet users are prominent in this list. A more reasonable
conclusion is, however, that social media has simply not much to do with it. Perhaps
the presence of vulnerable subgroups, remaining difficulties in access, distrust in
doctors, scientists, and more generally institutions, can explain, with possibly some
exceptions, why some countries had negative vaccination trends.15
There is no motive to deny that the false consensus effect works on uncommon
beliefs, making them illusorily widespread, but consensus does not by itself guar-
antee acceptance. It has been a recurrent motif in all the chapters so far we are wary
learners; we evaluate with flexibility the social cues attached to cultural traits; we
follow the majority and listen to the influentials, until we do not. Finally, as we
discussed at length in the previous chapter, it may be that the ideas of selective ex-
posure to information online and consequent reinforcement of own beliefs have
been overstated. Yes, people can find enough like-​minded others on social media
to become trapped in the proverbial bubble, but this is not what happens neces-
sarily, or not as it is at times pictured. If you think that Trump is an alien, or if you
think that vaccines are dangerous, you can find online other who agree with you,
but there are not strong reasons to suppose that you are necessary blind to contrary
evidence also present online.

Reputational costs and anonymity

Another consequence of connectedness that, according again to Boyer, may pro-

mote the spread of online misinformation is related to what we called reach and
opacity. Imagine again you are convinced that the Queen is a cannibal, or any other
“unconventional” belief that is unlikely many other people will hold. Trying to
spread this belief in your circle of offline contacts may not only be unsuccessful, but
it will damage your reputation. At the very least, your friends and colleagues will
think you are a little offbeat and, more worryingly, they will tend to trust you less
when you try again to say something contentious. The risk is even higher if your
claim contains accusations about other persons, as misinformation online often
does. If you declare that your colleague Peter has manipulated data in his last paper,
 Reputational costs and anonymity  125

you are implicitly betting that your co-​workers will be on your side, but this is in
general far from sure and, if they are not, it is a high-​risk situation for you. (this is
true in any case, but remember we are assuming that these accusations are false).
Boyer makes a comparison with accusations of witchcraft in small-​scale, trad-
itional, societies: “You never know for sure that people will not rally around the al-
leged witch. You may pay dearly if you are the only one to level the charge against a
particular individual. That is why public accusations of this kind only occur after a
long period of discreet consultations, and in some places are never made public.”16
The opposite, it seems, of what happens online.
As usual, reach is the other side of the coin of availability that we discussed be-
fore: you can always find at least someone who believes that the Queen is a cannibal
if you are wondering about it (availability) and you can always find at least someone
you can convince that the Queen is a cannibal if you spread this belief (reach). In
addition, and more importantly, you can easily spread misinformation and rumors
anonymously online, especially if they concern accusations about someone else. An
even better strategy is to first to spread a rumor anonymously and, if it gains traction,
then claiming it and thus giving it more credibility. The diffusion of anonymous ru-
mors does not present risks for the spreader, but anonymous rumors are less trust-
worthy. When an anonymous rumor reaches a certain threshold of acceptance so
that one can be reasonably sure that other people “will notrally around the alleged
witch”, the spreader—​or someone else—​can endorse it, possibly boosting its spread.
Many online rumors have similar dynamics. A recent example is the notorious
“Pizzagate” affair, a debunked conspiracy theory according to which several high-​
level officials from the US Democratic Party were involved in a pedophilia ring
in a pizza parlour. “Pizzagate” was initially pushed by anonymous sources, and
was endorsed only after reaching a sufficient diffusion by well-​known right wing
media outlets. The story is well known. Private emails from John Podesta, Hillary
Clinton’s campaign chairman, were made public by WikiLeaks. Users on Twitter
and 4chan elaborated a quite incredible narrative that started from James Alefantis,
the owner of the Comet Ping Pong restaurant in Washington and Democratic
supporter and fundraiser. Evidence was scant—​to say the least. Proofs included,
for example, photos of children in Alefantis’ Instagram profile or the fact that
the restaurant was allegedly decorated with “graffiti relating to sex.” I cannot re-
sist dwelling a bit: one of the “proofs” consisted of the presence of a logo with two
crossed Ping-​Pong rackets in the menu (the name of the restaurant is Comet Ping
Pong). They represented, according to the conspirators, a butterfly, a symbol used
by pedophiles to identify their sexual preferences. And that was not all: on the logo
was written “Play—​Eat—​Drink” which, of course, can be abbreviated PED.17
Anonymity and the unlikelihood of real, physical, contacts can indeed promote
the spread of misinformation. However, it is not a necessary result. Remember
the anonymous Wikipedia contributors we met earlier in the book? Their edits
were at least as good as the edits of the registered users. Not only did the absence
126 Misinformation

of reputational returns not prevent them from making useful contributions, but
also did not encourage them from making harmful ones. Adding swear words in
a Wikipedia article is different from spreading misinformation but, as we will see
later, the pattern is similar: fake news or conspiracy theories are, as much as van-
dalism in Wikipedia or extremist, polarized, websites, very visible and talked about
online, but their relative magnitude and effect may be lower than we often believe.
Interestingly, anonymous forums can also self-​regulate and generate mechan-
isms that allow, up to a point, the harmful effects of anonymity to be managed.
4chan, one of the seeds of the Pizzagate conspiracy theory, is an image-​based bul-
letin board, with boards dedicated to different topics. In one of the 4chan boards,
known as /​b/​or “random,” variously referred to by the media recently as a “meme-​
generating cesspool,” or the internet’s “rude, raunchy underbelly,” anonymous
posts account for more than 90 percent of the total. /​b/​is raunchy indeed: go and
check at your own risk and not if there are other people around you (to its par-
tial merit—​sort of—​/​b/​is where LOLcats memes and the practice of rickrolling
originated). Reputational devices aimed at tracking users’ identity spontaneously
emerged in /​b/​. One is “timestamping”, or posting a picture of themselves with
a note containing the current day and time. Others are aimed at claiming group
identity, such as using peculiar terms or even patterns of characters that cannot be
copied and pasted from one post to the other, but need to be recreated with com-
plicated Unicode strings of codes. In sum, even when anonymity is quasi-​enforced
and positively valued, as it is on /​b/​, users can take advantage of it to spread misin-
formation and accusations, but they can also head the other way, displaying their
identity to signal influence and credibility.18
Furthermore, if the positive outcomes of /​b/​are not much more than launching
the tradition of sharing disguised hyperlinks pointing to Rick Astley’s Never
Gonna Give You Up, anonymity may be positive, even necessary in some contexts.
Anonymity online has been shown to contribute fostering community identity
and group commitment, providing greater scope for adversarial opinions, espe-
cially from individuals in disadvantaged groups or in lower hierarchical positions,
or stimulating creative thinking. Obviously, there are negative as well as positive
aspects:  community identity promotes helping behavior and cooperation, but
also hostility toward out-​groups; adversarial opinions can be legitimate and con-
structive or not. The point is: anonymity and the absence of physical proximity
could promote the spreading of misinformation and false allegations as they lower
the reputational costs for the spreader, but they do not necessarily do so.19

Optimization for (shallow) engagement

A subtler factor that may favour the spread of misinformation online is the way
in which algorithms decide what we see in our social media newsfeeds or in our
 Optimization for (shallow) engagement  127

Google searches. In the chapter Cumulation, we will explore in more detail the role
of algorithms and their pitfalls and advantages, but for now it is enough to say that
our Google results or the tweets we see in our timeline are not an unbiased reflec-
tion of what is present online, as everybody know. Twitter, for example, changed
in 2016 the way tweets were shown to users, from a purely “chronological” presen-
tation to a selection made by an algorithm of “interesting” tweets or tweets from
“top-​users.” The change was received with much criticism and several articles ap-
peared about how to opt-​out from the new algorithmic timeline and back to the old
chronological one. As far as I know, two years later the buzz was over and every-
body quietly accepted the algorithmic version. It seems that, when faced with such
an abundance of information there are simply no viable alternatives -​more on this
later on.
We do not know—​regrettably—​the details of how social media algorithms
work, but some guesses are not too difficult. Facebook may present in the time-
line posts that have received more “likes” and more comments and links that have
been clicked more than others. An obvious consequence is that posts prompting
for an immediate reaction will climb to the top of the newsfeed. Pictures of your cat
doing a funny face (or of your kid doing a funny face) are a relatively benign con-
sequence, as much as the wedding post that precipitates dozens of congratulatory
messages. The same, however, is true for controversial political opinions, which
will cause many comments, or for sensational stories that will generate more inter-
ests and will be “liked” more.
In 2013, the website Upworthy became an internet sensation by utilizing a pe-
culiar style for titles which become known as the “Upworthy model” or, more to
the point, “click-​bait model”. Upworthy top-​hits in 2013 included titles such as
“His first 4 sentences are interesting. The 5th blew my mind. And made me a little
sick,” “9 Out Of 10 Americans Are Completely Wrong About This Mind-​Blowing
Fact,” or “This Kid Just Died. What He Left Behind Is Wondtacular”—​the highest
ranked.20 The Upworthy model has been, not surprisingly, harshly criticized and
has generated an endless series of parodies, but it worked. People clicked on the
links, and Upworthy’s articles were pushed to the top of social media feeds.
Upworthy’s titles exploit a basic emotional language and arouse curiosity but
do not provide enough information without clicking on them. It does not really
matter whether the stories linked are true or false and, in fact, the stories do not
need to be particularly attractive themselves, as long as users are tempted to click
on the link. In such a system, misinformation can be advantaged. “False” news can
be manufactured building on features that make them attractive in an almost un-
constrained way, whereas “true” news cannot, simply because they need to cor-
respond to reality. Misinformation can be designed to spread more than real
information. Notice this does not need to be a conscious process, as is the case of
Upworthy: giving enough combinations of stories, or simply titles, some of them
will have the right features.
128 Misinformation

This is how I interpret the results of a recent study, where researchers from MIT
tracked more than ten years of the spread of “false” and “true” news on Twitter,
collecting more than 4.5  million tweets. They reached the conclusion, echoing
Jonathan Swift, that “Falsehood diffused significantly farther, faster, deeper, and
more broadly than the truth in all categories of information. [ . . . ] Whereas the
truth rarely diffused to more than 1000 people, the top 1% of false-​news cascades
routinely diffused to between 1000 and 100,000 people. [ . . . ] It took the truth
about six times as long as falsehood to reach 1500 people and 20 times as long as
falsehood to reach a cascade depth of 10.”
There are a couple of interesting points to highlight here. First, the conclusion
is not as grim as it seems. To collect “false” and “true” news, the researchers con-
sidered all the tweets where someone had posted a link to a fact-​checking web-
site, such as Snopes, so what they were comparing was more accurately rumors that
were debunked versus confirmed rumors. Nobody would check on Snopes whether
Donald Trump has been elected President of the US, a “true” news that, for this
reason, was not considered in the study, and surely reached more than 1000 people.
However, but someone could check whether Queen Elizabeth wore a brooch
gifted by Obama while meeting Trump (rumor debunked) or whether Amnesty
International accused Trump of violating and endangering human rights in the
United States and around the world (confirmed). In sum, what the study says is
that rumors that were subsequently debunked “diffused significantly farther, faster,
deeper, and more broadly” than rumors that were subsequently confirmed.21
Second, consistently with what we discussed in the previous chapters, “struc-
tural elements of the network or individual characteristics of the users involved
in the cascades” did not contribute to explain why false rumors were more suc-
cessful than true ones. In other words, false rumors are not popular because they
are diffused by influential accounts, or accounts with many followers. In fact, the
opposite is true: users who spread false rumors have on average less followers, and
are less likely to be verified users. (Verified users are accounts of public interest—​
politicians, journalists, celebrities, etc.—​that are certified authentic by Twitter,
hence likely to be “influential” figures. A blue badge next to the name identifies
them.) The researchers also checked the role of the infamous bots, and re-​analyzed
the data after excluding the activity that was likely to be initiated by them: the re-
sults were the same, so that they concluded that false news is widespread “because
humans, not robots, are more likely to spread it.”
Where is the difference between false and true rumors then, and what make the
former more successful? The answer, according to the study, lies in their content.
Analyzing the emotional content of the tweets sent in reply to the news links, using,
as we already discussed, the presence of certain keywords associated to emotions,
the difference was that the reactions to false rumors were characterized by more
surprise and more disgust. Thus, “viral” news is not false news, as opposed to true
 How pervasive is online misinformation?  129

news, or news diffused by influential accounts or by bots, but is—​even online—​

news that elicits certain emotional reactions.
As usual, the take-​home message is not clear-​cut: as we argued, real news (and
not “debunked rumors”) is likely to spread more than false news in any case;
other emotional reactions, such as joy or trust, were found higher in the replies
to debunked rumors than in the replies to confirmed rumors, so that it is not the
emotional content by itself that makes some news better spreaders than others.22
However, as we will see shortly, these results go in a direction that should not be too
surprising for cultural evolutionists.

How pervasive is online misinformation?

To sum up, there are various reasons why the spread of misinformation can be fa-
vored online, but none of them seem decisive. Yes, everybody can spread informa-
tion quickly and cheaply, but this does not imply, taken alone, that individuals are
more likely to diffuse misinformation in respect to other content. Any, or almost
any, bizarre belief can be confirmed online, and this may push people to spread
it further, but confirmation bias, or myside bias, as we called it in the previous
chapter, is not a blind force—​neither online nor offline—​and it does not explain
why one would hold and be willing to spread the bizarre belief in the first place.
Online diffusion guarantees anonymity and it makes it very likely that you will
find somewhere, someone, willing to believe you, but anonymous individuals
often share true information and collaborate, without apparent gain, to collective
projects. Algorithms that optimize shallow engagement, such as “like” and links
clicks, can favor misinformation, but they do it as an unintended consequence of
favoring attractive content. As we will see later, it might be that to understand and
possibly counteract the spread of online misinformation one needs to understand
what content is attractive, and why is it so, more than the idiosyncrasies of online
transmission.
First, though, the usual question: is it really the case that misinformation on-
line is more pervasive than misinformation offline? It is not easy to answer this
question directly, but we can at least try to quantify more precisely the strength
of misinformation online. As we just saw, the study showing that falsehood dif-
fuses online better than truth needs to be considered in its specific context, that is,
the comparison of confirmed versus debunked rumors. What about true truths?
Contrary to the prevailing opinion, many quantitative studies show that the spread
of misinformation online is, after all, limited. A recent example is an analysis of
11.5 million tweets on a politically charged theme such as immigration during the
first month of Donald Trump’s presidency, from 20 January to 20 February 2016.
Researchers at the Pew Research Centre looked at tweets with links to articles on
130 Misinformation

immigration-​related topics, and identified the most popular 1,030 websites that
were linked by the tweets.
The researchers used two complementary ways to assess whether the links were
pointing to fake news. The first simply involved comparing the sites found in the
analysis to the websites included in three authoritative lists of websites known for
publishing hoaxes and misinformation (produced by BuzzFeed, FactCheck.org, and
Politifact). Of the 1,030 sites included in the analysis, only 18 (2 %) were found in
these lists. As a second indicator, the researchers used an indirect proxy: whether
websites were newly created or not. Many websites spreading misinformation had
been created ad hoc during the US electoral campaign, either with the goal of simply
making money (remember Boris at the beginning of this chapter) or with the pur-
pose of conditioning political outcomes, or both. Websites registered only after the
first of January 2015 were considered “newly created.” The results are slightly more
complicated to present, because the researchers separated the websites into various
categories, under the main division in News Organizations (including both online
versions of “traditional” news outlets and web-​born outlets) and Other Information
Providers (including blogs, non-​profit organization sites, official government sites,
academic websites, and “digital-​native aggregator sites,” that is websites that do not
produce original content). In all categories, except one, the majority of websites
were established before January 2015, more than one year before the elections. The
only category that presented a mix of old and new websites was, not surprising, the
digital-​native aggregator sites, where newly created websites represented 48 per-
cent. However, digital-​native aggregator sites accounted for only 3 percent of the
total of the 1,030 highly linked websites, so that half of 3 percent is a tiny minority.
As we have already seen when discussing echo chambers, extremists, or polarized,
or misinformation-​spreading, or often a combination of all websites do exist, but
they tend to have a much smaller audience with respect to legitimate ones.23
Analogous results come from European data from France and Italy. In 2017, the
reach of 38 “false news” websites in France and 21 in Italy was compared with the
reach of the most prominent news websites in these countries. In Italy, for example,
the most successful false news website, Retenews24, reached, each month, 3.1 per-
cent of the online Italian population. The other websites follow with figures that
quickly go below 1 percent (the sixth website in the rank is reached by 0.9 percent of
the population). In comparison, the websites of the newspapers La Repubblica and
Corriere della Sera, by far the two most popular online news outlets in Italy, reach
on average every month around half of the population:50.9 percent and 47.7 per-
cent, respectively. The situation is similar in France, with the difference that the
access to prominent legitimate news websites is more fragmented. The most prom-
inent one, Le Figaro, reaches “only” 22.3 percent of the online population.
If we look at the cumulative Facebook interactions (the total number of com-
ments, shares, and reactions, such as “like,” “love,” “angry,” etc.) of the websites
the results are apparently more worrying. Two websites classified as “false news”
 The effects of online misinformation  131

providers—​Santé+, a health-​related site, and Le Top de l’Humour, a mix of satire,

memes, and “traditional” misinformation—​ outperformed all legitimate news
outlets in France. In Italy the situation was better, with the big news providers
remaining in top positions also for Facebook interactions. The most successful
false news website was Io Vivo a Roma, a site that publishes local news about the
Italian capital in addition to articles identified as misinformation (at the time of
writing, the most-​liked article on Facebook on the homepage of Io Vivo a Roma
is about the antibodies allegedly present in dogs’ saliva, with an Upworthy style’s
title: “Does your dog lick you often? You can’t imagine what happens—​An incred-
ible discovery,” which in about one month was liked more than 70,000 times on
Facebook). The factsheet, compiled by Oxford researchers, concludes however that
“in most cases, in both France and Italy, false news outlets do not generate as many
interactions as established news brands.”24
Another, more recent, research, conducted again in France before the 2017
presidential election, gives some reassuring news too. The team of researchers used
the names of key players in France’s political elections and specific political-​related
keywords to collect almost 60 million tweets for a period of 11 months. They also
relied on an external list of possible “fake news” targets. In this case, however, it
was not a list of websites, but a database compiled by the French newspaper Le
Monde of all the URLs and Facebook posts associated to 179 debunked stories,
such as “More than 30% of Emmanuel Macron’s campaign was financed by Saudi
Arabia,” They found less than 5,000 shares of these links, which corresponds to less
than 0.01 percent of their sampled tweets. The researchers comment, with some
understatement, that, whereas there is the possibility that these figures are under-
estimated, “fake news are not heavily shared in France by people interested in pol-
itics on Twitter”.25

The effects of online misinformation

Perhaps less than what we fear, but misinformation spreads online. However, it is
one thing if many of us enjoy reading the odd silly story or taking a distracted look
at the last dietetic fad from time to time, but it is a different thing if misinformation,
fake news, and the like have strong effects on our day-​to-​day lives. Many academics
and political commentators, as is well known, have identified online misinforma-
tion as one of the main reasons behind recent political events such as the result of
the Brexit referendum in the UK or the election of Donald Trump as President of
the United States.
When researchers tried to quantify the real effect of misinformation in the
case of Trump’s success, however, they found mixed results. Some of them are ex-
pected:  social media had a relatively important role as “gateways” to fake news
websites. Using web traffic data from a sample of a few thousands of Americans in
132 Misinformation

the period before the election, researchers were able to show, first, that their con-
sumption of fake news (identified, as before, from the same authoritative lists of
suspect websites) was correlated to their Facebook usage. The more people use
Facebook, the more they are likely to visit websites associated with fake news. Of
course, this is an indirect measure, and possibly biased: people who spend more
time on Facebook spend more time on the internet overall, so that the result may
simply be the outcome of bigger exposure to online information in general.
However, there is a second, more direct, measure that confirms this sugges-
tion: checking the URL visited before landing on a fake news website, the same
team of researchers found that Facebook accounts for 22 percent of those. Almost
one in every four times someone visited a fake news website they were on Facebook
immediately before. Again, this could be due the fact that people are often on
Facebook, but in this case one can compare it with the number of times people
were on Facebook before landing on a legitimate news website: in this case the pro-
portion is only around 6 percent. So, we are comparatively more often on Facebook
before visiting a fake news site than visiting one which is legitimate.
Another result that fits well with the common perception is that misinforma-
tion was generally pro-​Trump. The great majority of fake news was pro-​Trump
(around 90 ). As a consequence, 40 percent of people in the sample that identi-
fied as supporting Trump and 14 percent of Clinton supporters visited at least one
article from a pro-​Trump fake news website in the one month period considered.
Pro-​Clinton websites were visited at least once by only around 3percent of Trump
supporters and by 11 percent of Clinton supporters (notice that Clinton supporters
visited more pro-​Trump fake news websites than pro-​Clinton fake news websites).
Overall, around one quarter of Americans were on one of the websites classified
as suspect for publishing misinformation at least once. This is, again, a big number,
but the interesting and possibly surprising fact is that the great majority of them
visited these websites rarely. By dividing the sample into deciles (ten equal-​sized
groups) according to their political attitudes—​from liberal to conservative—​and
by looking at the actual number of visits, one finds that a single decile, that is the
10 percent of the most conservative individuals, account for 60 percent of all fake
news consumption. This casts doubts, at a minimum, on the idea that fake news is
mainly propaganda to change people minds, while suggesting that may also be an
easy way to engage, and indirectly cash in advertising money from people who are
already convinced by something and enjoy basking in it.26
Supporting results come from a comparable study by economists Hunt Allcott
and Matthew Gentzkow. As above, they used a combination of web traffic and
survey data to assess the impact that fake news could have had on the 2016 US elec-
tion. They agree that social media was the most important channel through which
people access misinformation, and they also concur that, aggregating by simple
clicks, fake news was quantitatively abundant. Sampling 65 fake news websites, they
found 41 pro-​Clinton and 115 pro-​Trump articles that were shared on Facebook
 The effects of online misinformation  133

38 million times overall. To quantify their effects on individuals, the researchers

presented, after the election, a list of fake news, true news, and, as a control, made-​
up fake news (i.e. non existent news that was never presented as fake news, such
as “Clinton Foundation staff were found guilty of diverting funds to buy alcohol
for expensive parties in the Caribbean”), asking how many of them the respond-
ents remembered and how many of them they actually believed. They showed that
about 15 percent of US adults remembered, and 8 percent believed, the average
fake news headline. However, the results were “statistically identical” for the made-​
up fake news headlines, suggesting a high rate of false recall. Adjusting for it, the
researchers conclude that it is unlikely “that the average voting-​age American saw,
remembered, and believed more than 0.71 pro-​Trump fake stories and 0.18 pro-​
Clinton fake stories.”
There is a lot of averaging here of individuals and of news. As discussed above,
there is plenty of individual variation in consuming fake news, so that it is likely
that a few individuals and possibly a few fake news items account for these quan-
tities, and the averages may not be too meaningful. And still, they are small num-
bers. If one looks at specific news items, the most recalled and remembered seem
to be associated with the headline “Wikileaks was caught by Newsweek fabricating
emails with the intent of damaging Hillary Clinton’s campaign,” for which around
half the Americans asked answered “yes” or “not sure” to whether they recalled
seeing it. Among those, around 30 percent believed it was true. This makes for
around 15 percent of respondents of the survey remembering and believing it. As
comparison, for the best performing fake fake news, that nobody could have seen,
this figure is about 10 percent. (Notice this is higher than, for example, the propor-
tion of Americans who remembered and believed the Pizzagate conspiracy).
In the Appendix, Allcott and Gentzkow present analogous survey data for
“classic” conspiracy theories. In 1963, for example, half of Americans believed that
President John Kennedy was assassinated either by a right-​wing extremist or by a
Communist (I am pulling together the data assuming that the number of people
who believed that Kennedy was assassinated by both is negligible).In 1991, more
than 30 percent believed that President Franklin Roosevelt knew about the Japanese
plan to bomb Pearl Harbor but did nothing to prevent it. Of course, these—​as well
as the 2016 election ones—​are surveys results and need to be taken with a grain
of salt. Moreover, there may be a survivorship bias for the big conspiracy theories
like the Kennedy assassination. We talk about them now exactly because they were
successful, a fact that could be enough to explain the large number of believers as-
sociated with them. However, this should help put in perspective our fears of living
in a “post-​truth” era.27
Is it the case that misinformation itself is now more available, like everything
else, so that it looks more diffuse? The fact that misinformation is not that pervasive
and that its effects are plausibly not that big, does not mean we should not strive to
have less of it. Perhaps, however, we need to think about it in a different way.
134 Misinformation

What is the situation after 2016, and the political fake news shenanigans that ac-
companied it? BuzzFeed, one of the authoritative sources for questionable websites
associated with misinformation, compiled a list of the biggest fake news “hits” on
Facebook in 2017, that is, the false articles that generated more engagement in the
social media. The top 50 hoaxes of 2017 caused 23.5 million “shares” on Facebook.
It is worth spending some time on what exactly the big hits were about. Quoting
the report:  “top-​performing fake news story in the analysis typifies the kind of
hoax that succeeded on Facebook in 2017.” Here are the top 10:

(1) Babysitter transported to hospital after inserting a baby in her vagina

(2) FBI seizes over 3,000 penises during raid at morgue employee’s home
(3) Charles Manson to be released on parole, to Johnson County, TX
(4) Police: Chester Bennington Was Murdered
(5) Morgue employee cremated by mistake while taking a nap
(6) Angry Woman Cuts Off Man’s Penis for Not Making Eye Contact During Sex
(7) Female Legislators Unveil “Male Ejaculation Bill” Forbidding The Disposal
Of Unused Semen
(8) President Trump Orders the Execution of Five Turkeys Pardoned by Obama
(9) Elderly woman accused of training her 65 cats to steal from neighbors
(10) Couple hospitalized after man gets his head stuck in his wife’s vagina

Scrolling through this list, it is hard to think of a well-​oiled propaganda ma-

chine: five out of ten are about sex, and the top-​50 list gives the same impression.28

Cognitive appeal

It is time to recapitulate the story I am trying to tell. Misinformation, fake news,
hoaxes, falsehood, you name it, are successful online spreaders. However, they may
be less successful than is sometimes feared. Legitimate information and true news
are also—​possibly more—​successful online spreaders. Particular features of online
cultural dynamics, such as the fact that everybody has the possibility of reaching an
enormous audience, that self-​reinforcing information of all kinds are potentially
available, that anonymous interactions are possible, and that algorithms opaque to
users favor emotionally laden information, may all boost the spread of online mis-
information, but not necessarily so. It is far from clear whether misinformation on-
line is thriving more than misinformation offline, echoing what we discussed in the
previous chapter about polarization. The role of fake news as a weapon of political
propaganda is at least debatable, and the content of misinformation, as we shall see
in more detail soon. seems to be important in determining its success.
Looking at misinformation through the lens of cognitive anthropology could
help make sense of this. Cultural evolutionists talk about content-​based biases
 Cognitive appeal  135

when intrinsic features of certain cultural traits favor their success. Sometimes
content biases are obvious. We mentioned, earlier in the book, how Aka pygmies
quickly replaced bows and arrows with crossbows, a more effective hunting tool.
Sometimes their effect is subtler. Olivier Morin documented how, independ-
ently, in two different cultural traditions—​European renaissance and Korean
Joseon dynasty (spanning five centuries, until the beginning of the twentieth cen-
tury)—​direct gaze portraits increased in popularity over time, likely because of a
psychological bias for which we find images of faces directly looking at us more
attractive, more attention-​catching, and more memorable than images of faces
that do not. In this case, we would not expect all portrait traditions, as well as all
portraits in a single tradition, converging toward direct eye gazing. First, there are
many aspects that can favor the success of a portrait. An averted eye portrait can
have other aesthetical features that make it likeable. Non-​aesthetic factors can have
a fundamental role when determining which portraits are successful or not in a
given society: Morin reports the case of official Korean portrait where as court eti-
quette prohibited gazing directly at others, portraitists followed the same rule).
Artistic success often is favored, at least in contemporary culture, by disruption of
the norm, so that if direct gaze portraits are successful in a certain period of time,
we may expect an averted gaze “reaction” following it. Second, there is individual
variation in our psychological preference for direct-​eye pictures: some of us can
be more sensitive to them, preferring, say, a poorly realized portrait but with a face
that looks straight at us to a good one with an averted gaze, and vice versa.29
Importantly, content-​based biases can vary in respect to their generality. Some
of them can be eminently idiosyncratic. Individuals, because of their past experi-
ences, as well as because of intrinsic differences in their gustative systems, have
different tastes. Some people love the taste of cilantro (I do), finding it zesty and re-
freshing, others can simply not eat it and they describe it as “rotten.” A content bias,
broadly defined as such, will favor cilantro in a part of the population, and work
against it in another part. Others content biases can have a wider scope, but their
reach is still limited to specific situations. Anthropological examples abound of
aesthetical preferences that are confined to a particular society or to a particular pe-
riod of time. Local ecological factors, such as the availability of certain raw mater-
ials or the weather conditions, favor, for example, different ways of house building
in different countries. The content-​based bias favoring crossbows over bows and
arrows for Aka pygmies is bound to disappear if shotguns are introduced, or when
hunting is replaced by another form of securing food—​as it is already happening.
Claiming that there is effect of a content-​based bias does not, in itself, explain
anything, but points to where we should look for explanations. Some preferences,
however, can be sufficiently wide-​ranging to be applicable in many different cir-
cumstances, such as the partiality for direct gaze pictures we just considered.
Anthropologists and psychologists have compiled a growing catalogue of universal
inclinations that make us prefer certain content towards others. The ubiquity of
136 Misinformation

cultural artefacts such as portraits (whether they look at us or not), caricatures,

masks, or made-​up faces cannot be understood without considering that we have a
natural disposition to find real faces very interesting stimuli, that we naturally pay
attention to them, that we remember them better than other stimuli, and so on. The
most common cases of pareidolia, the phenomenon for which we perceive patterns
that do not exist in arbitrary stimuli, are faces. We see faces in clouds, rocks, build-
ings, and many other configurations. Drawing two points inside a contour of any
shape is generally enough to suggest that our goal is to depict a face.30
Although it does not necessarily need to be the case, these wide-​ranging prefer-
ences are often the consequence of psychological features, resulting from the de-
sign of the human mind. In particular, an approach known as “epidemiology of
representations” or, more recently, “cultural attraction theory,” mainly associated
with the work of anthropologist and cognitive scientist Dan Sperber, has stressed
this point. While other factors—​local and idiosyncratic preferences, or ecological
influences—​are also deemed important, researchers in this tradition have often
privileged explanations based on universal cognitive principles, and for good
reasons.31 This is important, not because psychological explanations are better
than “socio-​cultural” ones, but because they allow circularity to be avoided. We do
not say that because portraits, masks, and caricatures are widespread then they are
attractive. We do say that portraits, masks, and caricatures are widespread because
they are better than alternatives (images of other body parts, for example) based
on testable, independent, explanations drawing on psychology. If new experiments
find that humans do not react preferentially to face-​looking stimuli than to other
stimuli, or that this happens in some population where portraits and masks are still
common, we would need to find a different explanation for the diffusion of por-
traits and masks.
By and large, widespread, recurrent, cultural traditions are explained with their
universal psychological appeal, which is exactly what we expect to happen, but it
also reinforces the suspicion of circularity. Ideally one should find cases in which
universal psychological appeal did not produce widespread cultural traditions, and
be able to predict the direction in which cultural evolution could develop. This is
certainly possible. Anthropologist Natasha Dow Schüll described how designers,
programmers, and psychologists work with the producers of video pokers and
electronic slot machines to create the most possible successful cultural artefacts—​
from their point of view—​that is, the most possible addictive machines. Old slot
machines were mechanical devices with limited space for engineering. The re-
sults were displayed and determined by the actual position of the reels. However,
electronic slot machines allow gamblers to be tricked by, for example, showing a
higher proportion of quasi-​wins, that is, the presence of lucky symbols just below,
or just above, the actual, and usually not winning, central sequence. Up to a cer-
tain point—​too many quasi-​wins and players are frustrated—​this is more ap-
pealing than the scarce amount of quasi-​wins showed in the mechanical device.
 Cognitive appeal  137

Slot machine designers are probably not too interested in debates on cognitive an-
thropology, but they produce successful cultural artefacts by virtue of their inde-
pendent psychological appeal. One could predict that, in general, gambling devices
that offer an illusorily higher perception of the possibility of success would be fa-
vored over alternatives.32
Attractive cultural features that derive from psychology have the advantage
of being general and of providing non-​circular explanations, but they do not
have a special status. In particular, their effect is not stronger than other factors
influencing cultural success. If anything, it is weaker in the majority of cases. A law
against gambling could, in a cultural-​evolutionary blink of an eye, disrupt our sup-
posed trend towards devices that deceive individuals with many quasi-​wins. The
influence of the rich and the powerful is—​sometimes, as we saw at length above—​
able to overcome psychological preferences, especially when it is in the interest of
the adopters (Morin uses the example of the not-​exactly-​crowd-​pleaser Brezhnev’s
biography, that still enjoyed abundant diffusion). Anti-​intuitive theories, such as
quantum mechanics, can still successfully spread—​indeed with some effort and,
especially, ample institutional support.
Feeble as they are, however, general psychological factors have the advantage, if
we are interested in explaining why some cultural traits are widespread and others
are not or why some cultural traits are easy to adopt and others not, that their dir-
ection is constant. It would not be very wise betting, at a portrait auction, that the
highest selling portrait will be a direct-​eye gaze one. There are too many factors
influencing what happens. The most viral piece of misinformation can get its status
for many reasons. Still, on many, many, auctions, in different times and cultures,
we may have some grounds to believe that direct-​eye gaze portraits will enjoy a
relative advantage. If we collect enough fake news, we may see that on average they
possess characteristics making them appealing to human psychology. Tiny effects
that go in the same direction create cumulative effects and, if we zoom-​out enough,
discernable patterns. Social influences—​popularity, prestige, and so on—​do not
have this property, so that one ends up explaining (unsatisfactorily, for me) that
Brezhnev’s biography was successful because Brezhnev was in power.33
Finally, generally attractive factors do not enjoy much sympathy among the ma-
jority of cultural evolutionists. Why is that the case? There may be several reasons.
One is likely to be that, as mentioned before, there is the impression that the work-
horse becomes our psychology, so that cultural phenomena are considered “eco-
logical patterns of psychological phenomena” (which is indeed a basic tenet of the
“epidemiology of representations” approach mentioned before).34 Another, pos-
sibly more compelling, reason is that cultural evolutionists, as we saw early in this
book, are especially interested in cultural evolution as an adaptive process that
outsmarts single individuals. The contextual biases we explored earlier do indeed
produce this outcome, on average and given the right conditions. General psycho-
logical preferences, conversely, do not generate cultural-​evolutionary outcomes
138 Misinformation

“smarter than us,” but they generate cultural-​evolutionary outcomes that are as
smart as we are or, more often than what we would like, as dumb as we are. Which
makes for a good time to go back to online misinformation.

The power of negative content

That a content eliciting a strong emotional reaction will be more likely to spread
successfully is not that surprising. However, we can be more detailed, and we can
characterize more precisely the kind of emotions that will make content successful.
First, negative emotions are better spreaders than positive emotions.
Psychologists and cultural evolutionists use, among others, the so-​called method
of serial reproduction, easier to remember as the “transmission chains” method, to
understand which content is expected to spread and to be retained along chains of
cultural transmission. The transmission chain method is the laboratory equivalent
of the Chinese Whispers Game. A participant listens to a story and then they have
to repeat it to the next one, and so on, along a chain of transmission episodes. How
does the story change along the passages? Which features are retained and which
ones are lost? A schematic representation of a transmission chain experiment is
found in Figure 6.1
Here is an example. In an experiment, researchers fed several chains of partici-
pants with a story that included both positive and negative events. The story was
about a girl flying to Australia and contained various details: negative events were
represented by particulars such as “the man in the seat next to her seemed to have
a nasty cold,” whereas a positive detail was, for example, “when (the air hostess)
returned she told Sarah that she would be moved to business class.” After a few
passages, the researchers calculated how many positive and negative events were,
on average, still in the story: negative events were remembered and transmitted
around twice as many times as positive events. Sixty percent of negative details
were still present, against only around 30 percent of positive ones. The researchers
additionally included ambiguous details in the story, such as “Walking down the
concourse, Sarah saw a young man take an old women’s bag.” Statements like the
one above can be interpreted, and retold, both in a positive light (a kind young man

Original (Transformed) (Transformed)

material material material
Reproduction Reproduction
Reception Reception

Participant 1 Participant 2

Figure 6.1  A schematic representation of a transmission chain experiment.

 The power of negative content  139

wanting to help the old woman) and in a negative one (the young man is stealing
the old woman’s bag). The majority of these ambiguous events disappeared from
the story, but for the ones that remained ambiguity was in general resolved, with a
preference, again, for a negative interpretation of the ambiguous details.35
In another experiment, participants were recruited for an online study entitled
“Truth or Trash? How Believable is the News Today?” and they were asked to
evaluate whether alleged news items were believable or not. The researchers pro-
vided the same information, but framed them either positively (“When civil litiga-
tion cases go to trial, 40 percent of plaintiffs succeed and win money”) or negatively
(“When civil litigation cases go to trial, 60 percent of plaintiffs lose, winning no
money, and often having to pay attorney fees”). They found that participants were
more inclined to believe the information that was framed negatively, even though
the content was exactly the same.36
These experiments extend a broad psychological literature pointing to the exist-
ence of a negativity bias, for which negative information is, everything else being
equal, more attention grabbing and memorable. This is likely, as the first study re-
ported hints, to have aggregative effects: negative stories will be more culturally
successful than positive ones. This suggestion is not only limited to laboratory ex-
periments. In my research, I studied how the expression of emotions changed in
printed books, mainly in English language, in the last centuries. Using different
materials—​from the more than 8 million books in the Google Books corpus to
“small data” corpora that we built by ourselves, containing classic English language
novels or unknown authors sampled from the Gutenberg library—​we consistently
found a steady and large decrease in the usage of words denoting emotion through
the last two centuries. This decrease is entirely driven by a decrease in positive
emotions, while words denoting negative emotions remain around the same. We
have here, on one side, a tendency towards a general decrease of the emotional
tone of narrative, of which we cannot yet pinpoint exactly the causes; on the other,
a tendency to privilege negative emotions, so that the general decrease is achieved
by reducing only positive emotions. Another way to see this dynamic is that the
ratio between negative and positive emotions increased in time. To get some sense
of this, in the Google Book Corpus, at the beginning of last century, for 100 words
associated with positive emotions there were around 57 words associated with
negative ones (positive emotion words are always on average more frequent than
negative ones). In 2000, the negative emotion words were approximately 69, an in-
crease of around 20 percent.37
Someone even sampled four centuries of operas, and found that in one third of
them (meaning in 112 operas of 337) there were a “completed suicide [ . . . ], non-​
fatal suicidal acts, or suicidal thoughts.” Sure, operas are notorious for their overly
dramatic plots and emotions, so what about children’s animated films? In a paper
(titled CARTOON KILLS, which appeared in the special Christmas issue of the
British Medical Journal) 45 top-​grossing children’s animated films were compared
140 Misinformation

to 90 top-​grossing dramatic films for adults. It was found that two thirds of chil-
dren movies contained at least one “on-​screen death of an important character.”
The authors write in the Discussion:

Our sample of animated films included three gunshot deaths (Bambi, Peter Pan,
Pocahontas), two stabbings (Sleeping Beauty, The Little Mermaid), and five
animal attacks (A Bug’s Life, The Croods, How to Train Your Dragon, Finding
Nemo, Tarzan), suggesting grisly deaths are common in films for children.38

That news privileges the negative is a cliché. The universally famous adagio is
“man bites dogs:” it is not “dog wins the lottery,” an equally surprising and unusual
event, but positively connoted. We can, however, check whether this cliché is true
or not and possibly quantify it. Study after study of news coverage found a con-
sistent negative bias, which goes back in time long before the diffusion of social
media and internet. The media coverage of economic news, for example, in articles
from the New York Times and broadcasts of ABC World News Tonight from 1998 to
2002, was more likely to be framed as negative than as positive: “even when eco-
nomic stories contain positive information, there may often be hedging or warning
on the part of financial experts that consumers should not become overconfident.”
Media coverage of unemployment is high when unemployment rates are relatively
higher, so perceived as negative outcomes, and low when they are lower. While
commercial flying is becoming safer and safer, in the last 25 years, media atten-
tion to aviation incidents has become greater and greater (at least in the sample of
Dutch newspapers the authors considered), suggesting that “as a consequence of
increased commercial pressures on news media institutions, media’s penchant for
the negative and the exceptional has become more prominent over time.” The data
considered, and the increasing trend, start from 1991.39
Misinformation, precisely because does not have the constraint to correspond
to reality, may be crafted to appeal to general psychological preferences. This does
not need to be a conscious process. We can imagine websites that publish posi-
tively valenced fake news and produce headlines like “The government cut taxes!,”
“Employment is increasing!,” or “Immigration rates are at historical low.” It sounds
unlikely, right? Indeed such websites, provided they exist, would not be successful.
Whether as a consequence of a conscious process—​fake news creators know what
appeal to the audience—​or of an unconscious one—​misinformation websites that
publish unattractive content are rarely visited and end up disappearing—​we can
predict that misinformation online will have certain features. One of this is that
they will be, on average, more negative than positive oriented, and this is what
happens.
As an illustration, I sampled fake news outlets from two of the authoritative lists
of suspect websites mentioned above in this chapter (provided by BuzzFeed and
Snopes) and I found 26 websites that, as I am writing, are still actively publishing
 Threats and online misinformation  141

a mix of hoaxes, political misinformation, and more or less explicit satirical news.
I extracted the ten articles that appeared first on the home page in each site, for a
total of 260 articles. It is a relatively small sample, but, exactly for this reason, it has
the advantage that the articles can be actually read and categorized for their content
by humans—​this is what we did together with a few collaborators. As we expected,
the general emotional content of the articles was heavily leaning toward the nega-
tive. In fact, we categorized only 22 articles (out of a total of 260, so less than 10%)
with a positive tone. The rest were more or less equally divided between having a
negative (128) or neutral (110) content. Another way to put this is that negative art-
icles were five times more numerous than positive ones. Not surprisingly, there was
some variation among different coders but, in general, articles that were categor-
ized as negative were not mixed up with positive and vice versa. The coder who cat-
egorized more positive articles (around 18%) also categorized more negative ones
(around 66%), so that the proportion remained similar. One coder classified only
one article as having a positive content out of the 50 she categorized.40

Threats and online misinformation

But why, then, should we have a cultural preference toward negative information
that is reflected in many different domains? Evolutionary biologists and psycholo-
gists have uncovered a fundamental asymmetry between negative and positive.
Rozin and Royzman have put it nicely, though in the specific context of fear of con-
tamination: “Brief contact with a cockroach will usually render a delicious meal
inedible. The inverse phenomenon—​rendering a pile of cockroaches on a platter
edible by contact with one’s favorite food—​is unheard of.” From an evolutionary
point of view, the avoidance of potential threat has a special status. In general, and
this is true for any species, the avoidance of dangers has a greater effect on fitness
than the pursuit of advantages. Serious threats can result in consequences with
heavy influences on an animal’s survival perspective—​think about a broken leg—​
not allowing them to pursue fitness advantages for a while—​until the leg is func-
tioning again—​if not producing the ultimate fitness disadvantage, death. On the
contrary, missing a positive opportunity is less costly: you can try again later or try
something else and, in any case, at least you are still alive.41
Humans, as repeated over and over, are special in their reliance on information
provided by conspecifics. This fact has a couple of interesting consequences that
enhances the asymmetry between negative and positive information, making in-
formation on possible threats more likely to be considered relevant and credible, as
we described above in the experiment where news framed negatively was reputed
more believable than the same information, but framed positively. First, it is better
not to test information concerning possible negative outcomes, but the opposite is
true for information concerning possible positive outcomes. Think about it in this
142 Misinformation

way: if someone tells you that diamonds are below a rock (information about pos-
sible positive outcomes) the logical thing to do is to go and lift the rock to check if
the diamonds are really there, with the consequence of testing the veracity of the
information. If someone tells you that poisonous snakes are below a rock (informa-
tion about possible negative outcomes) you are wise if you do not touch the rock
at all and, as a result, you are not verifying whether the information is true or not.
A good strategy, that of course does not need to be conscious, is to tentatively con-
sider information on possible threats as credible, and instead to test information
on positive outcomes.42
Another angle to look at the same informational asymmetry is to assess the
relative costs of false positives and false negatives. Consider again a piece of infor-
mation on possible benefits, like the presence of diamonds below a rock. A false
positive here is to trust the information when there are no diamonds. A false nega-
tive is instead not to believe the information, when the diamonds are actually
present. The cost of the two errors depends on the extent of the benefit, thus it
plausibly varies from situation to situation. Consider now the piece of information
on possible hazards: there are snakes below the rock. The cost of a false negative
(not believing the information and poking around the rock) again depends on the
extent of the hazards—​and it is potentially very high—​but the cost of false positive
(not poking around the rock) tends always to be minimal. It is for this reason that
devices such as smoke or movement detectors are tuned to err on the side of false
positives. We prefer, up to a point, to switch off manually our fire alarm when toast
burns, rather than risk not being notified of a real fire.43
This points to a more restricted attractive factor: not negative emotions broadly,
but specific negative information, related to threats. While the majority of empir-
ical studies in cultural evolution investigated a general negativity bias, there are a
few suggestions that the success of threat-​related information is independent from
it and could possibly explain the more general tendency. Using again the “trans-
mission chain” method, researchers had participants reading a description of a
new product, such as a new running shoes brand (Lancer™). These descriptions
contained both generally negative information (“Lancer™ special fabric may smell
if not cleaned properly”) and specifically threat-​related information (“Lancer™’s
strap design can cause sprained ankles when used for activities other than run-
ning”). The latter items were transmitted more than the former along the transmis-
sion chains.
In the same study, the authors tackled another puzzling phenomenon. Rumors
on hazards and threats seem often to be scarcely relevant for the people who share
them. “Morgue employee cremated by mistakes while taking a nap”, the fifth most
successful fake news on Facebook in 2017, shared by almost one million people
according to BuzzFeed, does indeed concerns a threat, but a particularly unlikely
one, especially if one does not work in a morgue. To address this, the threat-​related
items of the previous examples were modified to include explicit information on
 Disgust  143

their rarity (“In 2% of users Lancer™’s strap design can cause sprained ankles when
used for activities other than running”). The results were analogous to those of the
previous experiment, showing that participants did not take into account the un-
likelihood of the threat.44
In our fake news sample we found indeed a surprisingly high amount—​almost
30  percent—​of articles describing threats of various nature. Misinformation is
often about killers, kidnappers, bombers, sexual offenders, and the likes, no matter
how credible their antics are and how relevant for the readers the threats can be, as
in the example of the morgue employee cremated during a nap.

Disgust

A negative emotion that has been extensively studied in relation to its effects on
the diffusion of stories is disgust. Stories with particulars that elicit a disgusted
reaction are present everywhere and they seem to enjoy considerable success.
Psychological research shows that disgust is especially provoked by information
about contaminated food, usually by animals (everybody will have heard the dec-
ades old story about the presence of worm meat or rats in McDonald’s hambur-
gers) or body products. Diseases, mutilations, body products in general, and sexual
acts considered “unnatural” are also primary disgust evokers.
Philosopher Shaun Nichols, using the same hypothesis we are exploring here—​
that the differential success of some cultural traits can be explained by their appeal
on universal cognitive preferences—​analyzed the evolution of western etiquette
norms according to the idea that some of them prohibited “core-​disgusting” actions.
Nichols examines an extremely popular and influential book from the sixteenth
century, Erasmus’ On Good Manners for Boys, and consider, within the book, the
norms that are likely to elicit disgust, such as “To repress the need to urinate is in-
jurious to health; but propriety requires it to be done in private”),and the ones that
do not: “If given a napkin, put it over either the left shoulder or the left fore-​arm”. He
shows that the majority of norms that elicit disgust remain to this day an integral
part of contemporary etiquette, but the norms that do not elicit disgust do not, and
they “now seem simply arbitrary or even run against contemporary etiquette.”45
Urban legends, rumors, and children’s stories often have motifs that elicit dis-
gust. Analyzing a sample of 260 urban legends, Joe Stubbersfield and colleagues
found that 13 percent of them contained disgust-​evoking content. In a different
study, a “disgust-​scale” (a quantitative measure of how disgusting is a story) was
a good predictor of online success of urban legends. Transmission chain experi-
ments, with the same logic as the ones described above, also show that disgust is a
powerful factor in determining the success and the survival of stories in a Chinese
Whispers Game-​like setup. Kimmo Eriksson and Julie Coultas presented to parti-
cipants two different versions of the same story. One story, for example, concerned
144 Misinformation

an imaginary character, Jasmine. Jasmine was involved in a charity cake sale and
ended up having her homemade cake featured in a newspaper. In one version, the
mention was due to the “delicious” cake being “sliced up for visiting dignitaries.”
In the other version—​you imagine—​Jasmine found out that the flour she had used
was infested with maggots, with the newspaper reporting about the likely horrified
“dignitaries.” The latter version of the story was transmitted more successfully than
the former.46
The same authors, together with Míchéal de Barra, performed the same trans-
mission chain experiment, but this time they considered participants from the US
and from India. They confirmed that disgust-​evoking stories were more successful
among US participants, but this did not happen with Indian participants. Indian
participants recalled and transmitted better the version of Jasmine’s story in which
there were no maggots in the cake. They suggest thus that there are cross-​cultural
differences in how attractive we found disgusting stories. This is true, but we do
not need to interpret it necessarily as a clue that disgust may not be a general psy-
chological factor of attraction. There was also a difference in how Indians and US
participants rated the “disgust-​scale” of the two versions of the story, with Indians
being less sensitive to the disgusting one. In other words, it could be that it is not
the case that Indian participants were not sensitive to disgust, but that Indian par-
ticipants were simply not disgusted enough by the story to find it memorable and
interesting. As Eriksson and colleagues reason: “high exposure to disgust elicitors
may decrease their power as triggers of cultural transmission, if by habituation
their salience as disgust cues is reduced.” I remember a BBC documentary I saw a
few years ago, where a group of Kombai—​an indigenous population from western
New Guinea—​were apparently laughing amused at the western documentarist to
whom they gave a disgusting-​looking maggot to introduce in the ear in order to
clean it. However, as the documentarist explains, “the joke was on me:” they later
told him that they were not laughing because of what he was doing, but because
they had given him the wrong maggot.47
Misinformation exploits disgust:  we catalogued more than 15  percent of the
news in our sample as aiming explicitly to elicit disgust. Although this proportion
is lower than the one that I presented above for threat-​related information by about
a half, it is interesting to compare this with real news. Intuitively, a sizeable amount
of legitimate news will also be about threats, even though it may be lower than the
30 percent that we found in fake news. After all, threat-​related information is cog-
nitively attractive, so given that one can find events such as homicides, violent rob-
beries, and so on, in real life, they will be likely to be over-​represented in legitimate
news outlets. Disgust, however, is practically absent in mainstream news outlets. In
this respect, while the absolute amount of threat-​related information in fake news
is higher than the absolute amount of disgust-​related information, it is possible
that the relative importance of disgust, when compared with true information, will
be higher.
 Sex  145

Whereas in our sample of fake news disgust information was relatively benign—​
besides my fear of someone accidentally seeing my laptop screen while I was con-
centrated on reading “Plane accidentally empties toilet tank over cruise ship, 23
injured” (I promise to avoid mentioning articles any more sickening)—​the in-
fluence of our intuitive disgust system has been linked to other, more worrying
trends, as the diffusion of anti-​vaccine beliefs.
Anti-​vaccination campaigns did not start with the internet, but they did with
the introduction of vaccinations itself. There are several reasons that conspire to
make vaccination unappealing to our minds. One, likely to be the most important,
is that the effects of vaccines, if all goes well, are invisible. You do not feel better
after being vaccinated in the same way you do after taking a painkiller for your
back ache. Even more worrisome, if enough people are vaccinated, there are no ob-
servable negative effects for the people who do not. That is why, even if they are also
making “big pharma” rich, protests against doctors’ prescriptions of antibiotics are
rarely seen. Vaccines, however, can also tap into our disgust-​related psychology.
After all, a vaccination is nothing else but the insertion of a contaminated external
substance into our bodies. The fact that the quantity is not sufficient to harm us is
not important. Disgust is independent of magnitude. A short contact with a tiny
cockroach can contaminate a full meal, as we mentioned before. This makes sense
as that’s exactly how contamination works in general—​tiny amounts of polluting
substances can and do contaminate our bodies—​but it leaves us not well prepared
to reason about vaccines (and, conversely, about homeopathy).48

Sex

Perhaps surprisingly, cultural evolutionists did not collect data or perform experi-
ments, as far as I know, on the role of sexually-​related information as a psycho-
logical factor of attraction. There are no equivalents of Jasmine’s cake stories with
spicy details. We do not know if narratives with sexual content are remembered
and transmitted better than narratives without; we do not know if sexually-​related
particulars are retained through transmission chains while the neutral ones are
not. As obvious as it may seem, the “sex sells” adagio remains untested in cultural
evolution.
The hints we have from other fields, however, seem to indicate that sexual de-
tails are not important for the success of cultural traits. For example, do erotic
scenes, nudity, and sexual innuendos increase movies’ box office performances?
Apparently not. Each time scientists have looked at the data—​at least regarding
relatively recent movies—​they have found that the presence of graphic sexuality is
not a good predictor of a movie’s success. The website Screen It! provides scores on
various features of movies before they are released, such as “sex/​nudity,” “blood/​
gore,” “smoking,” “tense family scenes,” and similar (the tagline is “Here’s how you
146 Misinformation

can eliminate your worries about the sex, nudity, profanity, violence & more that
your kids are seeing in today’s movies”). Analyzing almost one thousand films
from the early 2000s, and their Screen It! scores researchers found no correlation
between the “sex/​nudity” scores and their box office revenues or their critical ac-
claim. It could also be, however, that this reflects the fact that, in recent years, porn-
ography is widely available, so that contemporary movies do not try, so to speak, to
compete with it. The situation could have been different a few decades ago. In Italy,
for example, a subgenre of comedy characterized by the presence of female nudity
and sexual references thrived amongst the mainstream public in the 70s and early
80s of the last century, and this may plausibly be the case for other countries.49
What about advertisements? Again, the data we have go against the common
intuition -​but they should not be that surprising if you arrived at this point in the
book. Recent meta analyses of several laboratory studies on the role of sex in ad-
vertisements hint at the conclusion that sex does not necessarily sell. It needs to be
said, sex in advertisements does have some effects, and they are in the direction
one would expect looking at it through an evolutionary lens. Advertisements with
erotic content and nudity are more memorable than the advertisements without,
and there is a gender difference in this outcome, with men being more sensitive
than women. Overall, however, a recent meta analyses, taking into consideration
more than 70 studies and accounting for more than 10,000 participants overall, did
not find any effect for factors such as “purchase intention,” and even for “brand re-
call”. Especially if males, we remember that there was sex in the advertisement, but
not what the advertisement was about.50
In the fake news sample we analyzed, sexually related information was present in
a substantial quantity: 45 articles, around 17 percent of the total, were categorized
as containing sexually-​related topics. As for disgust, this seems larger than what
we would expect in real news. It is interesting, however, that the majority of art-
icles concerning sex and disgust were also categorized as containing threat related
information: 25 out of 45 for sex and 27 out of 40 for disgust. Considering this, it
may be that it is not that sex or disgust by themselves are important for the success
of cultural traits, but they are used (again, this does not need to be a conscious
choice) as a vehicle to produce threatening narratives. This gives some support to
the idea, discussed above, that the evolutionary rationale for the general preference
for negativity would be related to the adaptive function of threat-​detection.

Ghosts, Bigfoots, and fishy tomatoes

Imagine someone is going to tell you a story. It is a story about a lizard. The lizard
can be described as (a) a lizard that eat insects off the ground and crawls around
quickly on all four of its feet, or as (b) a lizard that has a long, thin tail and can
never die no matter how old it is, or, finally, as (c) a lizard that always melts in
 Ghosts, Bigfoots, and fishy tomatoes  147

the hot sun, can never die no matter how old it is, and can hear other creatures’
thoughts. Which of the protagonists would you choose for the story? The first one
looks like a pretty boring animal. The third, on the other hand, seems pushing it
too hard, as if the storyteller got carried away. Psychologist Konika Banerjee and
colleagues presented the stories to children aged between 7 and 9 and found that,
as you might expect, they also preferred version (b) of the story (to be precise, they
were not asked which version of the story they preferred, but, similar to many ex-
periments we discussed in this chapter, the researchers measured how many details
of the stories the children remembered). The same happened for a story about a
mailbox: a mailbox that “was covered with rust and was crying because it was sad”
was preferred to a mailbox “that was made of metal and had sharp edges along its
corners” and to one “that was floating in midair, was crying because it was sad, and
ate fire every morning to get energy for the day”. An invisible banana won against
a banana with “bright yellow skin” and against a banana that “felt angry when it
rained, turned invisible every few minutes, and could live in outer space without
needing any oxygen.”51
The hypothesis behind these experiments is that children, like the rest of us,
have some intuitive expectations about how the world works. All living beings,
including lizards, die at some point; non-​living beings, including mailboxes, do
not feel emotions. All physical objects, be they living or not, including bananas,
are solid and we cannot see through them. A mailbox “that was crying because it
was sad” explicitly violates these intuitions. Minimally counterintuitive concepts
are concepts that violate only a few intuitions and confirm, or simply do not in-
volve, others. These concepts are thought to be particularly memorable and atten-
tion catching. Add too many violations and the concepts are not good-​to-​think any
more. Add none and they are the old boring average mailbox.
Pascal Boyer argued that minimally counterintuitive concepts represent a cog-
nitive optimum: they require more attention and they need to be processed with
more details than intuitive concepts, and at the same time we are still able to draw
inferences about how they are likely to behave, which may not happen with con-
cepts that are too counterintuitive. Boyer believes that the success of religions,
supernatural entities in stories, and some forms of superstitious beliefs is linked
to the fact they involve minimally counterintuitive concepts.52 Gods have coun-
terintuitive properties, they may be omniscient and immortal, but they also have
features that make them very human: they are jealous, possessive, they get angry
or they are happy with what the believers do. Santa Claus flies in the Christmas
sky, and has a magic bag that contains presents for all children in the world, but he
needs his sledge to fly, he likes a few cookies (at least in the north of Italy), and chil-
dren need to write him a letter to tell him what their wishes are.
Does the presence of minimally counterintuitive concepts enhance the success
of cultural traits? Ara Norenzayan, Scott Atran, and their collaborators analyzed
the success of Grimm Brothers folktales from this perspective. They used Google
148 Misinformation

hits to determine the popularity of a folktale and produced a final sample of 42

tales: half of them successful, and half of them not successful. The first category
includes the universally known Cinderella (the folktale with the highest number of
hits), Hansel and Gretel, or Rapunzel. In the second there are the likes of Hans My
Hedgehog and The Knapsack, the Hat, and the Horn (ever heard of them? that is the
point). For each folktale, they counted the number of counterintuitive elements
(units such as the tears of Rapunzel restoring the sight of the blind Prince) and they
found that most of the successful folktales had around two or three counterintui-
tive elements, whereas unsuccessful folktales were more or less uniformly distrib-
uted among all possible numbers of counterintuitive elements. Two thirds of the
stories with two or three counterintuitive elements were in the culturally successful
sample, but only around one third of the folktales with less than two, or more than
three, elements were.53
Joe Stubbersfield and Jamie Tehrani tested the hypothesis of the cultural effect-
iveness of minimally counterintuitive concepts in contemporary legends. They first
considered various variants of the story of Bloody Mary, an urban legend in which
a female character who suffered violent death reappears if her name is repeated
three times in front of a mirror. They found that the average number of counter-
intuitive elements present was between two and three, providing some support
for the idea of an optimal number of counterintuitive elements in narratives just
discussed. They also found, however, that there was not a relative survival advan-
tage for counterintuitive elements in respect to intuitive ones. They reconstructed
a plausible phylogenetic tree linking the various versions of Bloody Mary, thus
being able to infer, with some likelihood, which version was derived from which.
They saw that the derived versions were just as likely to conserve both intuitive and
counterintuitive elements from the “parent” version.
In the same analysis of contemporary urban legends we discussed above, they
found that only 6 percent of them contained minimally counterintuitive elements.
This seems surprising, given that ghosts, Bigfoots or aliens seem instinctively to
be the protagonists of many of them. Stubbersfield and colleagues discuss that this
low figure may simply be an artefact of the sampling procedure, as many of the
modern folklore stories that feature minimally counterintuitive elements, such as
stories about ghosts or UFOs, will not be submitted at all to Snopes, which is ultim-
ately a fact-​checking website. (In a sense, this is the opposite problem to what we
had for the “fake-​news-​spreading-​farther, faster, deeper” study that we discussed
above: Snopes does not consider news that is either obviously true or obviously
false.) In any case, as we will see in a moment, a limited presence of minimally
counterintuitive elements in urban legends seems to be consistent with what we see
in online fake news.54
Minimally counterintuitive elements were present in around 13  percent of
the articles we sampled, which is less than the other categories we considered
so far: threat, disgust, and sex. In a way, this figure is again in stark contrast to
 Ghosts, Bigfoots, and fishy tomatoes  149

legitimate news as, at least in theory, they should be completely free from super-
natural components. However, violations of intuitions that could be considered
“supernatural” in the common sense of the term were even less, only around
one third of them (12 out of 33). In addition, these were generally presented as
explicitly satirical pieces (“T.D. Jakes’ Wife Could Be Summoning Demons By
Practicing Yoga”; “Kanye West is dead. Current Kanye is an Illuminati Clone”).
Only a handful of articles were purposely-​misleading news with a genuine super-
natural element (“People Go Missing And Dead Aliens Found After Mysterious
“Flying Object” Invades western States”). Of course, it may also be that, to maintain
some credibility, fake news, as much as contemporary urban legends, need to limit
the amount of supernatural elements.
The class of minimally counterintuitive concepts we found in the majority of
news (21 out of 33) was a specific, and interesting by itself, category, concerning the
violation of ordinary essentialist thinking. These articles are not necessarily about
supernatural events, though some of them may be, but they go against other intui-
tive expectations we have. We are essentialist in respect to living beings. We believe
that living beings, in contrast to artefacts, have a hidden essence which does not
change and which is responsible for their physical appearance and for their be-
havior. For children, a bird that ends up looking like an insect because of an envir-
onment contaminated by toxic waste is still a bird. If one takes a white horse and
paints black stripes on it, there is no zebra around, but simply a horse with painted
black stripes. A lion without a mane and with the same painted black stripes is not
a tiger, even if it looked indistinguishable from one. In contrast, by taking out han-
dles from a mug and putting flowers inside, the mug becomes a flowerpot. There
are not rigid boundaries between chairs, armchairs, couches, and stools.55
An unexpected—​for me at least—​amount of articles among the ones we sam-
pled were about topics such as “experimental” transplants of organs from one
species to another (you can imagine the organs in questions), inter-​species sex,
chirurgical sex changes, and the like. Not surprisingly, some of the articles cat-
egorized as violating our essentialist intuitions were about genetically modified
organisms (“First genetically modified human being is raising concerns for re-
searchers”). It has been suggested that the opposition to genetically modified
organisms can stem, among other things, such as intuitive disgust, or breach of
intuitive teleological thinking that results in the idea that we should not “interfere
with nature”, from a violation of our intuitive essentialist expectations. DNA has
become the material equivalent of the hidden essence that determines the iden-
tity of living beings and regulates their behavior. Painting black stripes on a horse
is fine, but what about inserting a catfish gene inside tomatoes’ DNA? In a survey
from 2004, 85 percent of the US respondents were more or less equally divided
between being unsure, or thinking that it was correct, that transgenic tomatoes
“would probably taste fishy. In sum, transferring genes messes up the essence of
the organisms.56
150 Misinformation

Social interactions and gossip

The transmission chain method has a relatively long history. It was pioneered by
the British psychologist Fredric Bartlett in the 1930s. Bartlett used various forms
of stimuli, such as a Native American folktale, “The war of the Ghost”, but also
pictorial material, such as abstract human faces, that tended to become more and
more schematic and void of details when passing the drawing from one partici-
pant to another along the chain. The modern usage of the method, with an ex-
plicit cultural evolutionary perspective, is however relatively recent. One of the
first experiments was carried out by Alex Mesoudi and colleagues, who examined
the influence that social information had on the retention and the transmission of
stories. The researchers were broadly inspired by the social brain hypothesis which
we explored at the beginning of this book, namely the idea that our big brains—​and
the cognitive abilities associated to them—​evolved to manage social relationships
in primate groups. Following this logic, Mesoudi and colleagues hypothesized that
narratives containing social information would be more likely to capture our atten-
tion and memorable than narratives without, and as a consequence they would be
transmitted more successfully.
The experiment involved four stories. Two were unrelated to social dy-
namics: one story was about global warming, and how forest fires contribute to it
(physical content), another was about Nancy, a student who misses her morning
lecture because of a broken alarm clock (individual content). The other two were
the social ones: in the first, Nancy asks for directions to a swimming pool from an
“old man” at a bus stop, who in turn asks a bus driver who helps Nancy (social con-
tent), while in the second Nancy has an affair with a professor, gets pregnant, and
tells the professor’s wife, who leaves the professor (gossip content).
Both the gossip and the social stories were better remembered than the non-​
social ones. Quite surprisingly, the gossip content (defined as containing “par-
ticularly intense and salient social interactions and relationships”) and the social
content (defined as containing “everyday interactions and relationships”) had
comparable results. In both cases, after four iterations, the stories were constituted
by around five or six sentences, out of the 14 of the original story that seeded the
chain. The non-​social stories instead were reduced to two or even, for the physical
content, one sentence.57
In the investigation of my sample of fake news, we used a different way of coding
the social content, for two reasons. First, the social category as used by Mesoudi and
colleagues seems too broad for the analysis of online misinformation. Practically
all the articles involve at least some form of social interaction or relationship.
Second, the articles contain actual material about celebrities such as pop stars, poli-
ticians, and actors, so that I thought it could be useful to quantify how many news
items were reporting this kind of social information, in respect to news regarding
unknown—​in fact, in the majority of cases, probably imaginary—​individuals. My
Online misinformation from a cultural evolution viewpoint  151

category of social is thus closer to the social-​gossip category above (“particularly

intense and salient social interactions and relationships”), and concerns reports
of infidelity, denigrations, forming and destroying alliances, and so on, and they
need to be at the core of the narrative whether they concern famous or unknown
individuals. I also consider a further category (celebrities) when the news is about
a known person, independently from the actual social content. An article about
Kim Kardashian’s new diet will be coded as celebrities, but not as social; one dis-
cussing the role of Justin Bieber in a never-​happened suicide (“Bieber Accused Of
Being Responsible For Teen Girl’s Suicide After ‘Beached Whale’ Insult”) as both,
and finally, an article such as “Elderly People Riot Nursing Home After field Trip
To Dunkin’ Donuts Was Denied, 29 Injured” (I am not making it up) only as social.
Articles with social information—​salient social information, as defined above—​
were indeed numerous: 129 out of a total of 260, almost exactly half of the full
sample. This result is consistent with the findings of Stubbersfield and colleagues,
who also found that around half of contemporary urban legends contained social
information “at the core of the narrative”, even though they distinguish a gossip
category, possibly more similar to the one we used, which accounted for only 9% of
the legends. The high number of fake news items containing salient social informa-
tion could be partly linked to the presence of a relatively high number of articles on
politics, which, in the majority of cases (around 60% of them) involved social in-
formation: alliances, debates, and the likes. Political fake news was, however, still a
minority in the sample: we coded 105 articles (out of 260) as being explicitly about
politics. This is worth noticing and it fits in well with what we discussed at length
before in the chapter: misinformation is not necessarily, and probably not mainly,
about politics.
The presence of celebrities was also relevant, with one or more famous people
being among the main characters in 125 of the articles coded. Interestingly, how-
ever, the majority of them were political figures. Quite obviously, political articles
need to mention the political figures involved often. The count of news about ce-
lebrities, excluding political articles, drops to 43, that is, around the 16 percent of
all the fake news we considered, a figure comparable with the articles about sex or
eliciting disgust, and lower than the articles containing descriptions of threats.

Online misinformation from a cultural

evolution viewpoint

Let me summarize the main points I tried to make in this long chapter. First, some
of the features of our digitally mediated interactions can, and do, facilitate the
spread of misinformation. Everybody, or almost everybody, is able, with a min-
imum of technical knowledge and with basic equipment, to produce and effectively
spread, at least potentially, information. One can always find someone online who
152 Misinformation

already confirmed almost all possible ideas and many of their combinations, thus
feeling they are not the only ones thinking that the Queen is a cannibal or, more
worryingly, that vaccines are harmful. One can always find someone online who
will believe almost all possible ideas and many of their combinations. One does not
need to disclose oneself online and face the risk of not being believed by anybody
else. All these features are, however, not necessarily negative: quite the contrary,
they facilitate the spread of misinformation as much as they facilitate the spread
of everything else. As we discussed, we also spread plenty of useful and correct in-
formation thanks to these features, so they do not explain, by themselves, why only
misinformation should apparently thrive online.
Another feature that characterizes our online activity is what I called “optimiza-
tion for shallow engagement”. Algorithms magnify the popularity of news and
posts that are liked, clicked or shared, making them more visible to us. While, as we
discussed earlier in the book, the influence due to the perception that something is
popular is easily overrun by many other factors, pure availability makes the news
more presented by the algorithm favored over others. This mechanism may favor
sensationalist and oversimplified information and, in general, information that we
find intuitively appealing. I believe this is a real danger, and I will discuss more
about this and other effects of algorithmic selection later. On the other hand, the
fact that algorithms favor what we like simply begs the question about why we find
some news items more appealing than others.
I also discussed that misinformation, while present and no doubt in some cases
widely successful online, may have been quantitatively overestimated and, in par-
ticular, the real-​world consequences of misinformation, especially political mis-
information, are far from being understood. If anything, from what we know,
political misinformation is not the most shared, and its efficacy may be less strong
than what many think. Perhaps, what has changed is that, as with everything else,
misinformation itself is more available, so it looks more diffuse. But the effect is of
course the same for any kind of information.
An interesting aspect is that, in the sample we considered, quite a few web-
sites more or less explicitly present their articles as satirical news. The author of
all the articles on Real News Right Now claims to be “an internationally acclaimed
independent investigative journalist specializing in international politics, health,
business, science, conflict resolution, history, geography, mathematics, social is-
sues, feminism, space travel, civil rights, human rights, animal rights, fashion,
film, astronomy, classic literature, religion, biology, paranormal activity, the oc-
cult, physics, psychology, and creative writing” and to have received, among other
awards, “three Nobel Peace Prize nominations.” The tagline of the website National
Report is “America’s shittiest independent source,” which did not prevent its articles
being shared and liked on Facebook thousands of times. It could be that these and
other suspect websites turned satirical after being included in the BuzzFeed and
Snopes lists, eight months ago, but it could also be that they were satirical all along
Online misinformation from a cultural evolution viewpoint  153

and people sharing their pages were aware of what they were doing. It could be, in
other words, that the quantitative assessments of diffusion of fake news, which we
already saw being less worrying than what is usually thought, could even be in-
flated by the conscious diffusion of satire.
We should expect that some misinformation would be available online, as
it would anywhere else. After all, misinformation has the obvious advantage, in
contrast to correct information, of not being constrained by reality. We can tailor
misinformation to be appealing, attention-​grabbing and memorable more than we
can with real information. If this is correct, misinformation will have the features
that make all narratives, online and offline, culturally successful. Many of the same
ingredients we analysed in misinformation are also found in urban legend, folk-
lore, and indeed novels and movies. The interest provoked by, say, threat-​related
information works online as well as in a laboratory where a handful of students tell
stories to each other.
In my view, there is not much that makes the spread of misinformation a specif-
ically online phenomenon. This perspective differs starkly from the popular idea
that misinformation is low-​quality information that succeeds in spreading because
of our limited attention: bombarded with too much information we end also taking
in some junk. Quite the opposite, misinformation, or at least some of it, is very
high-​quality information! The difference is that “quality” is not about truthfulness
or depth of analysis, but about how it fits with our cognitive predispositions.58
Of course, misinformation, defined as “factually false claims” as we implicitly
did in this chapter, is a general label that covers various phenomena. Non-​truthful
news can be anything from satire (as we have seen), poor and unintentionally mis-
leading reports, fake news explicitly aimed at political propaganda, and so on. The
motivations to share an article about Pizzagate can be different from the motiv-
ations to share an article about Kim Kardashian or about unknown people involved
in some unconventional sexual behavior. Still, the success of all these cultural traits
is due in large measure to the fact that they are appealing to us and not to some
undesirable feature of the circulation of digital information. We can and should
do our best to fine-​tune the mechanisms that regulate the presentation and the dis-
semination of online news, but we will be more likely to succeed if we ground this
effort in some knowledge of why some of them are attractive and others are not.

Notes

1. Cited in Craik (1916), https://​en.oxforddictionaries.com/​word-​of-​the-​year/​word-​of-​

the-​year-​2016, for classic accounts of rumors pre-​digital age, see, for example, Allport
and Postman (1946) or Morin (1971).
2. http://​hoaxes.org/​archive/​permalink/​the_​great_​moon_​hoax/​
3. https://​www.wired.com/​2017/​02/​veles-​macedonia-​fake-​news/​
154 Misinformation

4. http://​r rresearch.fieldofscience.com/​2010/​12/​arsenic-​associated-​b acteria-​nasas.

html, https://​s cholar.google.co.uk/​s cholar?cites=12247241756468139397&as_​
sdt=2005&sciodt=0,5&hl=en
5. http://​andrewgelman.com/​2016/​09/​21/​what-​has-​happened-​down-​here-​is-​the-​winds-​
have-​changed/​
6. Faulkes (2014)
7. https:// ​ w ww.huffingtonpost.co.uk/​ e ntry/​ oreos-​ m ore-​ a ddictive-​ t han-​ c ocaine_​
n_​4118194, https://​www.theguardian.com/​science/​blog/​2013/​oct/​21/​
oreos-​addictive-​cocaine
8. https:// ​ w ww.conncoll.edu/ ​ n ews/ ​ n ews- ​ a rchive/​ 2 013/​ s tudent-​ f aculty-​ research-​
suggests-​oreos-​can-​be-​compared-​to-​drugs-​of-​abuse-​in-​lab-​rats.html
9. Levy et al. (2013)
10. http://​www.chicagotribune.com/​chi-​oreo-​1-​story.html, http://​www.foxnews.com/​
food-​drink/​2017/​03/​06/​6-​things-​didnt-​know-​about-​oreo-​cookies.html
11. Boyer (2018), https://​www.youtube.com/​watch?v=_​3axPn65MGM, http://​
deardirtyamerica.com/​2012/​06/​queen-​elizabeth-​ii-​confirmed-​cannibal/​, https://​www.
amazon.com/​Donald-​Trump-​Alien-​Prof-​Sanchez-​ebook/​dp/​B01NBWZLQI
12. Ross et al. (1977), Wojcieszak (2008)
13. Betsch et al. (2012), McClure et al. (2017)
14. Data from: https://​data.worldbank.org/​indicator/​SH.IMM.MEAS
15. Yaqub et al. (2014)
16. Boyer (2018)
17. https://​www.snopes.com/​fact-​check/​pizzagate-​conspiracy/​
18. Bernstein et al. (2011)
19. Solove (2007), Haines et al. (2014), Ren et al. (2012)
20. https://​pando.com/​2013/​11/​12/​upworthy-​closing-​in-​on-​50m-​monthly-​uniques-​lists-​
11-​greatest-​hits/​
21. https://​www.snopes.com/​fact-​check/​amnesty-​donald-​trump-​human-​rights/​; https://​
www.snopes.com/​fact-​check/​queen-​elizabeth-​trump-​brooch/​
22. Vosoughi et al. (2018)
23. http://​www.journalism.org/​2018/​01/​29/​sources-​shared-​on-​twitter-​a-​case-​study-​on-​
immigration/​
24. Fletcher et al. (2018)
25. Gaumont et al. (2017)
26. Guess et al. (2018)
27. Alcott and Gentzkow (2017)
28. https://​ w ww.buzzfeednews.com/​ a rticle/​ c raigsilverman/​ t hese-​ a re-​ 5 0-​ o f-​ t he-
​biggest-​fake-​news-​hits-​on-​facebook-​in
29. Morin (2013)
30. Sperber and Hirschfeld (2004)
31. Sperber (1996), Morin (2016)
32. Schüll (2012)
33. Morin (2016)
34. Sperber (1985)
Online misinformation from a cultural evolution viewpoint  155

35. Bebbington et al. (2017)

36. Fessler et al. (2014)
37. Morin and Acerbi (2017)
38. Pridmore et al. (2013), Colman et al. (2014)
39. Niven (2001), Hester and Gibson (2003), van der Meer et al. (2018)
40. Acerbi (2019)
41. Rozin and Royzman (2001)
42. Boyer (2018)
43. Fessler et al. (2014)
44. Blaine and Boyer (2018)
45. Nichols (2002)
46. Stubbersfield et al. (2017), Heat et al. (2001), Eriksson and Coultas (2014)
47. Eriksson et al. (2016), https://​vimeo.com/​6081251
48. Miton and Mercier (2015)
49. Cerridwen and Simonton (2009), https://​en.wikipedia.org/​wiki/​Commedia_​sexy_​
all%27italiana
50. Wirtz et al. (2018)
51. Banerjee et al. (2013)
52. Boyer (1994)
53. Norenzayan et al. (2006)
54. Stubbersfield and Tehrani (2012), Stubbersfield et al. (2017)
55. Keil (1992)
56. Blancke et al. (2015)
57. Mesoudi et al. (2006)
58. For an example of misinformation as “low-​quality” information: Qiu et al. (2017). This
article has been retracted in 2019.
 7
Transmitting and sharing

Memes, of course

In the last chapter of The Selfish Gene, Richard Dawkins turned his attention spe-
cifically to human behavior. He argued that to understand it “we must begin by
throwing out the gene as the sole basis of our ideas on evolution.” Humans, and
their brains in particular, Dawkins famously claimed, are the battlefield of a “new
replicator:” the meme.

Examples of memes are tunes, ideas, catch-​phrases, clothes fashions, ways of

making pots or of building arches. Just as genes propagate themselves in the gene
pool by leaping from body to body via sperms or eggs, so memes propagate them-
selves in the meme pool by leaping from brain to brain via a process which, in the
broad sense, can be called imitation.

The recent history of the concept of meme is quite interesting. Dawkins himself
seems to have had an ambivalent attitude toward it. On the one hand, he has re-
peatedly claimed that it was not his intention to formulate a new theory of culture,
but that he simply wanted to create an analogy to make more tangible the idea of
replicator. Replicators are independent from their specific material substrate, and
genes are only one of the possible instantiations. Anything, given the right condi-
tions, can be a replicator, including “tunes, ideas, catch-​phrases, clothes fashions,
ways of making pots or of building arches.” On the other hand, he has shown—​
understandably—​to be pleased that the idea of meme has been a good meme itself
and has successfully spread in popular and, up to a point, scientific culture.1
A quick review of how memetics spread among researchers interested in human
culture, and especially of the reasons why, in a relatively short period it was dis-
counted by the majority of scientists—​with a few notable exceptions, such as the
philosopher Daniel Dennett—​is useful to introduce the ideas that cultural evolu-
tionists have about the importance of fidelity in cultural transmission, the topic of
this chapter. These ideas are fundamental for understanding how a genuine evolu-
tionary approach to culture can be developed and also to understand how digital
media can have an influence on future cultural evolution. Bear with me, and after-
wards we will move to Grumpy Cat.
After the publication and the exceptional success of The Selfish Gene, discus-
sions about memes began to occupy a niche in academia. In particular, in the

Cultural Evolution in the Digital Age, Alberto Acerbi. Oxford University Press (2020) © Oxford
University Press.
DOI: 10.1093/oso/9780198835943.001.0001
 Memes, of course  157

second half of the 1990s, two books popularized the concept reaching a wide audi-
ence: Aaron Lynch’s Thought Contagion: How Belief Spreads Through Society, and
Richard Brodie’s Virus of the Mind: The New Science of the Meme. Shortly after, in
1997, the first issue of the Journal of Memetics—​Evolutionary Models of Information
Transmission appeared online. In 1999, psychologist Susan Blackmore published
what is probably the most well known account of memetics: The Meme Machine.2
The majority of researchers trying to apply evolutionary approaches to culture,
however, remained unconvinced and were cautious when using the meme analogy.
Today it is uncommon to hear people talking about “memes” in cultural evolution,
and the Journal of Memetics is no longer published (the last issue dates back to
2005). Why is that so? Let’s examine two critical accounts of memetics which are
skeptical in different ways.
One possible criticism of memes is that what we know about the way informa-
tion is stored and transmitted from one person to another does not fit with the pro-
cess of replication as implied by memetics. What are, in a meme, the equivalents of
the genotype (roughly, the DNA) and of the phenotype (its physical expression)?
Imagine you ate a tasty lasagne at a dinner at your friends’, Alice and Bob, and you
want to reproduce the dish. This is a fairly reasonable example of cultural transmis-
sion. Now, following the memetic analogy, the lasagne you observed at Alice and
Bob’s place would be the phenotype of the meme “Alice and Bob’s lasagne”, while
the genotype should be something like the internal representation of the lasagne
that you copied and that you can use now to reproduce a new version of the lasagne.
There are a couple of problems with this description. First, you never had occasion
to actually copy the internal representation of the lasagne so, if transmission hap-
pened, that is, if we agree that your lasagne is a new instance of the cultural trait
“Alice and Bob’s lasagne”, it is not clear what meme actually passed from them to
you. Of course, they could have given you their secret recipe. This is an interesting
case and we will come back to it, but for the current purpose it is important to note
that in many cases of cultural transmission we do not have recipes or anything
similar.
In addition, it is unlikely, from what we know, that your mental representation of
the lasagne will be equivalent to the mental representation of Alice and Bob (and,
naturally, it is highly doubtful that Alice and Bob will have the same exact mental
representation of their lasagne). Thus, even assuming that you copied something,
it is questionable that there is anywhere a “replica” of Alice and Bob’s lasagne.
According to cultural evolutionists such as Robert Boyd and Pete Richerson, that
something analogous to memes will be discovered at the level of the information
stored in the brain is an open empirical question. More important, however, it is
that the feasibility of an evolutionary approach to culture does not depend on the
answer to this question. “We do not understand in detail how culture is stored and
transmitted, so we do not know whether memes are replicators or not. If the appli-
cation of Darwinian thinking to understanding cultural change depended on the
158  Transmitting and sharing

existence of replicators, we would be in trouble. Fortunately, memes do not need to

be close analogues to genes.”3
An evolutionary approach to culture does not need replicators. This may seem
counterintuitive:  how does evolution happen without replicators? Henrich and
Boyd think that, given the action of contextual transmission biases like the ones
we examined earlier in the book, even if transmission is highly inaccurate at the in-
dividual level there is still inheritance at the population level. Here is a description
of one of their models, showing how conformity can have an effect in this respect.
Imagine there are two cultural traits, A and B, and one of them, say A, is more dif-
fuse than the other. They can be any exclusive and discrete choice—​you have one
or the other, but not both—​such as being vegan or not, owning a car or not, or
playing tennis or not. These traits are copied from one generation to the next, but
copying is extremely ineffective. When one individual wants to copy A, they have
20 percent probability of acquiring B instead, and vice versa. In this situation, in a
few generations, the frequencies of A and B will fluctuate randomly, and given a
large enough population they will become more or less equally distributed. This
happens because the copying error penalizes the variant with higher frequency,
simply for the fact that, if there are more As in the population, there will be more
errors transforming As to Bs, than errors transforming Bs to As. In other words: no
inheritance, thus no evolution.
However, imagine now that individuals do not copy others at random, but
choose from whom to copy according to a conformist bias, that is, a dispropor-
tionate tendency to copy the majority. Now, it is still true that the popular trait A is
more likely to be mistakenly copied and become the unpopular B but, crucially,
more people will also try to copy A so that, as long as transmission is not com-
pletely random, A may remain in the majority. The logic of this model is analo-
gous to the one we discussed in the Popularity chapter, showing that conformity
can maintain within-​group homogeneity in the face of migrations between groups.
A  similar reasoning can be done for contextual transmission biases that makes
us more likely to copy particular individuals. The success-​and prestige-​bias we
discussed earlier are an example of those. Again, the possibility of inheritance at
population level is related to the proportion between the magnitude of the error
and the strength of the bias, and Boyd and Richerson and their colleagues believe
that, in most situations, it is a realistic assumption that the latter will be able to
overcome the former.4
This formal and rigorous answer to the “evolution without replication” problem
depends, however, on two assumptions. First, transmission should be faithful
enough to model the process of cultural diffusion as a selection process, that is, a
process in which the individual-​level processes of selection determine the success,
at the population-​level, of the variants. In other words, we still model individuals
as choosing A or B and copying A or B, even if the error may be important. Second,
the strength of contextual biases such as conformity and success or prestige bias
 Cultural attraction  159

should be strong enough to overcome the error in copying. As you may suspect,
I am quite skeptical about the force of contextual biases. What about fidelity of
transmission? Is there any reason to doubt that cultural transmission and evolution
can be thought of as a process of copying at all?

Cultural attraction

The objection of the majority of cultural evolutionists to the claims of memetics is

that we do not know whether there is anything analogous to replicators in many
cases of cultural evolution, but, mainly, that it does not matter. Even without repli-
cation, a selection-​like process at individual level, realized with general transmis-
sion biases that have the effect of favoring a variant, or a group of variants, over the
others, produce a genuine mechanism of inheritance and a genuine evolutionary
process. Whether we call them “memes” or “cultural traits” does not really matter
in the end, but it is better to avoid the term “memes” so as not to suggest a too-​strict
analogy with the biological replicators, the genes. Other cultural evolutionists dis-
agree. Dan Sperber, who we encountered in the previous chapter as one of the ad-
vocates of the importance of general cognitive factors to explain cultural dynamics,
has also developed, together with collaborators, a more general criticism to this
majority view in cultural evolution.
Sperber’s argument is not that cultural transmission is not faithful enough,
but that the similarity between two tokens of the same cultural trait—​Alice and
Bob’s lasagne and my copy of their lasagne—​is not due, in the majority of cases,
to faithful copying but to convergent reconstruction. Let’s first unpack this argu-
ment, and then let’s see why it is important. In his criticism of memetics, Sperber
proposes a thought experiment: imagine a transmission chain, like the ones we
described in the previous chapter, where every person in the chain is passing
a drawing on to others. First they look at it, and then they have to reproduce it.
Imagine this drawing is a scribble of sufficient complexity, like the one on the left
of Figure 7.1. At each passing of the drawing, we expect some error in copying: the
longer the chain, the more distant to the original the copied scribble will be. If you
remember from the last chapter, Bartlett, the inventor of the modern transmission
chain method, did something quite similar, even if instead of scribbles the original
drawings were abstract human faces or stylized animals. In any case, errors and
modifications were indeed quite large.
Although the fidelity of transmission in these chains surely will be very far from
the fidelity of biological replication, one can still accept that, given some condi-
tions, they will produce an evolutionary process. This argumentative strategy is the
one we discussed above: a little bit more fidelity, a good way to select the items to
copy from,) and the cultural evolution machinery will work. However, this is not
what Sperber thinks. Imagine now a similar chain, but where the drawing is not a
160  Transmitting and sharing

Figure 7.1  The scribble and the star in Sperber’s thought experiment.

Reproduced from Sperber, D., “An objection to the memetic approach to culture.” In: Darwinizing
culture: The Status of Memetics as a Science, Robert Aunger (ed.), pp. 165–​166, Figures 8.1 and 8.2 ©
Oxford University Press, 2001.

scribble, but instead, as depicted on the right side of Figure 7.1, a five-​pointed star
of the same complexity (with complexity here being, for example, the total length
of the lines used in the two drawings). What will happen now? Intuitively, the
“copying error” will be lower. However, according to Sperber, we are not copying
at all. Imagine again giving to someone else the original scribble and all the subse-
quent drawings in random order, and asking them to guess the order of the trans-
mission chain, that is, which drawing was the copy of the original, which was the
copy of the second, and so on. This may not be easy, but the task is, in principle,
feasible. As errors accumulate, the scribbles become more and more distant from
the previous ones, and the positions in the chain of two relatively similar scribbles
are likely to be close to each other.
Now, would it be possible to do the same for the star? Would we expect the last
star to be the most dissimilar to the original one? Would we have good reasons to
put two relatively similar stars one after each other in the chain? The answer is no,
according to Sperber, and, if this is not intuitive for you, I invite you to try for your-
self. All the stars will be more or less similar to an ideal version of the star, say with
all the edges the exact same measure and the “right” symmetry among the points,
but there are no reasons to think that one could put them in a chain from closest to
furthest to the original. Or, what would happen if after one month one would ask
the participants to the experiment to redraw the stimulus? It is likely that the par-
ticipants in the “star condition” will do it with no problems, unless some of them
completely forgot what the experiment was about. The participants in the “scribble
condition”, however, would not be able to do it at all.
We are just not copying the stars as we are copying the scribbles. There is a causal
link between each star in the chain, of course, but it is not a process of “copying.”
Five-​pointed stars trigger previous knowledge, and this previous knowledge is
used to reproduce a new version of the stimulus. Errors or unnecessary features of
the drawing will generally be ignored. The point is not that one process (copying
 Cultural attraction  161

the scribble) is low fidelity and one (copying the star) is high fidelity. The point is
that, in this case, fidelity is not a product of a general ability to copy or to imitate,
but it is exactly what we need to explain. If this would be due to a general “capacity
to imitate,” Sperber reasons, there should not be a difference between the two con-
ditions. The existence of memes, or cultural traits that look like replicators, is not
an explanation of human behavior. On the contrary, we need to explain how in
some particular cases—​stars but not scribbles—​this can happen.5
A way to make this clearer is to consider, as philosopher Mathieu Charbonneau
suggests, that cultural evolutionists use the concept of fidelity in two different
ways. The first, that Charbonneau calls episodic fidelity, concerns the similarity
between the traits transmitted. In this sense, when the star that I draw is very
similar to the one that I observed, one can say that there has been faithful trans-
mission. The second—​propensity fidelity—​refers to the tendency of the mech-
anisms involved in the process of cultural transmission to preserve the features
of the trait transmitted. The two concepts of fidelity are very different and being
aware of their usage could clarify debates in cultural evolution. In the star case, for
example, there is episodic fidelity but not propensity fidelity. In what follows I will
use the term “similarity” (or “stability” when this is applied to chains of transmis-
sion, or populations of items) to refer to episodic fidelity, and simply “fidelity” for
propensity fidelity.6
The stars/​scribbles case is not an oddity due to some idiosyncrasies of the
thought experiment, but illustrates a general property of cultural evolution.
Cultural traits have of necessity some stability, otherwise they would not be called
cultural to begin with: they are the five-​pointed stars of the experiment. This sta-
bility is due to the fact that, even if the various cognitive mechanisms that support
the transmission of information are not high fidelity per se (as the imprecise repro-
duction of scribbles demonstrates), some cultural traits, or some of their features,
are more likely to be reproduced than others.
Another common example is the transmission of stories. Think about a famous
folktale, such as Cinderella, and imagine you have never heard it. Now, someone
is telling you the story for the first time, with all the usual details: the wicked step-
mother, the prince, the glass slipper, and so on. If you want to repeat the story later
on, it is highly unlikely that you will repeat it word for word in every passage. Some
details may be even forgotten and become lost: what is the color of Cinderella’s
dress at the Prince’s ball? Which animals were transformed into footmen to escort
Cinderella to the Royal palace? Were there footmen at all? Others, instead, will be
repeated each time by different narrators. They can be fundamental to the plot,
like the glass slipper. They can be particularly memorable and attention catching,
like the pumpkin coach, an example of a minimally counterintuitive concept, as
we discussed in the previous chapter, or the unfair treatment of Cinderella by the
stepsisters and the stepmother, a possibly common situation, as child abuse and, in
extreme cases, infanticide are more likely to occur between genetically unrelated
162  Transmitting and sharing

people, such as stepparents and stepchildren, than between biological parents and
offspring.7
These details work like the five points of the stars. Cinderella is still with us not
because of a general-​purpose mechanism supporting high fidelity transmission,
but because the reconstruction of stories is not random and, through long chains of
cultural transmission, cultural traits or features of cultural traits that are less likely
to be reconstructed will disappear. The differences we observe from one passage
to the other of the chain, being a chain of stars or of the retellings of Cinderella,
in other words, are not random errors, as usually modeled by the majority of cul-
tural evolutionists, but biased transformations, an essential and constructive
part of cultural transmission. The various versions of the cultural trait “story of
Cinderella,” both the actual narrations and our mental representations of it, are all
similar enough, because of biased transformations, to be considered the same type.
Sperber and colleagues call these types attractors, hence the name of the approach.8
Notice that the usage of the term attractor here is similar, but different, from
how we used it in the previous chapter when talking about psychological attraction
and the spread of misinformation online. As we will see shortly, one fundamental
characteristic of digital media is that they support, in various ways, high fidelity
transmission. In many cases—​think about a social media share—​transmission
is virtually replication of a message that hardly requires an active role from the
individuals involved, besides a motivation to share it. Attraction, in the online
misinformation case, refers to the existence of factors that are general enough (re-
member threat-​related information, disgust, and so on) to make the cultural traits
that include them more successful, on average, than cultural traits that do not.

A continuum of fidelity

The point made by Sperber is an important one. Thinking that we are endowed
with a special copying ability and that it is because of this ability that culture exists
is simply a way to avoid recognizing what is interesting about culture. How does
the noisy process of cultural transmission cumulatively give rise to what we call
culture? Why do some cultural traditions survive and others do not? Pascal Boyer
puts it brilliantly when he writes that the fact the processes of interaction and
transmission “could lead to roughly stable representations across large numbers of
people is a wonderful, anti-​entropic process that cries out for explanation.”9
However, this is not the whole story. Consider, as illustration, the extreme ex-
ample of laughing that Sperber uses in the same critique of memetics. Laughing is
socially transmitted, in two senses. First, it is contagious: it is easier to laugh when
others laugh and, in fact, it is also common to laugh simply because others laugh,
even when not knowing the reasons why they do so. Second, laughing is cultur-
ally variable and, within a culture, subject to specific guidelines: there are occasions
 A continuum of fidelity  163

were laughing is allowed and others where it is not, or we expect different members
of the population, for example children and adults, to have different rules about the
appropriateness of laughing. Still, laughing is not copied in any meaningful sense of
the term. We do not replicate the laughter we observe, but a pre-​existent biological
disposition is triggered by the encounters with other examples of laughter and, in
addition, fine-​tuned by the rules we learn during our lifetime. But let’s go back at
the example of Alice and Bob’s lasagne: while we acknowledged that it is difficult to
consider our attempt to cook it again as a case of replication, it would also be odd to
think that “a pre-​existent biological disposition is triggered by the encounters with
other examples of lasagne.” Somehow, some new information has been passed on.10
Even in Sperber’s thought experiment, one could argue that, if we ask the parti-
cipants to do their best to copy the meaningless scribble, they may produce a series
of drawings very similar to each other. Thom Scott-​Phillips realized a real-​life ver-
sion of the thought experiment and reproduced the same results. He showed that
a drawing of the first three letters of the Latin alphabet (ABC, the equivalent of the
five-​pointed stars) were easily reproduced both when participants could place a
new piece of paper on top of the image and trace it, and when they were shown the
image only for a few seconds. The scribble, instead, could be reproduced only when
participants could trace it. This point is important: given the means to increase
the fidelity of transmission, in this case the instruction of placing the new piece of
paper on the top of the original drawing, and the intention to do it, we can, and do,
implement hi-​fi copying.11
If you think about that, this is what around 90 percent of all children in the world
(according to UNESCO) do when they learn how to write. Letters or characters do
not trigger any pre-​existent knowledge, in the same way that the “ABC” stimulus in
the experiment would not be reconstructed by a participant only knowledgeable of
the Han characters of the written Chinese (and, of course, vice versa). Interestingly,
writing systems are not completely arbitrary either, and some of their features re-
flect the properties of our visual system. Olivier Morin, analyzing more than 100
scripts, showed that letters tend to be constituted by horizontal and vertical lines
(with respect to oblique lines) more than what would be expected by chance, that
there is a tendency to separate letters between pure cardinal (containing only hori-
zontal and vertical lines, like E, H, L) and pure oblique (X, W), and, finally, that
there is a prevalence of vertically symmetric letters (A, H, M) with respect to hori-
zontally symmetric ones (C, B, D). All these characteristics fit with the way we pro-
cess visual stimuli: horizontal and vertical lines are more common in both natural
and human-​made environments and they are easier to recognize, and the same
holds for vertical-​symmetrical shapes.12
In sum, a reasonable expectation is that in various cultural domains, or even
in different instances of cultural diffusion, the stability will be obtained by a dif-
ferent degree of triggered reconstruction and faithful copying that has the po-
tential to disseminate, in the individuals who are part of the transmission chains,
164  Transmitting and sharing

new information. When kids learn to write, faithful copying is more important
than triggered reconstruction but, when one knows how to write, it is the other
way around. We mentioned earlier in the book the examples preferred by cultural
evolutionists. Opaque technologies, counterintuitive procedures like manioc pro-
cessing: they all seem to require a good degree of hi-​fi copying. On the other hand,
the oral transmission of a story may be mainly supported by reconstruction and
redundancy: the same attractive story is told again and again, until it sticks.
Here is another way to see it: as hinted above when exposing the complications
of the memetic approach, defining what a cultural trait or a meme exactly is, cre-
ates a quandary. What is the cultural trait “Cinderella”? If we define it as “a story
about a young woman oppressed by her stepmother and her stepsisters who finally
gets to marry a prince,” then most instances of transmission can be legitimately
considered faithful copying, unless not even this basic plot element is retained.
If we define Cinderella as the set of all the sentences we uttered when telling the
story, then it is unlikely they will all be copied faithfully and thus reconstruction is
taking place. Is there a right level of granularity? I think everybody would agree it
depends on what we are interested in explaining. If, for example, the focus of the
analysis is the presence of minimally counterintuitive elements it does not matter
which animals were transformed into footmen to escort Cinderella to the Prince’s
palace, as long as the transformation (i.e. the counterintuitive element) takes
place and is transmitted in further versions of the story. However, if we are folk-​
zoologists interested in the representation of animals in tales it definitely does (by
the way: they are rats).13
In fact, when explaining the diffusion and the stability of cultural traits, both
reconstruction and faithful copying matter. Think about the story of Dracula. Its
success can be linked to a hodgepodge of elements. Some of them fit with the re-
constructive description: vampires, as well as zombies, superheroes, and gods, are
hallmarks of minimally counterintuitive elements, and as such they can be more
memorable and easy to reconstruct. The same goes for Dracula’s thirst of blood,
which may be linked to our disgust and threat-​detection psychology. Legends of
vampires are, not surprisingly according to this view, diffuse in many populations.
Still, the fact that we refer to one particular Transylvanian vampire cannot be sep-
arated from the enormous success of Bram Stoker’s Gothic novel. Some features
that are today considered an integral part of vampire folklore, like the capacity to
shape-​shift into a bat, become so only after being included in the novel and being
faithfully copied in all the subsequent versions.14

Fidelity amplifiers

Humans, after all, are also quite good at copying. One line of evidence comes from
studies in which humans and animals of other species are compared for their social
 Fidelity amplifiers  165

learning abilities. In general non-​human animals can and do learn from each
other, but they do it less and with less accuracy. Claudio Tennie and colleagues,
for example, tested all four species of great apes (chimpanzees, gorillas, bonobos,
and orangutans) and human children at the Leipzig zoo, in a social learning task
involving the observation of a demonstrator opening a box and retrieving a reward
from inside. All participants, both humans and non-​humans, were able to open the
box and access the reward, but the interesting part is how they did it. The box could
be opened either by pushing or pulling it, and the children, unlike apes, also re-
produced the specific action that the demonstrator used. Notice what is important
here it is not that children were able to copy the actions of the demonstrators while
apes concentrated only on the final result—​we now know that apes can, in certain
conditions, pay attention and also reproduce actions—​but that the actions they re-
produced were unnecessary. When a set of actions that brings a result is necessary
or particularly effective, the fact that observers reproduce it does not tell us much
about the fidelity of the transmission process. They might have figured it out by
themselves, or physical constraints may, in any case, make them converge on that
set of actions. We need to focus on behaviors or information that are superfluous,
or that are highly unlikely to be reproduced by individuals alone: in this case, hu-
mans seem to be special.15
The curious human ability, so to say, of copying unnecessary actions has even
got a name: overimitation. This is not just a child’s quirk. In another experiment,
it has been shown, on the contrary, that adults copied the superfluous actions per-
formed by a demonstrator more than children did. Again, picture a box with some-
thing that needs to be collected from a front door. The twist here is that the box
also has a door on the top, as shown in Figure 7.2. Demonstrators inserted a tool in
this door that was completely irrelevant for retrieving the reward (indicated with
the arrow labelled 3 in the figure); in fact, a barrier within the box blocked the tool
from even reaching the area in which the reward was. The only necessary action
was to open the front door by turning it (arrow 4 in the figure) or sliding it (arrow
5). Around half of the actions performed by 5-​year-​old children and more than two
thirds of the actions performed by adults, after watching the demonstration, never-
theless involved the insertion of the tool in the top door.16
Not only this, but, as the explicit instruction of tracing the scribble exemplifies,
faithful transmission can be supported and enhanced by various tools, in a broad
sense, be they actual physical artefacts, such as, say, a transparent paper to facili-
tate the tracing of the scribble, or additional information that accompanies cultural
transmission, such as suggesting the possibility of tracing): we can call these fidelity
amplifiers.
As we hinted earlier, the transmission of Alice and Bob’s lasagne can be greatly
enhanced if they give you their recipe. A recipe is a set of directives that describes
how to prepare a dish, so that it will be most similar to the original. The cultural
evolution of recipes would be an intriguing story to be told by itself. I am not talking
166  Transmitting and sharing

1
3

Figure 7.2  The box and the possible actions in the overimitation experiment.
Reprinted from British Journal of Psychology, 102(1), Nicola McGuigan, Jenny Makinson, and Andrew
Whiten, From over‐imitation to super‐copying: Adults imitate causally irrelevant aspects of tool use
with higher fidelity than young children, pp. 1–​18, doi.org/​10.1348/​000712610X493115 Copyright ©
2011, John Wiley and Sons.

about how ingredients and preparations changed through time, but about how the
descriptions in the recipes did. There are common guidelines about how to write a
good recipe: starting by setting out the number of servings and the time required
for the preparation, listing all the ingredients, writing step-​by-​step instructions for
the preparation in the same order they need to be performed, and so on. As with
all technical idioms, cookery language has developed a series of specific terms (to
sauté, to simmer, to reduce, etc.) that decrease ambiguity and, again, favor preser-
vative transmission.
Recipes, IKEA instructions, tutorials on how to play the guitar are algorithms,
sequences of instructions that, ultimately, could be carried out by a mindless de-
vice, such as a computer. Of course, this is not what happens. A recipe presumes a
great amount of tacit knowledge—​what is a knife? How do you operate it? What is
a gram?—​that needs to be reconstructed, and the time that a machine will read a
regular recipe and prepare the dish for us, even not considering, for example, the
physical complications of operating “hands”, is not in sight. Still, the logic is the
same: a good recipe should provide new information that is not triggered, and be
able to limit its ambiguity.
 Fidelity amplifiers  167

Recipes make explicit that we need to copy actions, and not just the final product,
as in the experiment with children pushing or pulling the box according to what the
demonstrator was doing. This is useful because sometimes it is extremely difficult
to reverse-​engineer a final product. For Alice and Bob’s lasagne, if you already know
how to do lasagne in general, it may not be that hard. An important aspect is that
one can make a series of trial-​and-​error attempts and get closer and closer to the
final result. If the first time it is insipid, you can add more salt. If the final result looks
a little soggy, you can avoid putting in water when preparing the ragu sauce or try to
use less béchamel. But what if you need to invent lasagne from scratch? Good luck.
Not only it is extremely unlikely that you would figure out by yourself the steps
of the preparation: another problem is that for some preparations you do not have
feedback when performing an action similar to the correct one, so trial-​and-​error
learning is practically impossible. There is no need to copy how much salt to put
in the water when preparing pasta (even though we actually do copy it!), but you
can try, taste the final product, and, the next time, put more or less salt according
to how it tasted. The final product gives you feedback that you can use to change
your actions. But, what about, say, tying a Windsor knot? If you do all of the actions
90 percent correct, or nine actions out of ten 100 percent correct and one com-
pletely wrong, in the majority of cases your final product will not be a 90 percent
good Windsor knot. It will likely be some random mess of a tie that, importantly,
you cannot use to improve your actions and tie a 95 percent good Windsor knot.
A way to think about this is that some tasks, like the salt in the pasta water, have
smooth search spaces that can be explored productively by individuals, following
a gradient that brings them to the optimal solution, while others, like tying the
Windsor knot, have flat search spaces, with few optimal peaks. These spaces are
very difficult to explore individually, as there is no gradient until you are very close
to the solution. There are only a few possible ways to tie a correct Windsor knot.
For these tasks, recipes or the copying of actions are essential to succeed.17
Fidelity amplifiers do not necessarily need to be explicit or deliberated. Tasks
that are hard for children to learn, such as tying their shoes or learning the al-
phabet, can be encoded in vivid images and rhymes such as Bunny Ears (“Bunny
ears, Bunny ears, playing by a tree. Criss-​crossed the tree, trying to catch me . . .”)
or the ABC song. Children’s songs such as “Bunny Ears,” in turn, remain stable be-
cause of the usage of rhetorical expedients that favor memorization. David Rubin
analyzed the properties of oral traditions, from counting-​out rhymes to ballads to
epic poems, and proposed that their stability—​especially in the shorter forms, such
as children’s songs—​is enhanced by devices like rhymes (tree/​me, hole/​bold, in the
case of one version of Bunny Ears), alliteration (the repetition of similar conson-
ants at the beginning of words close to each other: the “m” in Eenie meenie miney
mo), and word repetition (“one potato, two potatoes . . .”).18
In addition, different media can enhance or decrease the likelihood of faithful
transmission. If word repetition and rhymes are especially important for stabilizing
168  Transmitting and sharing

stories that are orally transmitted, the same devices may have less importance in
written texts that do not need to be memorized. You can observe the actions of
Alice and Bob while they prepare the lasagne, but they also can tell you their re-
cipe while chatting after dinner, and explain it step-​by-​step. It will also probably be
better if they write it down for you, or if they point you to the cookbook or website
where they got it from initially. As a minimum, in this case the instructions will de-
pend less on your memory and you will be able to retrieve them whenever needed.

Digital transmission

Drawing on the sketch above, it may be useful to distinguish two ways in which
fidelity of cultural transmission can be enhanced. Let’s take again the case of Alice
and Bob’s lasagne. When Alice and Bob give you a written copy of their recipe, in-
stead of just telling you how they made it, they are increasing the probability that
you will be able to retrieve the information they wanted to provide to you. There
is a quantitative gain: more information, if all goes well, is preserved during the
transmission. Another option is to change the attributes of the information, or its
quality: Alice and Bob can write a nice and clear step-​by-​step recipe, or a Joycean
stream-​of-​consciousness of their evening in the kitchen. Both types of information
can be preserved well, but the former will probably be more useful to you if your
goal is to reproduce the cultural trait “Alice and Bob’s lasagne.”
Thinking in this way, digitally mediated cultural transmission provides a tre-
mendous support for fidelity from the quantitative point of view. Digital transmis-
sion has many of the features that make analogic written transmission good in this
respect: it is durable (it is not stored in human minds, but in external artefacts), it
is based on a discrete system that makes copying effortless (we copy the word “to-
mato” as the word “tomato,” no matter what our mental representation of a tomato
is exactly), and it is redundant (errors can be easily recovered, if I write “tonato,”
you—​as well as the software I am using to write—​know what I mean).
However, digital transmission provides other fidelity amplifiers. First, the cost of
copying is even lower than in the case of analogic written transmission. Transcribing
a text by hand is a relatively demanding task, and indeed it is has never been a
common activity, except in specialized contexts, or as an exercise specifically aimed
to improve the act of writing in itself, at school for example. There are possible al-
ternatives. Humans invented endless technologies to reproduce texts, from carbon
paper to modern copying machines and obviously, the printing press. Copying ma-
chines and the printing press make the copying process straightforward, but the
technology necessary to implement them never developed in such a way as to be
accessible to single individuals. Beside very rare occasions, people do not wake up
in the morning, write down their thoughts and then duplicate them with a copy
machine and distribute them among their friends and colleagues. Books, articles in
 Digital transmission  169

newspapers and magazines, and scientific publications enjoyed enormous diffusion

because of printing technology, but the amount of information that could be dupli-
cated has been limited by the fact that the costs of production and distribution have
been relatively high in respect to contemporary digital technologies.
Second, and related to this, when information is both digital and available on-
line there is no need to build several physical copies of the same information, un-
less only as a form of backup. Digital online information provides the possibility
of virtual replication. I do not need to store in my hard disk the scientific articles
I consult while writing this book, as long as I saved the links to access them or,
more and more often, simply remember a few words of the title and Google them.
It is often said that information is a non-​rival good, that is, a good that it is not re-
duced when consumed. If I have a cake and I share it with you, we will have half a
cake each, but if I tell you what time it is, we both have the whole information. This
is true, and important, but when one goes to the nuts and bolts, things are not that
simple. As we just discussed, physical copies of information enhance transmis-
sion fidelity, and physical copies are rival goods. If I give a book to you, I will not
have it any more. If I want to spread my thoughts via a Xerox machine I need to
print many copies of them. If I want to spread my thoughts via social media I just
write them down on my smartphone or laptop and they can be retrieved for free
anywhere.
Of course, they are not completely for free.
Online connections and devices to access the information present online have a
cost, and there are big differences in access in different areas of the world, or in dif-
ferent demographic groups in the same area. In 2016, for example, only 30 percent
of Americans without a high school diploma had a broadband connection at home,
versus 91 percent of college graduates. Digital storage is, as a matter of fact, also a
rival good: the last picture you shared on Facebook is stored somewhere and takes
some physical space that another Facebook picture cannot take. However, when
comparing this to other forms of information storage, it is safe to consider digital
supports (that reduce the space needed to store information) coupled with online
connections (that reduces the need to duplicate it) as a matter of fact, non-​rival.19
Third, copying errors are reduced in digital transmission, when comparing it to
written, analogic, transmission. Technologies such as printing and copying ma-
chines are reasonably precise but few people, as we have just seen, can use them to
diffuse their cultural traits. Individual writing, which may be compared, for diffu-
sion, to digital technologies, supports a more precise transmission than oral com-
munication, but is still very unfaithful in respect to digital reproduction. Scholars
studying ancient manuscript transcriptions have developed a lexicon to describe
the categories of errors that scribes made:  there is “homoteleuton” (the scribe
paused and, when resuming, skipped a passage because of the similarity of two
ending lines), “homeoarchy” (the same, but because of the similarity of the begin-
ning of two lines), “dittography” (repeating the same word twice) and many others.
170  Transmitting and sharing

Some research analyzed the work of a “skilled female typist” in the 1960s and
1970s, in order to identify errors in typewriting. “To amass a database of 3,000
errors”—​the study reads—​“it was necessary to scan an estimated 1,300,000 key-
strokes. Her error rate in this context was therefore 0.23 %.” This looks like a low
error, but it still amounts to “about three keystroke errors per typical manuscript
page.” An error rate of 0.23 percent (that is, one character out of around 450) is in
fact in the low end of average error rates of expert typewriters. Online sharing, on
the other hand, is virtually error-​free as much as copying-​and-​pasting portion of
texts -​even though, as we will see soon, errors are not absent in the latter.20
Digitally mediated cultural transmission also provides, at least potentially, im-
proved fidelity from the qualitative point of view. It has been suggested that one
of the shortcomings of analogue, asynchronous, written transmission is the ab-
sence of mechanisms of repair. In an everyday conversation, we frequently inter-
rupt each other requiring clarifications, more information, additional context, and
so on. This is done with short, often one-​word, sentences (“What?” “Who?” re-
peating a word we are not sure we understood with an interrogative tone, etc.).
These kinds of interactions make transcriptions of oral conversations look strange
when we read them, but they go practically unnoticed when we talk, and they are
generally effective, in the sense that the other speaker gives the correct additional
information requested. Linguist Mark Dingemanse and colleagues recorded hours
of conversations in 12 languages, taken from ordinary social interactions such as
preparing foods or playing games, and they found that these “repair” requests were
similar in all languages, and extremely common: on average once every 1.4 min-
utes across all languages.21
These are not signs that oral communication does not work smoothly, quite the
contrary: it has been proposed that this may explain the relative difficulty of evo-
lution of graphic codes for asynchronous communication (relative that is, with
respect to the evolution of codes for synchronous communication that emerge
everywhere there are humans). In other words, natural languages are present in all
societies, but fully functional written codes are more rare, also because, as Socrates
lamented at the beginning of this book, “if you question them, they always say one
and the same thing.” As Morin and colleagues reflect:

Even when graphic messages can be exchanged rather rapidly between different
places, technological constraints (until a few decades ago) implied that repair had
to be restricted and considerably delayed. Furthermore, other-​initiated repair is
simply impossible when receiving messages from the dead (a crucial aspect of
cultural transmission).22

(Just to be sure, they are not thinking about Ouija boards: the “dead” are the au-
thors of, for example, canonical books. There is not much point of saying “Huh?”
to Plato.)
 Internet memes  171

It does not seem too far-​fetched to consider digital transmission, online digital
transmission, as a form of written communication that allows repair. In the case of
online chats this seems obvious. Some repair mechanisms may be different from
the mechanisms we use in oral communications (for example, just writing a ques-
tion mark, or a puzzled face emoticon, after a sentence we do not understand),
while other more similar (I imagine “Huh?” is as successful online as offline). The
analogy can be less immediate for other forms of online communication such as
social media posts, but they also provide explicit mechanisms -​think about an-
swers or comments -​that may used in the same way.23
In addition, transmission based on digital technologies can be enhanced by
the fact that including non-​written content, such as images and video, is easy and,
again, practically costless. While images are somewhat common in books, videos
were not possible to include: video tutorials are a digital-​age phenomenon. There is
not much research in cognitive anthropology and cultural evolution on how videos
increase the fidelity of cultural transmission, but it is not too far-​fetched to specu-
late that, at least for certain domains, the effect may be relevant. Think about skills
that require visual feedback, such as how to apply make-​up or how to process your
pictures digitally. Think about skills that require complex action sequences, such
as the already-​mentioned necktie tying (I myself look on YouTube on the rare oc-
casions I need to wear one), playing a musical instrument, realizing a complicated
recipe, dancing, and many others. For most of these skills, written transmission
would be ineffective—​or, if you prefer, the reconstructive part of the transmis-
sion would need to be much higher—​and the only possible alternative would be
to have the chance to be in the presence of a real-​life demonstrator. A YouTube
channel on “Primitive Technology” has, as I write, almost nine million subscribers.
The description states that that the videos teach how to “build things in the wild
completely from scratch using no modern tools or materials. These are the strict
rules: If you want a fire, use a fire stick—​An axe, pick up a stone and shape it—​
A hut, build one from trees, mud, rocks etc.” Whereas the proportion of the nine
million subscribers who actually tried to follow the tutorials and build a “round
hut” or a “natural draft furnace” is no doubt very low, it would likely be more diffi-
cult, way more difficult, to learn to do something like that with a written transmis-
sion. Amusingly, the maker of the videos, when asked how he learned his skills,
answered that it was by looking up “anything that interests me online” where he
“can usually find information”. In sum, if you want to learn primitive technology,
the web is the answer.24

Internet memes

We can summarize what we have said so far by characterizing more clearly the
“continuum of fidelity” image. The two extremes of the continuum are two extreme
172  Transmitting and sharing

views of human culture. On one side, let’s call it the “laughing” side, what we call
culture is the expression of a previously present repertoire that is triggered by ex-
ternal conditions. Evolutionary psychologists John Tooby and Leda Cosmides have
famously depicted it using an analogy involving jukeboxes. They ask us to imagine
a population of identical jukeboxes, all provided with plenty of records, the same
for each jukebox. These jukeboxes are spread in various parts of the world—​some
of them in Italy, others in Brazil, and so on—​and they are activated by the geo-
graphical coordinates that their GPS registers (yes, they have one). Different co-
ordinates make the jukeboxes play different records: all the jukeboxes set in Rome
will play the same song, while the jukeboxes set in Rio de Janeiro will play another
one, the same amongst them, but different in respect to the one played in Rome.
In addition: if a jukebox is moved, for any reason, from Rome to Rio de Janeiro,
its GPS will record the new coordinates, and it will start to play the new song, con-
forming, so to speak, to the local population. More gadgets can be added to the
jukeboxes: for example, a sensor for the temperature (in this case, Rio’s jukebox will
be likely to have a song for the summer and one for the winter) or even a micro-
phone that detects a few notes of the songs present in the environment, and then
chooses the corresponding record (the app Shazam does exactly this—​that would
be the laughing analogy). In all cases, there is nothing new under the sun.25
On the other side of the continuum, let’s call it the “Xerox machine” side, cul-
ture is passed from a person to the other as an independent abstract entity, and
we just copy what we receive. Of course, sometimes there are errors: spots on the
page make some part of the message illegible; foreign substances such as dust on
the scanner glass produce unwanted streaks and lines. These errors are passed on
in the next generations of copies—​Xerox machines do not adjust the copies—​and
they tend, on average, to appear at random and to deteriorate the messages. We will
be able to spot patterns in the errors: some machines are indeed better than others,
and we can categorize the different kinds of malfunction and when they are more
likely to appear. Overall, however, what we want is for our machine to work well,
and just copy whatever we give to it.
Human culture is nowhere in the two extremes, but everywhere between them.
Thinking about culture in this way may be useful. Different domains can be placed
in different levels of the continuum, as well as different aspects of the same cul-
tural phenomenon (remember Dracula). Moreover, different supports of cultural
transmission allows moving in this continuum and, as we mentioned, it is worth
reflecting on whether the diffusion of digital media can be considered as producing
a push towards the Xerox extreme, and asking what the consequences might be.
In this sense, the semantic shift that the term meme has undergone in popular
culture seems quite appropriate. When talking about memes, the majority of
people will not think about the original meaning Dawkins gave to the term more
than 40  years ago, but we will immediately mention Grumpy Cat or Distracted
Boyfriend. Today, a meme is a peculiar internet phenomenon, and the term
 Internet memes  173

identifies a content that spread with success online. This usage started in the 1990s,
and it was made popular by the same Mike Godwin we encounter earlier in rela-
tion to his “law” about online discussions (as they grow longer, “the probability of a
comparison involving Hitler approaches one”). In a 1994 Wired article he called the
Nazi-​comparison an “out of hand meme” and his “law” a “counter-​meme.”26
Memes—​internet memes—​are indeed characterized by high fidelity, since their
transmission is supported by the digital infrastructure. However, perhaps sur-
prisingly, they are more changeable than what the level of fidelity allowed by their
channel of transmission would suggest. You may have seen one, often called the
Equality versus equity meme. In its more recognizable variant, it is composed of
two drawings. In both drawings there are three individuals of different heights—​an
adult, a young person, and a child—​trying to watch a baseball match from behind
a fence (see Figure 7.3). In the drawing on the left, each of them is standing on a
same-​sized wooden box: as a result, only the adult and the young person can see
the match above the fence. This is equality. In the drawing on the right, the same
three boxes are redistributed: the adult is tall enough to see the match without his
box, which is now given to the child. With two boxes, he is higher than the fence
and he can see the match too. The situation is unchanged for the young person,
who needs his box. Everybody, using the same number of boxes as in the previous
instance, can see the match. This is equity.
This meme was invented in 2012, and its author tracked some of the variants that
he could find on the web. As he writes, “my original graphic was being adapted,

Figure 7.3 The equality versus equity meme.

© Craig Froehle, 2012.
174  Transmitting and sharing

modified, and repurposed in a mind-​blowing variety of ways, and then shared and
redistributed all over the place.” First, the drawing was revised in several aspects.
The original image, shared on Google +, was a Microsoft PowerPoint collage of a
“public photo of Cincinnati’s Great American Ball Park, a stock photo of a crate,
clip art of a fence” and three individuals who look like they were created using the
PowerPoint available basic “shapes.” Subsequent versions were, for example, fully
hand-​drawn or realized with a professional-​looking design. One version replaces
the baseball match with a cricket match, another with a view of Utah’s Delicate
Arch. Other versions change the gender of the protagonists: sometime the young
person is a girl, sometime they are all females. In other variants, the goal is not to
see a match behind a fence any more, but, for example, to reach the fruits hanging
from a tree, to ring some bells placed on the top of the images, to reach books on a
high shelf, or to write on a blackboard, you name it. Several variants, while keeping
the same logic of the more common ones, started to have, plausibly without the
original change having been intended to be meaningful, more boxes on the eq-
uity side, giving the impression that equity may be more resource-​intensive than
equality. In fact, the original version was not even concerned with “equality versus
equity,” which became what the meme is about, but on “equality to a conservative
versus equality to a liberal.” Other variants label the two images “equality versus
justice,” “equality versus fairness,” “sameness versus fairness,” and so on.
And, of course, more radical modifications and new takes are very common.
Satirical versions of the meme depict “reality” as a third condition, in which even
more boxes are given to the adult, or another figure (a politician) takes away all
the boxes and leaves with them. There is no shortage of versions of “The problem
with that equity vs. equality graphic you’re using” or similar, that shows that the
“real” solution is to give more boxes to everybody (instead of redistributing them),
to replace the wooden fence with a transparent one, or, indeed, just get rid of the
fence (removing the “systemic barrier”). On the other hand, the author of the ori-
ginal image notes, “there was no shortage of complaints that the kids were just free-
loaders and should buy a ticket to be inside the stadium if they want to watch the
game.”27
This is a nice, and relatively well documented, case, but my point is that if one
would take the time to search for the history and of the variants of their favorite
meme they would find, in the majority of cases, adaptations, modifications, false
starts, unsuccessful types, prolific versions, and dead ends. What is exciting is that
we can, given that all these trickeries happen in digital media, track and quantify
them and, hopefully, understand something more about the process. A group of
researchers lead by Lada Adamic analyzed thousands of memes that spread on
Facebook over a period of 18 months, from April 2009 to October 2011. They were
able to gather the impressive number of around 460 million single instances of
sharing. At the time, Facebook did not have a built-​in share button, so that the
pieces of information had to be copied and pasted manually by the users of the
 Internet memes  175

social network, if not re-​typed word by word. In this perspective, there is not a neat
separation between “memes” and “not-​memes:” two strings of text are the same
meme if they are similar below an arbitrary threshold (we will see examples in a
moment) and a meme counts as such if it is diffused enough according to a speci-
fied threshold. How many people need to copy and paste the same or similar status
update for it to be considered a meme is, as above, an arbitrary decision. The re-
searchers decided to define a meme as any cluster of similar status updates that had
at least one variant with 100 copies in the period considered. Notice how this def-
inition fits well with the logic of cultural attraction theory we discussed before, for
which a cultural trait is a cluster of items that are similar enough to each other to be
considered the same type. The same happens here, even though the digital medium
of transmission would provide fast and precise replication at basically no cost.
One of the meme clusters they considered had as a most popular variant (copied
exactly in this form more than 470,000 times): “No one should die because they
cannot afford health care and no one should go broke because they get sick. If you
agree please post this as your status for the rest of the day.” Successful variants had
modifications such as starting with “thinks that” (which will follow the name of the
user in the status update, so reading like “John Doe thinks that no one should . . .”)
or the addition of the short sentence “We are only strong as the weakest among
us” between the two main sentences. Other variants altered the message, but they
were still considered members of the meme cluster. There were ironic mutations
(“no one should be without a beer because they cannot afford one;” “no one should
be frozen in carbonite because they couldn’t pay Jabba the Hutt”) and there were
variants that appealed to the opposite political spectrum (“no one should go broke
because government taxes and spends . . .”).
Overall, the “mutation rate” was surprisingly high: on average, 11 out of 100
copies were different from the original form which they were likely to be copied.
Think about it in these terms: you copy and paste a sentence from one place to an-
other and 11 times out of 100 you end up with a different, albeit only slightly, one.
Over all the memes that were considered in the research, interestingly, mutations
were not random. While it is likely that the first person introducing Jabba the Hutt
was aware of what they were doing, many changes were minimal but, as the results
of the psychological preferences we considered in the previous chapter, consistent
in their direction. Within the same meme cluster, shorter variants tended to be
more popular. A shorter text is more likely to be copied with fewer errors—​this
is obvious if you have to remember it by heart, but apparently it works the same if
you have to swipe your mouse—​and, provided it can convey the same message, it
will be favored. Variants that were too short, indeed, were less popular. In addition,
mutations were more likely to arise at the beginning or at the end of the sentence.
The central sentence of the most popular variant was preserved, on average, 7 times
out of 10, but the first and the last sentence were preserved 6 and 5 times out of 10,
respectively. Again, this may be due to the simple fact that when selecting a text to
176  Transmitting and sharing

copy and paste, it is more likely to miss words at the beginning or at the end of the
selection.28
Perhaps these effects are not that unexpected, but what is odd is that they result
in a digital media, where, as we discussed, high fidelity copy is trivial to imple-
ment. Of course, the mutation rate is due to the fact that the share function was
not implemented yet. It is, however, interesting that even today some Facebook
memes spread with an explicit instruction to be manually copied and pasted, even
though they could simply be automatically shared. This convention could be due
to different factors. The main reason may be that shared statuses have a direct
link with the original. If the original status is, for whatever reason, reported by
other users and as a consequence cancelled by Facebook, all the shares will follow
the same path. Copied and pasted versions, on the other side, cannot be tracked
back automatically and they can survive independently. Another reason, however,
may be closer to what we just discussed. A copied and pasted meme can accumu-
late mutations, whether intentional or not, and these mutations can enhance its
success.
The same happens when sentences are copied from an article, or a blog, to
another one. Sébastien Lerique and Camille Roth used a dataset of hundreds of
thousands of quotations from the internet—​where a quotation is simply a series
of words delimited by quotation marks, such as “The Bank of England said, ‘these
operations are designed to address funding pressures over quarter-​end’ ”—​and
they looked at the substitutions of words within them. The quotation above, for
example, is also reported as “these operations are intended to address funding pres-
sures over quarter-​end.”
Lerique and Roth do not report the frequency of changes, but they show that
these changes are not random. High-​frequency words, which are easier to recall,
are replaced half the number of times that would be expected if words were re-
placed at random, and the trend is the opposite for low-​frequency words, which are
replaced twice as many times. For the same reasons, words of many letters are re-
placed more than short words, and words that are acquired, on average, at an early
age, are substituted less than words that are learned later. Replacements are not
random either: they tend to be words that are more common, shorter, and learned
at an earlier age. Here is their example: in 2008 the Burmese poet Saw Wai was ar-
rested because he diffused an eight-​line poem in which the first word of each line
spelled-​out the sentence “Senior general Than Shwe is foolish with power.” Other
websites replaced “foolish” with “crazy,” a word that is more common (“crazy” ap-
pears around 4,100 times in their dataset, “foolish” appears 675 times), shorter (5
versus 7 letters) and learned at earlier age (5.22 years old versus 8.94 years old).29
More generally, researchers studying internet memes have emphasized that
defining them as stable, unchangeable, pieces of information that proliferate in
the web misses the point. Think, for example, of the widespread meme known as
Downfall or Hitler Reacts. The meme is based on a scene of the acclaimed 2004
 Internet memes  177

movie Downfall, depicting the last days of Adolf Hitler and his suicide in his Berlin
bunker. In the acme scene (which may have become the acme scene also because of
the meme), Hitler is notified by his generals that the last remaining active military
unit is too weak to defend Berlin. As a consequence, there are no more obstacles
between the Red Army and the German capital. Hitler orders everybody out and
he remains in the office with the top generals only. With the rest of the staff in the
corridor eavesdropping behind the door, Hitler starts a three-​minutes rant against
the generals, while at the same time understanding that the war is lost. The scene is
emotional, and the interpretation of Bruno Ganz, the actor playing Hitler, is unani-
mously applauded.
In 2006, a YouTube user posted the scene with the original audio, but subtitled
in Spanish. Hitler’s breakdown, according to the subtitles, was, however, about the
absence of new features in the demo trial of Microsoft Flight Simulator X. Soon,
the English translation appeared. In a few months, and in the next years, thou-
sands of variants of Hitler Reacts flooded YouTube. A dedicated YouTube channel
Hitler Rants Parodies has around 1,700 different videos as I write—​Hitler phones
an Indian call center being the most viewed with more than two million views.
The topics are endless: Hitler is angry because some character in a TV series died,
Hitler is angry because France won the World Cup; Hitler is angry, in a turned-​
meta parody, because there are too many Hitler Reacts takes. Many international
videos are present, including one subtitled in Hebrew, in which Hitler is mad about
the chronic lack of parking spaces in Tel Aviv, which apparently generated con-
cerns when it appeared in 2009. The most famous version of the meme, Hitler gets
banned from Xbox Live, became itself the source of derived variants, all different
from one another.
Limor Shifman, a meme researcher, called this kind of diffusion “remix.” In
this case, a meme is a series of instances of similar material, resulting from the
digital editing of a pre-​existent template. Everybody can make a HitlerReacts video,
adding their own subtitles to the Downfall scene. There is no need to be familiar
with video editing: a website allows you to build it directly in your internet browser,
with the subtitles at the right time of the scene.30 A  myriad of internet memes
follow the same logic. Grumpy Cat (dear reader, we made it!) is probably one of the
most prominent LOLcats—​what the internet is really about, according to many.
LOLcats are mainly images macro, pictures or artworks with a caption that can be
edited by everybody. Grumpy Cat image macros feature a real cat that, because of
a feline dwarfism condition, looks particularly grouchy. Captions like “I had fun
once—​It was awful” or simply “Nope” are added to the figure. As before, the point
is that everybody can produce their own Grumpy Cat meme. The meme is a con-
stellation of cultural traits, which include the various versions of the image macros,
but also, in this specific case, different image macros (some of them with other
cats), videos of the cat, one book (at least) and merchandising going from mugs to
posters and t-​shirts.
178  Transmitting and sharing

A similar dynamic, called “mimicry” by Shifman, is again based on the idea that
a meme is a template for creating different variants. Whereas remix requires a rela-
tively low effort from the authors that digitally elaborate already present content,
mimicry involves the actual creation of a new content, inspired by a source, so it is
not surprisingly less diffuse. Examples of mimicry are the parodies of Gangnam
Style or Charlie bit my finger (both famous memes in their own right). In the first
two weeks of February 2013, around 40,000 videos similar to each other were
uploaded on YouTube. The videos usually showed a single person dancing, sur-
rounded by a group of people apparently unaware of the dancer. The music features
a sample of a voice shouting “con los terroristas” (“with the terrorists” in Spanish)
and a synth snare drum. After 15 seconds the bass hits, the video cuts, and every-
body in the scene, or other people, often with odd outfits or in actual disguise, start
to dance too, performing bizarre movements. In 30 seconds it is all finished. You
may have recognized the Harlem Shake meme. Variants are countless.
Unlike the imitations of Gangnam Style or Charlie bit my finger, there is not an
original version that is parodied. The song Harlem Shake was released a year be-
fore the videos started to appear, but it went by and large unnoticed. And, in any
case, besides providing the first 30 seconds of music, there is not much else: the
song itself was not accompanied by a video. The script of the meme was estab-
lished by the initial versions of the video, realized first by a not-​widely-​known
Japanese-​Australian entertainer, George Miller, also known as “Joji,” and then by
five unknown Australian teenagers. These early versions are not, however, the most
famous and their temporal priority does not give them any special status in re-
spect to others. Again, a meme is a cluster of cultural traits, associated, more or less
tightly, to a common template.31
In addition, although to a different degree, memes do not live only a separate life
in the digital world, free from the imprecisions that dampen faithful transmission
in the analogue realm. In the autumn of 2016, a clown craze spread for a couple of
months, first in the US and then in the rest of the world. Sightings of “evil clowns”
in the real world were accompanied, not surprisingly, by circulation of suspicious
pictures and news in social media, which in turn produced more sightings of real
world clowns. Several accidents, including a murder of a teenager in Pennsylvania
were connected to killer clowns, but all claims were deemed unsubstantiated. The
murder was real but, apparently, it was not the murderer, but the victim that, for
unrelated reasons, was wearing a clown mask. The situation seemed tense enough
that in October McDonald’s company decided that Ronald McDonald, the clown
mascot of the fast-​food chain, would be better to stay away from public attention
for a while, and the German Interior Ministry announced a “zero-​tolerance policy”
against scary clown costumes and masks just before Halloween.
Memes, such as the clown craze, have their real (or allegedly real) counterpart
in the world offline, and real (or allegedly real) events that happen online, such as
clown sightings, become digital memes. It is even more difficult, in cases like these,
 Online sharing  179

to define what exactly is the cultural trait. If Grumpy Cat is the constellation of all
image macros of grumpy cat and Harlem Shake is the collection of all videos with
some specified features (plus, plausibly, the actual performances that were filmed),
the scary clown craze is a set of information spreading on social media, newspaper
articles, police communicates, events that may have happened in several different
countries, and so forth.
Incidentally, the scary clown craze also exemplifies many of the themes that we
have discussed in previous chapters. It went “viral” not because it was pushed by
personalities or because of an orchestrated marketing campaign (even though it
has been conjectured, with no evidence, that the all the fuss may have been gener-
ated as a promotional act for the forthcoming film’s adaptation of It, a celebrated
novel by the horror writer Stephen King that features a clown as a main villain). It
went viral because digital media provided a channel for quasi-​universal distribu-
tion and because enough people were susceptible to it, and enough were willing to
contribute with their own material. It went viral because it has plenty of the ingre-
dients that we described in the previous chapter: negative emotions, threat-​related
information, minimally counterintuitive concepts: clowns are like us, but they also
behave in an unpredictable and out-​of-​the-​ordinary manner. Clowns are even
super-​stimuli to our natural disposition to find faces attention-​catching and mem-
orable, with their make-​up that exaggerates facial features.

Online sharing

In the previous section I described several examples of internet memes that show
that it would be naïve to think that digital diffusion is always a matter of replica-
tion of information. Still, online cultural transmission can be error-​free at prac-
tically no cost, and, in many cases, this is what happens. Whereas Gangnam Style
generated thousands of parodies, the original video itself was the first video to
hit 1 billion (one with nine zeroes) views on YouTube in December 2012, and the
count has surpassed 3 billion views as I write. Gangnam Style is not an isolated
phenomenon:  in 2018, all the 100 most-​viewed videos on YouTube generated
more than 1 billion views. On a smaller scale, the fake news hits we encountered
in the previous chapter were shared, in many cases, more than 1  million times
each—​and I am talking about “FBI seizes over 3,000 penises during raid at morgue
employee’s home.”
A social media share is a paradigmatic example of error-​and cost-​free transmis-
sion. How does this fit with the experiments carried out by cultural evolutionists?
As we discussed at length, transmission chain experiments are basically controlled
versions of the Chinese Whispers Game: a participant listens to a story (or reads
it) and they have to remember it and repeat it to the next one. The focus is on how
stories are distorted and on which details are remembered and which are instead
180  Transmitting and sharing

forgotten. But, when one shares a piece of content on Facebook, and the same con-
tent is re-​shared again and again, what is happening in reality is a long transmission
chain where there is no distortion and no decay. Figure 7.4 depicts a schematic rep-
resentation of a chain simulating online transmission (compare it with the figure
of the transmission chain in the previous chapter). A hypothetical participant has
access to different pieces of content, selects them, and they are automatically trans-
mitted, error-​free, to the next participant. What are the differences? Up to a point,
it can be reasonable assuming that things work more or less in the same way. In the
previous chapter we saw that the same psychological factors of attraction that bias
transmission chains—​threat, disgust, minimally counterintuitive concepts—​are
also present in many suspect online news items. Ideally however, one would like
to compare situations in which the same information can be passed on in a typical
transmission chain set-​up and in an arrangement more similar to what happens
online, where people do not have to remember and repeat -​and, in fact, do not
even have to read the actual content.
Promising steps in this direction have recently been made. In one of the experi-
ments about disgust I described in the previous chapter (the one with the story of
Jasmine and her maggot-​infested charity cake), Kimmo Eriksson and Julie Coultas
explicitly consider three separate phases in cultural transmission. They discuss
how transmission chains usually take into account only one of these phases, which
they call encode-​and-​retrieve. Participants listen to a story and then have to repeat
it. In reality, we can decide whether to listen to a story or not (in the majority of
cases, at least), and we can decide whether to repeat it to someone else. They call
these two additional phases choose-​to-​retrieve and choose-​to-​transmit.
It does not look too far-​fetched to consider online sharing as cultural transmis-
sion stripped, at least potentially, of the encode-​and-​retrieve phase. Some stories
capture our attention and we can choose whether to share them or not, and if we
do, it happens automatically and error-​free. In this sense, online cultural trans-
mission looks nothing like experimental transmission chains. In one of their tests,

. . .
D F H
C E G
B D E
A A D
Selection Selection
Reception Reception

Participant 1 Participant 2

Figure 7.4  A schematic representation of a transmission chain experiment simulating

online transmission.
 Online sharing  181

Eriksson and Coultas did exactly this: participants were presented with headlines
summarizing the main events of a series of stories, for example “Jasmine sold a
cake that was so delicious she would rather have kept it for herself ” and “Jasmine
sold a cake that by accident had maggots in it.” Participants could choose whether
to click or not on the headline and, if they did, they could read the entire story and
then decide whether they “found the story worthy of being passed along to other
participants.” The results were equivalent to the classic transmission chain set-​up.
Starting from a set of eight stories, four containing elements eliciting disgust and
four not, given to 40 individuals per generation, the stories without disgusting de-
tails became extinct in only two steps of transmission. As we already know, stories
where elements eliciting disgust were present were more successful than stories
where the same elements were omitted.32
In this case, the absence of the encode-​and-​retrieve phase, which I suggested is
potentially missing in social media transmission, did not make any difference. Are
there reasons to believe that it could? Transmission chain experiments focus on
cognitive aspects such as comprehension, memory, and verbal or written repro-
ductions. These aspects are, no doubt, important for cultural transmission, but they
are only part of the reasons why a cultural trait is successful. Something can be very
memorable, and perhaps easy to transmit, but we may simply not be motivated to
pay attention to it, or to transmit it to others. The sequence “AAAAAAAAAA,” pro-
vided with the information “a succession of 10 As,” is very easy to memorize and
very easy to reproduce and transmit, and it would be hugely successful, according
to the standards of a transmission chain experiment where participants cannot de-
cide what to pay attention to and what is worth transmitting. In real life, I doubt
that would be a hit. The opposite is also true: we are at times motivated to listen to,
and to retransmit, cultural traits that are not particularly memorable and easy to
reproduce. As Olivier Morin puts it, “genealogies, lists of gods and ancestors, and
other staples of traditional oral cultures seem, after all, as memory-​friendly as a
phonebook.” Digital, online, cultural transmission may provide a test bed for the
effects of attention and motivation to transmit as opposed to comprehension (not
really necessary) and memory and reproductions abilities (not necessary at all).33
This is possibly a new way to look at this issue, and we do not yet have robust
hypotheses on how the differences between cultural transmission with an encode-​
and-​retrieve phase—​which has been the main, if not the only, way of transmission
for the great majority of human history—​could differ from cultural transmission
without it—​which may be the next winner. The experiment we just mentioned sug-
gested that there was no impact for disgust-​eliciting elements. Joe Stubbersfield
and colleagues, whom we met in the previous chapter, analyzed instead the dif-
ferent effects of social information and content analogous to threat-​related infor-
mation, which they call “survival” information. Social information, if you recall,
is information about intense social relationships and, in their experiment, was
exemplified by headlines such as “Man caught naked by surprise birthday party”
182  Transmitting and sharing

or “Father and daughter have accidental cybersex.” For threat-​related informa-

tion they used titles such as “Steroids in chicken cause ovarian cysts” and “Woman
killed by spiders in her hair.” They found that stories with social information were
recalled and transmitted better than stories with threat-​related information in the
classic Whispers Game scenario, but they were not more successful when parti-
cipants had simply to choose what they wanted to read and what they wanted to
transmit.34
It may be incautious to make a sweeping generalization from this experiment.
After all it is a single result, coming from testing a few dozens of students in an
English university, but we can at least spell out what it suggests. Social informa-
tion may have a memory-​and-​recall advantage in respect to threat-​related infor-
mation (but why that should that be the case?), but this advantage disappears when
choose-​to-​receive and choose-​to-​transmit phases replace the actual transmission.
Alternatively, one could interpret that threat-​related information recuperates, so
to say, its memory disadvantage because we are more motivated to pay attention to
them, and to share them. It has been suggested that individuals who share threat-​
related information are considered as more competent than individuals who share
the same information, but without mentions of threats. As we explored previously,
there are a series of reasons why information about threats should be particularly
attractive and, as a consequence, it makes sense that the individuals who share it
would be paid attention to. This, in turn, would make sharing information of this
kind particularly appealing. But then, why is there not the same effect for sharing
social information? By sharing social information one could signal one’s know-
ledge of relationships within the group, and even one’s ability to interpret other
behaviors.35
What about the emotional content? Transmission chain experiments, as well
as our analysis of recent fake news, suggest that there should be a strong advan-
tage for negatively-​marked information, no matter whether the message needs
to be remembered and retold or simply shared online. Studies that focused spe-
cifically on online sharing, however, paint a more complex picture. Jonah Berger
and Katherine Milkman collected three months of New York Times articles that
appeared on the homepage of the journal, for a total of almost 7,000 articles, and
they checked which of them made it into the “most emailed list.” They found that
articles that were more positive were likely to be emailed more than the negative
ones. In fact, they found that the negative-​positive axis was less informative, to pre-
dict what went viral, than an axis between content eliciting arousal (such as awe,
anxiety, and anger) and content eliciting what they called “deactivation” (such as
sadness), with the former, not surprisingly, being more effective.36
It is useful to spend some time on these results, also to reflect more generally on
the limits of our analysis. First, that the most successful New York Times articles
were awe-​inspiring and generally positive may have much to do with the fact that
the readers of the New York Times are not a random sample of the population. The
 Online sharing  183

most emailed article, as I write. is titled “Advice from a formerly lonely college stu-
dent,” and it is indeed awe-​inspiring and positive, if slightly sappy. A story about
a Cornell University freshman—​it is early October—​who learned to forget high-​
school friends and not rely too much on social media (who would have thought!)
and fully enjoy her new life at college. Second place is “Donald Trump versus the
jungle,” a criticism of Trump’s foreign policy. The two examples give a good pic-
ture of some features of the average New York Times reader: democratic and, more
importantly, educated, possibly with kids about to be freshmen. It is difficult to
imagine this reader sharing a story about a morgue employee cremated during a
nap (still, one million people did it on Facebook in 2017). The same thing can be
said, of course, about the readers of the suspect websites we considered in our ana-
lysis: they are hardly a random sample. As we discussed in the previous chapter,
psychological factors of attraction work on average, their effects are discernible
when zooming-​out, at the big scale, and there is substantial variation when consid-
ering different individuals.
Second, the New York Times analysis considered successful the articles that
were most shared by email. Now, sharing something by email is very different
than sharing something on social media. Email sharing implies a direct connec-
tion to the person, or the few people with whom you are sharing the article. If you
read the article about the Cornell’s student, not-​lonely-​anymore, you can think
of your friends who are also at college, or who have children there. Generally, it
may make more sense to share positive than negative content with your email
contacts, who are often familiars and friends. Social media sharing may be felt
to be less personal, as it is a form of one-​to-​many communication, not addressed
to someone in particular, but to all your friends or followers. Thus, on top of in-
dividual and demographic variation, there is variation produced by the specific
digital tool one is using. It happened a few times to me that, when I met in real-​
life people with whom I just have had Twitter contacts, they were surprised to
know that I had other interests beside science, and in fact besides a few limited
fields of science. Some even criticized me because, according to them, I do not
take political stances in my life. But, of course, this was my Twitter persona, not
myself!
Just as general labels of “misinformation” or “fake news” cover quite different
things, underpinned by different psychological motivations, as we suggested in the
previous chapter, the label ‘digital transmission’ also includes very different phe-
nomena. Is that an insurmountable problem? It depends. There are some com-
monalities, which I have tried to highlight here, and these commonalities allow
for some generalizations. Up to a certain point. As long as we remember that we
can and we need, sometimes, also to zoom-​in, both endeavors are useful. It would
be tempting (and possibly more remunerating) to offer here the laws to make your
content online “contagious.” There is no shortage of advice on how to “go viral” on
social media. Unfortunately I cannot.
184  Transmitting and sharing

However, here a few take-​home messages: general psychological factors, which

act online as well as offline, can contribute to explain the success of online misin-
formation. As we said, misinformation can travel wide, far, and long, as it is de-
signed, intentionally or unintentionally, exactly to do this, as it does not need to
be constrained by reality. We are not sensitive to an abstract notion of truth, but to
various cues that point to the importance of the content and which may be associ-
ated only on average with truthfulness.
Online communication is special for various reasons, but one, which I have em-
phasized in this chapter, is that digital, online, media provides a cheap, fast, and easy
way to implement high fidelity transmission. Different from the cases most studied
in cultural evolution, we do not need to remember, replicate, and in fact, not even
necessarily understand, what we share: motivation may be the only winner. What
the consequences of this revolution are, when we focus on what content can be
advantaged, is not yet clear, but I hope to have at least convinced you that this may
be a productive way to look at this phenomenon. We saw that the same psycho-
logical factors that impact encode-​and-​retrieve transmission chains also impact
the spread of online misinformation. This is indeed not surprising. Content that we
find attractive and interesting will also be, often, easier to memorize and reproduce
than content that we do not find attractive. Strong emotions are finalized to direct
attentional resources to the stimuli that produces them, so that we will also process
and memorize them better.
While we do not know if high fidelity online transmission favors specific kinds
of content compared to low fidelity transmission, we do know that fidelity, going
hand in hand with online availability, is generating an exponential growth of cul-
tural traits. How does this influence cultural change at large, at the population
level? Is cultural change getting faster because of that? The more the better? Or is
cultural evolution simply producing way more junk than before? These will be the
topics of the next chapter.

Notes

1. Dawkins (1976, 1999), Burman (2012)

2. Brodie (1996), Lynch(1996), Blackmore(1999). See also: Hull (1988), Aunger (2000)
3. Boyd and Richerson (2000)
4. Henrich and Boyd (2002)
5. Sperber (2000)
6. Charbonneau (2018)
7. Acerbi and Mesoudi (2015)
8. Sperber (1996)
9. https://​www.edge.org/​response-​detail/​25388
10. Sperber (2000)
 Online sharing  185

11. Scott-​Phillips  (2017)

12. Morin (2018)
13. Acerbi and Mesoudi (2015), Boudry (2018), Scott‐Phillips et al. (2018)
14. Clasen (2012)
15. Tennie et al. (2006)
16. McGuigan et al. (2011)
17. Acerbi et al. (2011)
18. Rubin (1995)
19. http://​www.pewinternet.org/​fact-​sheet/​internet-​broadband/​
20. Logan (1999)
21. Dingemanse et al. (2015)
22. Morin et al. (2018)
23. Research on repair mechanisms in online communication is limited, but see, for ex-
ample, Meredith and Stokoe (2014).
24. https://​www.youtube.com/​channel/​UCAL3JXZSzSm8AlZyD3nQdBA/​featured
25. Tooby and Cosmides (1992)
26. https://​www.wired.com/​1994/​10/​godwin-​if-​2/​
27. https://​medium.com/​@CRA1G/​the-​evolution-​of-​an-​accidental-​meme-​ddc4e139e0e4
28. Adamic et al. (2016)
29. Lerique and Roth (2018)
30. https://​www.captiongenerator.com/​make-​a-​hitler-​reacts-​video
31. Shifman (2014)
32. Eriksson and Coultas (2014)
33. Morin (2016)
34. Stubbersfield et al. (2015)
35. Boyer and Parren (2015). Other studies in cultural evolution exploring situations
in which participants did not need to recall and reproduce the material transmitted
are, for example, Mercier et  al. (2018) that focuses on the “willingness to transmit,”
or Stubbersfield et al. (2018), where participants could edit the stories on a computer
screen before passing them on.
36. Berger and Milkman (2012)
 8
Cumulation

Accumulation, improvement, and ratcheting

If someone could travel in time from 10,000 years ago to today, they would no
doubt be blown away by the differences between their world and ours. People from
10,000 years ago were in many respects like us: in fact, they were us. We have more
and more indications that genetic evolution has continued in relatively recent times
(an example cultural evolutionists are particularly fond of is the diffusion of genes
allowing people to process milk in adulthood, that became widespread as a conse-
quence of the domestication of cattle, goats, sheep, and in general milk-​producing
animals), but my bet is that we could not tell apart an individual born 10,000 years
ago from ourselves, based on physical differences—​of course, we could if we ana-
lyzed their genome.
Also, in the eighth millennium BCE, important features of what we would rec-
ognize as “modern” civilization were already in place. Humans were spread around
most of the globe; agriculture was starting to be developed across different areas;
pottery, art, religions were already millennia old. Human groups were different
from each other:  if someone coming from a settlement in North America and
someone from Southeast Asia would have met, they probably would have been
surprised by their differences.
Still, our worlds would be deeply diverse. An individual born 10,000 years ago
cast in a modern metropolis (let’s take Shanghai) would be first surprised—​and
probably frightened—​by the fact that so many humans can live together in such
a limited space. Whatever estimate we want to use, the population of Shanghai
today is four or five times the population of the world 10,000 years ago. Besides
this, what kind of things would blow them away? Computers and smartphones,
cars, trains, aeroplanes, machinery in a hospital or in a factory, yes; but also elec-
tric lights, glasses, fridges, and, definitely, guns. If we explained vaccines, space
travels, or digital global connectivity to them, that would also make an impres-
sion. These are all examples of what cultural evolutionists call cumulative cultural
evolution.
What is cumulative cultural evolution? There is no description accepted unani-
mously: in a recent review of the concept, Alex Mesoudi and Alex Thornton list
more than 30 different definitions. There are three aspects that seem to be espe-
cially important to describe something as a case of cumulative cultural evolution,

Cultural Evolution in the Digital Age, Alberto Acerbi. Oxford University Press (2020) © Oxford
University Press.
DOI: 10.1093/oso/9780198835943.001.0001
 Accumulation, improvement, and ratcheting  187

as opposed as being simply “cultural”: accumulation, improvement, and ratcheting

(see Figure 8.1 and descriptions thereafter).1
First, there should be accumulation, some “increase in diversity and complexity
over time.” This is the first line of the figure: after some time, as indicated by the
arrow, where there was only one cultural trait (A) there are now three (A, B, and
C). As anthropologists have long documented, human culture is impressively rich
and diverse. A few years ago I was teaching cultural evolution to masters’ students
in ethology at the University of Stockholm. In the first lecture, I used to present a
renowned paper on culture in chimpanzees. In the last few decades, primatologists
have described several behaviors that appear to be characteristic of particular popu-
lations of chimpanzees, including, for example, tool use, such as nut cracking, or
different techniques to collect ants or termites with sticks of various shapes and di-
mensions. These behaviors are regionally variable: some populations perform them,
and others do not, or the techniques vary from population to population. Moreover,
it is considered unlikely that different ecological conditions could account for all
behavioral dissimilarities, so that they are explained as the result of local traditions
supported by chains of social transmission, as we described for humans.
In the paper, 64 behaviors are analyzed, and for at least 39 of these a cultural
origin is proposed. After some discussion with the students on how likely the hy-
pothesis of the cultural basis of these behaviors was, I used to present a slide showing
a menu from a pizzeria in the suburbs of Stockholm. In this example—​admittedly

A C

A Accumulation
B

A
C improvement
A
B

C
B
A A Ratcheting

Figure 8.1  Aspects of cumulative cultural evolution.

188 Cumulation

not a pinnacle of human culture—​there were 40 different pizza options (including

“death by pepperoni”), one more than all cultural traits showed by chimpanzees.
Human culture, when compared with the culture of any other species, independ-
ently of whether we want to consider these traits genuinely “cultural” or not, is ex-
tremely diverse. Of course, an exhaustive investigation of possible pizza kinds and
toppings around the world would bring us very far from the topic of this book.2
Technological variation is a prime example of cultural diversity, and it has the
advantage of being relatively easy to quantify. There are hundreds, or thousands,
different Linux distributions; Wikipedia reports around 700 different “notable”
programming languages (excluding “dialects of BASIC, esoteric programming lan-
guages, and markup languages”). Things were not different in the past: we are still
negatively affected by the proliferation of power plugs and sockets that followed
the diffusion of electricity. Objects such as the contemporary fork, the design of
which seems self-​evident and relatively uniform, come in fact in various shapes,
and historically they went through stages in which several different models were
used at the same time. At the beginning of the last century, cutlery sets comprised
up to 146 distinct pieces. An image in Henry Petroski’s book The Evolution of Useful
Things shows 36 different forks in a silverware collection, including:

Oyster fork-​spoon, oyster forks (four types) berry forks (four styles), terrapin, let-
tuce and ramekin fork [ . . . ] large salad, small salad, child’s, lobster [ . . . ] mango,
berry, ice-​cream [ . . . ] fish, pie, dessert, and dinner fork.3

Linguists assess that around 7,000 languages exist today, and this figure is esti-
mated to be a fraction of the languages spoken before the Neolithic. Synonyms—​
happy, merry, jolly—​can be considered cultural variants. Google digitized more
than eight million books, representing “only” 6 percent of all books ever published,
and 400 hours of video were uploaded every minute to YouTube in 2015.4
Accumulation, still, is not enough. The second feature of cumulative culture
is improvement. Improvement requires an increase in effectiveness, or perform-
ance: the new cultural trait needs to be so to speak, a better version of what was
already there. If accumulation was the only distinctive attribute of cumulative cul-
ture, the increase in the number of pizza toppings, books, or even different LEGO
blocks (they did became more numerous in the last 60 years) would count as cu-
mulative culture. However, these examples do not work, or they do not until we can
establish that they also reflect an increase in effectiveness.5
Drawing an analogy with biological evolution, cultural evolutionists use to say
that cultural traits compete with each other. With respect to 100 years ago, contem-
porary sets of cutlery do not generally include more than 20 distinct pieces. Some
pieces have never competed with one another: the shapes of serrated and plain
edge knives serve different functions, the former being useful for cutting objects
with a hard surface and a soft interior such as bread or tomatoes, and the latter for
 Accumulation, improvement, and ratcheting  189

precision cuts, such as peeling an apple or chopping vegetables, and we still have
both. Others, such as the four types of oyster forks, went through a process analo-
gous to stabilizing selection, with one, intermediate, shape being used: a cutlery
set with more than one type of oyster fork (if any) would be an eccentric curiosity
nowadays. Synonyms can also be considered as traits that have the same function,
with some of them more successful than others. Technologies with the same, or
similar, function, replace each other, or become employed in specialized settings,
such as vinyl records being replaced by cassette tapes, then replaced by CD, and
then replaced by mp3s readers, which are today, in turn, being replaced by multi-​
function smartphones.6
Competition is competition for resources too, be they physical storage, time, or
generic “brain space.” Names represent a vivid example: you simply cannot have
more names than people alive. The most popular male baby name in the US in 2016
was, with 19,015 new-​borns, Noah. Fifty years earlier, however, only 157 babies
were named Noah, making it only the 598th most common name. A staggering
79,989 children were then called Michael—​the most popular name in 1966—​but
now it would be surprising to hear a toddler called as such: only 241 babies were
called Michael in 2016, ranking the name 897th.7 Names, like many other cultural
traits, including words, show a very skewed distribution of popularity, where very
few items are very successful and the great majority are not, as we discussed in the
chapter on Popularity. Only a few things, among all cultural productions, can be
used or remembered, even if the constraints are less obvious than the very material
count of living humans that makes first names compete. Only a tiny minority of the
130 million books ever published, according to estimates from Google, is widely
read. For the few successful YouTube videos, there are literally billions of them with
only a handful of views.
Competition, however, is not necessarily proof of improvement. Noah is not
better than Michael. “Happy” is used more than “jolly,” but is it better? Are forks
today better than the forks of 100 years ago? In the second line of Figure 8.1 above,
not only there are more cultural traits at time t + 1, but C is better than B, which is,
in turn, better than A (the size, in Figure 8.1, stands for any possible form of im-
provement). The idea of increase in effectiveness is not itself foolproof. A sharper
knife is better than a dull one, and could indeed provide a long-​term genetic
advantage for a population of sharper-​knife users: they will be less likely to cut
themselves while operating the tool, they will need less time to process food, they
will be more precise, and they will perhaps be able to cut things that were not pos-
sible at all before. A vaccine is better than bloodletting, but what about an mp3s
player and a cassette tape? The former is more effective, but its effectiveness is de-
fined with criteria that are internal to the cultural system in which it operates. It is
quite unlikely that the usage of mp3s players increases human genetic fitness. In
the same review mentioned above, Mesoudi and Thornton refer to this as “cultural
fitness.”
190 Cumulation

Finally, even accumulation and improvement are not enough to define cumu-
lative cultural evolution. Japanese macaques are among the stars of animal social
learning studies. They are also called “snow monkeys”: you may have seen them
in pictures where, in winter, pink-​faced and with snow in their hair, they seem to
chill in hot springs. In the 1950s, in the small island of Koshima, a female macaque
was observed performing a novel behavior consisting of washing sandy potatoes
with seawater. Sandy potatoes were recently introduced by researchers themselves,
but they were not appreciated by macaques, at least until they discovered how to
wash them and remove the sand. Within a decade, the new behavior spread among
almost all the individuals in the group—​especially the younger ones. While it is
an open question how exactly the washing potato habit spread, we can agree that
this behavior represents an improvement. Monkeys that washed the potatoes had
access to an additional food source, which in the long run represents a net fitness
advantage. In any case, sweet potato washing would not be considered a case of
cumulative cultural evolution: it never changed. Seventy years after the first obser-
vation of the behavior, the monkeys in Koshima are still washing potatoes, in the
exact same way as they did in the fifties.8
Or consider nut cracking in chimpanzees. Nut cracking is one of those primate
behaviors that are considered “cultural,” as mentioned earlier. Nut cracking is not
a small feat. Chimps choose the right hammers and anvils depending on the hard-
ness of the nuts they want to open. Different groups, and different individuals,
use hammers made either of stone or wood, and of various sizes, and they use
different kinds of anvils. Juveniles spend plenty of time trying to master the tech-
nique, and they seem to purposely hang around adults who are cracking nuts.
As for potato washing, the cultural nature of the behavior is debated. This is not
surprising: culture is a property, the set of cultural things has fuzzy boundaries,
and the debate is probably bound to go on and on (personally, I would put nut
cracking very close to the “reconstructive” side of the continuum we discussed
in the previous chapter). However, this is not what matters here: the interesting
fact is that nut cracking behavior, generation after generation, did not change, or
it changed minimally. In a recent review of the emerging field of primate archae-
ology, the authors write that “current evidence puts the emergence of chimpanzee
stone technology in the late middle Pleistocene, perhaps as recently as 200–​150
kyr ago.” “Recently,” though, depends on your perspective: a way to look at this is
that thousands of generations have passed, and chimps are repeating the same be-
havior again and again.9
The third, missing, ingredient for cumulative cultural evolution is thus that
innovations and social learning are iterated every generation and they produce
an aggregate improvement. The last line of Figure 8.1 depicts schematically this
idea: trait C is not only better than A, but it is been built based on B, which was
built, in turn, based on A.  Psychologist Michael Tomasello famously used the
metaphor of the ratchet:
 Harry Potter and space travels  191

a key feature of uniquely human cultural products and practices is that they are
cumulative. One generation does things in a certain way, and the next generation
then does them in that same way—​except that perhaps they add some modifi-
cation or improvement. The generation after that then learns the modified or
improved version, which then persists across generations until further changes
are made. Human cultural transmission is thus characterized by the so-​called
“ratchet effect’,” in which modifications and improvements stay in the population
fairly readily (with relatively little loss or backward slippage) until further changes
ratchet things up again.10

Even if potato washing and nut cracking are considered cultural, and even if they
represent a net improvement for the individuals that perform them, they are not
cumulative, because, if we look at the population level, they did not improve across
generations. Here is another way to put it: products of cumulative cultural evolu-
tion are impossible, or extremely unlikely, to be invented by a single individual.
When provided with the appropriate tools, chimpanzees can reinvent by them-
selves, without observing others, nut-​cracking behavior. Many human products
instead seem clearly out of the reach of the inventiveness of each of us. Provided
with the appropriate tools I doubt I will build a laptop or a car (remember, to make
things worse, I am supposed not to have seen any of these before). To be honest,
I also seriously doubt I would figure out a recipe for a decent lasagne by myself.
Enough with the definition of cumulative culture now, it is time to go back to our
hypothetical visitor from 10,000 years ago.11

Harry Potter and space travels

We agree, I think, that fridges, vaccines, electric lights, space travels, and all that
our visitor found mind-​blowing, are good examples of cumulative cultural evo-
lution. Now for a different question: are there other things that the visitor will not
find mind-​blowing? Even though we will not understand each other, for example,
the visitor may not be too impressed by our linguistic abilities. It is reasonable that
the languages spoken 10,000 years ago were similar, in many respects, to the lan-
guages we speak today. What about Harry Potter? No doubt, almost all the details
will dumbfound our guest, but the main plot may be more familiar than, for ex-
ample, the science behind vaccines. What about the rituals of the Catholic Church?
What about our body ornaments?
I am not claiming that there will not be differences, or that rituals, body orna-
ments, or languages have not changed through time, as this would be absurd. There
is a thriving academic field studying language evolution, which cannot be surveyed
here, but language does accumulate changes. First, experiments convincingly show
that some features of languages can be due to short-​term cultural evolution. Simon
192 Cumulation

Kirby and colleagues put people in the lab and asked them to learn an “alien” lan-
guage, through a procedure similar to the transmission chain experiments dis-
cussed above. Imagine you have a set of various stimuli, say a circle, a square, and
a triangle, and that these objects can perform different actions, say falling down,
circling or zigzagging from left to right. At the beginning of the experiment, the
“words” used to indicate the stimuli are generated at random: let’s say that a tri-
angle falling down is labelled “polire,” a triangle circling is labelled “qujyt,” a square
falling down “asprex,” and so on (see Figure 8.2). The first participants see some
stimuli and the words assigned to them, then they see other stimuli—​unlabelled,
this time—​and they have to name them. Some of the new stimuli are items that
participants have seen before, but others (unbeknown to them) are new combin-
ations. Then, new participants enter and they do the same thing, but they see the
label generated by the previous participants, instead of the random words used at
the beginning. By iterating this process through generations of participants some-
thing interesting happens: languages become structured. Participants start to use
the same label for the same actions and for the same objects: if, say, a triangle falling
down remains “polire,” a triangle circling may be referred to as “poscan” and a
square falling down as “mixlire.” (What would you call a square circling?)12
Also, if it is difficult to estimate how the languages spoken 10,000 ago sounded,
there are good indications that demographic factors influence the features of lan-
guages. Languages with many speakers are different from languages with few
speakers, so one can hypothesize an evolutionary trajectory going from languages

Shape Movement

Polire Po-lire

Qujyt Po-scan

Asprex Mix-lire

Figure 8.2  How an “alien” language evolves.

Data from Kirby, S., Cornish, H. and Smith, K., Cumulative cultural evolution in the laboratory: An
experimental approach to the origins of structure in human language, Proceedings of the National
Academy of Sciences, 105(31), pp. 10681–​10686, doi.org/​10.1073/​pnas.0707835105,  2008.
 Harry Potter and space travels  193

with few-​speaker-​features to languages with many-​speaker-​features. It has been

proposed that languages with many speakers have a simpler morphology than
languages spoken by a few. In Italian, for example, we have various ways to ex-
plicitly mark remoteness distinctions when using verbs in the past tense. We can
use, among others, passato remoto for an event that happened in the distant past
(“Io andai in campeggio nel 2006”) and passato prossimo to describe the same
event happening more recently (“Sono andato in campeggio ieri”). In English, one
would use the same form for both sentences, saying “I went camping in 2006” and
“I went camping yesterday.” The difference needs to be deduced by the context.
Languages with many speakers would be, on average, more similar to English than
to Italian in this respect. One hypothesis is that the reason behind this is similar to
the reason why alien languages in laboratory become structured. They need to op-
timize expressiveness with simplicity, as they are learned by more people, adults in
particular.13
These results are fascinating, and I do not have reasons (nor, indeed, the com-
petence) to doubt them. Still, cumulative cultural evolution of language, the type
we see in the lab with circling triangles and falling squares, is extremely fast. It
needs to be extremely fast, as the random pairings that are devised to kick-​start the
experiment are not a usable language. Our 10,000-​year-​old friend will certainly
have a structured language. The demographic effects point to a slower evolution,
and it could make sense that the visitor from the past would have a language mor-
phologically more complex than contemporary English, for example. But would
this blow their mind? After telling them we went to the moon, it is time to tell
them “hey, we have only one form for the past tense!” They would be able, in a bit,
to understand that to express the differences they have to refer to other parts of
the sentence. This is indeed what I did, with not much trouble, when I first learnt
English.
Art, also, shows some kind of cumulation. In his dissertation, Oleg Sobchuk
asks, among other things, when the detective novel was “invented.” Why is invented
in quotations marks? Well, because nobody invented the detective novel. Sobchuk
convincingly argues that what we call detective novel is a cumulative product—​
an “aggregate improvement”—​of various inventions made in different times by
different authors. For example, Caleb Williams, a 1794 novel written by William
Godwin, is considered the first to use a device that will be then found in countless
detective stories: the plot unfolds backwards, from effects to causes. There is, say,
a murder in the first chapter, and, after that, the detective goes back to reconstruct
what led to it: in many detective novels, the end is the revelation of the causes of the
event that happened in the beginning. Fifty years later, Edgar Allan Poe used the
same trick in The Murders in the Rue Morgue, adding other innovations, such as the
“locked-​room” puzzle—​that is, the scenario of a dead body found in a room closed
from inside, which will be used in countless other detective stories. The Moonstone,
in 1868, an epistolary novel from Wilkie Collins, introduced the “closed circle of
194 Cumulation

suspects,” a situation in which a crime is committed in a relatively isolated setting,

like a train or a country house, and where there is a limited number of possible cul-
prits, each with some, more or less likely, motivation (Agatha Christie, anyone?).14
The evolution of the detective story looks like a good example of the ratchet
in action. Today we have at our disposal a bigger bag of tricks than Poe had, and
surely, more are being invented and added to the bag. Innovations are made and
retained through generations; they depend on past innovations—​a “locked-​room”
puzzle without a backwards plot would not be very breathtaking—​and make the
cultural trait in question more fit.in the sense of cultural fitness: a novel with more
of these plot devices will be, on average, more psychologically appealing than a
novel with less. As for language, in principle I do not have any problem with this
description, and I do think that the claim that artistic forms show cumulative cul-
tural evolution (and have a fitness!) is acceptable and informative. However, again,
how much did they evolve, when compared to other cultural domains? The locked-​
room trick is an excellent one, and I fear that if I had to invent it myself I would be
unlikely to come up with it. If I had to bet, though, and choose if I would be more
likely to invent the locked-​room plot device or electricity, I would not have much
doubt about where to put my money.
In sum, my guess is that the visitor from 10,000 years ago would be astonished
in different degrees by our achievements in different domains. In other words, my
hypothesis is that cumulative culture works differently depending on the material
that can be accumulated and on the ways in which information is transferred.
Unfortunately, as discussed above, the definitions of cumulative cultural evolu-
tion are far from rigorous, so it is difficult to measure quantitatively the “degree
of cumulativity” of a particular domain. In addition, as mentioned, records from
the past are often incomplete: in the case of technology, for example, they have to
be reconstructed drawing on, by necessity limited, archeological findings. In other
cases, they are not missing at all, like for spoken languages, where they can be only
inferred backwards from contemporary observations.
A shortcut, as we saw for the morphological complexity of languages, is to con-
sider the relationship between population size and the characteristics in which we
are interested in. To an extent, the assumption that population sizes can be used
as a proxy for degrees of cumulative culture is an acceptable one. Not surprisingly,
many cultural evolutionists have investigated the association, or the lack thereof,
between complexity and demography and, again not surprisingly, results starkly
differ. Whereas laboratory experiments, in controlled situations, usually find a reli-
able correlation between group size and ability to solve a problem—​and to transmit
improved solutions of the problem along the transmission chains you are now fa-
miliar with—​when looking at data in the real world, things get more complicated.
One would not only expect the relationship between population and complexity
being influenced by many other factors, such as contacts between groups or local
ecological conditions, but, if you accept my hypothesis, one would not expect the
 Digital cumulation  195

existence of a general, one-​size-​fits-​all, relationship, as it depends upon the domain

considered.15
With Jeremy Kendal and Jamie Tehrani, we analyzed the link between com-
plexity of folktales and population size for a few hundred tales recorded in
European and western Asian populations (we excluded other geographical areas
to avoid biases, as European and western Asian populations were sampled more
than others in our sources). We used various measures to define complexity, such
as the number of different stories present in a single society, the number of mo-
tifs included in the tales (motifs are standard narrative blocks in which folktales
are coded, say “the hero encounters a magical animal” or similar), and, finally, the
number of traits for variants of the same story, for tales with high diffusion. For ex-
ample, there are several variants of Little Red Riding Hood in different societies, and
they differ in how many details are included. In some variants, the hunter cuts open
the wolf ’s belly to save Little Red Riding Hood and the grandmother; in others, this
detail is omitted. The more details present, the more we considered a variant com-
plex. Our analysis showed that there is not, for this specific domain, a consistent re-
lationship between complexity and demography: in other words, folktales in small
populations are as complex as those in big ones. This suggests that there has not
been a noticeable cumulative cultural evolution for folktales.16

Digital cumulation

Here is a—​speculative, I admit—​proposal, that will bring us back to the discus-

sion of the digital age: domains that are closer to the “reconstructive” end of the
continuum we described in the previous chapter are less cumulative than domains
closer to the “preservative” end. Domains that support high fidelity transmission
of new information will be more likely to accumulate innovations than domains
in which individual reconstruction is the workhorse. Remember the stars in Dan
Sperber’s thought experiment: nobody along the chain learns anything new when
tracing them, and indeed if we mix up all the drawings, there is no way to order
them on a timeline. No transmission of new information, no cumulative improve-
ment. Thus, we would expect cultural cumulation to have the potential to increase
in time (and in bigger populations) for high fidelity transmission domains.
On the other hand, a problem with the transmission of new information is that
copying errors can deteriorate what is passed on. Copying errors are more dele-
terious in small populations than in bigger ones, to the point of being possibly fatal
in the extreme case in which nobody in a small population retains the correct ver-
sion of a trait that has been passed on. At the beginning of the book, we discussed
how, when effective cultural population sizes are too small, cultural traits spread
with difficulty and can even be lost. However, Sperber’s stars will be very unlikely
to disappear, as even single individuals will tend to reproduce the same features.
196 Cumulation

Therefore, we would expect, for domains characterized mainly by individual re-

construction, that cultural complexity would not depend on population size, and
thus not increase, or decrease, in time, exactly as happened for the folktales we
studied.
The idea that fidelity of transmission is important for cumulative culture is
widespread in cultural evolution theory and it makes intuitive sense.17 The differ-
ence in my proposal is that fidelity of transmission is neither a necessary property
of culture per se, as many traditions can remain stable without being supported
by faithful transmission, nor an automatic consequence of some peculiar human
ability. Humans are good copiers, but selective ones, so that cultural domains need
something else to be maintained and developed.
From this perspective, if you accept the hypothesis that we examined at length
in the previous chapter, we can consider that digital media, by boosting fidelity
of cultural transmission, are opening new possibilities of cumulation. Couple this
with the enormous increase in availability that we discussed at the beginning of the
book and there are reasons to think that a widespread access to media supporting
hi-​fi transmission can radically change the way in which our culture evolves. Give
it another 100 years, and what will blow the mind of our visitor from the past?
Of course, this is speculative terrain. In a book in many respects comparable to
the one you are reading, Alex Bentley and Mike O’Brien describe “the acceleration
of cultural change,” produced by the fact that cultural transmission went “from an-
cestors to algorithms” (the two quotes make up the title of their book). One of their
favorite examples is science. The increase in number of scientific publications is
indeed exponential. It is difficult to provide a precise figure, but recent estimates
point to an increase of around 8 percent each year, which translates to a doubling
time of approximately nine years. This means that, today, the global scientific
output is double the scientific output of nine years ago and, as difficult as it is to
keep track of it today for scientists, in another nine years it will be the double that of
today. If you wish a long and productive career in science for yourself, you should
know that, if the trend continues, in, say, 36 years, the global scientific output will
be sixteen times the output of today.18
These are impressive numbers: science is a very good example of accelerating
cultural cumulation. It is worth pointing out, however, that the increase of articles
only measures accumulation:  we need to assume that improvement and ratch-
eting go together with it. Still, is the acceleration of scientific production a digital
age phenomenon? There are a few caveats. First, the same exponential growth has
been in place, according to the authors of the estimates, since the mid-​1600s, al-
beit slower (less than 1% up). There was a first acceleration in the middle of the
eighteenth century (2 to 3%) and, more importantly, the actual pace started in the
period between the two world wars, so that we do not (yet) know the importance
of digital media in this process. Second, there are several other plausible reasons
to explain the recent growth in the number of scientific publications. One is the
 Digital cumulation  197

fact that more people are working as scientists around the world: the number of
PhDs awarded also increased exponentially starting from the second world war,
both within countries, such as the US, and especially globally, as other countries
started to be involved in scientific activities. Another is a change in the system of
incentives within academia, encouraging scientists to publish more and more.19
Third, and most importantly, science is a domain in which procedures to in-
crease the fidelity of transmission have been implemented from the beginning. It is
not too preposterous that at least some aspects of the scientific practice as we know
it today (think quantitative measures, as well as theories expressed quantitatively)
have had the effect—​whether explicitly intended or not—​of making transmission
of ideas more faithful. What changed in the last century, and probably even more
in the last decades, is what we called availability, which is in fact what Bentley and
O’Brien are mainly interested in, and this may be key in explaining the lift of the
exponential growth.
As briefly touched on in the chapter on misinformation, what appears to be the
genuine scientific revolution of the digital age is not necessarily an increase in dis-
coveries (there are even some suggestions that the rate of scientific discoveries is
actually decreasing when compared to investments).20 It may mainly be, so far, a
change in the practices that surround scientific activity, from the process of peer-​
review to the circulation of raw data and codes used for analysis, from the feedback
in real time provided by social media to the diffusion of digital preprints. These
changes were either impossible before the diffusion of digital media (think widely
circulated preprints), or confined to niche groups and, possibly, often in informal
and oral conversations (think discussions on the peer-​review process).
At the risk of sounding eccentric, we could learn by more mundane ex-
amples: domains in which digital transmission has made information both more
available and, especially, more faithfully transmitted than it was before, and where,
accordingly, the degree of cumulation has so far been limited. Some ideas: food
preparations, children’s rhymes, songs, conspiracy theories, games, and, more
broadly, art and narratives, religions, and all the things that our 10,000 year old
visitors would have found puzzling but not mind blowing. Is there an intrinsic limit
for the cultural cumulation of these domains, or could availability and hi-​fi trans-
mission provided by digital media make them unrecognizable to us in the next
centuries, or even decades?
Culinary recipes, for example, have been transmitted, mainly orally, for mil-
lennia within familiar circles or within relatively close groups of acquaintances.
Recipe books are an old invention—​a millennia old invention—​but their wide-
spread diffusion and usage can be traced back only to the last century. But with on-
line recipe websites, everybody has at their disposal thousands of recipes, variants,
comments and suggestions, coming from different geographical areas, and with all
the “fidelity amplifiers” we described in the previous chapter. Here a quote from a
recent study of online recipe recommendations:
198 Cumulation

The desire to look up recipes online may at first appear odd given that tombs of
printed recipes can be found in almost every kitchen. The Joy of Cooking alone
contains 4,500 recipes spread over 1,000 pages. There is, however, substan-
tial additional value in online recipes, beyond their accessibility. While the Joy
of Cooking contains a single recipe for Swedish meatballs, Allrecipes.com hosts
“Swedish Meatballs I,” “II,” and “III,” submitted by different users, along with 4
other variants, including “The Amazing Swedish Meatball.” Each variant has been
reviewed, from 329 reviews for “Swedish Meatballs I” to 5 reviews for “Swedish
Meatballs III.” The reviews not only provide a crowd-​sourced ranking of the
different recipes, but also many suggestions on how to modify them, e.g. using
ground turkey instead of beef, skipping the “cream of wheat” because it is rarely
on hand, etc.21

During October 2018, on Pinterest, a social media site that allows users to share
(“pin”) videos and images, more than half of the top 1,000 shares were recipes.
Whereas Pinterest has a relatively small number of active accounts (still, relatively
small means around 250 million), the top 1,000 stories generated three million
“pins.” The above-​mentioned Allrecipes.com claims that 95 recipes are viewed
every second.22
What are the consequences? Many studies, as the one just mentioned, investi-
gated the diffusion of online recipes, focusing often on how to determine what fea-
tures make a recipe popular, or how digitalization of recipes allows reconstructing
so-​called “ingredient-​flavour networks”—​basically how pairings between ingre-
dients work. The question of whether recipes (or everything else we discussed)
show more cumulative improvement) because of the digital support where they
are transmitted remains largely unexplored. Are there more recipes? Are they more
“culturally fit”? Are innovations in past recipes retained through “generations” of
recipes?23
Or think about chess. The most common way to rate chess players is through
a system called Elo, after the name of its inventor. The Elo rating assigns points to
chess players according to their match results (winning, losing, or drawing) and
the Elo rating of the opponents. If a player A wins against a player B, for example,
A’s rating will be increased by a certain amount (the more the higher the rating
of B). The opposite happens when a player loses. At the end of 2018, among the
20 best players of all-​time, according to the Elo rating, 15 of them achieved their
best score after 2010. The average Elo rating of the top 100 players in 2018 is 2703,
whereas it was 2544 in 1975.24
Figure 8.3 shows the average Elo rating of the top-​100 players from 1971, when
the World Chess Federation officially adopted this metric, until 2018. From the
second half of the 1980s, the Elo rating of the best players started to increase con-
stantly. Elo ratings above 2500 are far, very far, from the average chess player, but
the same data shows that the increase extends to all the players listed by the World
 Digital cumulation  199

2700

Average Elo rating

2650

2600

2550

1971 1980 1990 2000 2010 2018

Year

Figure 8.3  Average Elo rating of the top-​100 chess players through time.

Chess Federation. Whether the best players today are stronger than the grandmas-
ters of the past is a question that fascinates chess enthusiasts and computational
scientists alike, and it would not be wise trying to settle it here. Some believe for
example that the Elo rating are is not suitable for comparison of players from dif-
ferent time periods, as it has been inflated in recent years due to an influx of more
and more players, so that the contemporary higher scores would not correspond
to an actual superiority. However, many think that players today are indeed better,
and they have especially improved in the last 20 or 30 years.25
Why should that be the case? Why the increase from the late 1980s? There are
two interrelated main factors that could explain the improvement, and both are
likely to be important. The first one is the diffusion of computers able to play chess,
and the growth of their ability. In the 1980s, home computers started to spread,
and chess games were among the most popular. The Chessmaster 2000, advertised
as “the finest chess program in the world,” was published in 1986, and was the first
title in what is considered today as the best-​selling chess software series. In 1997,
the defeat of the then-​world-​champion Gary Kasparov by IBM’s Deep Blue was
one of the iconic moments of our era, but the truth is that today any chess applica-
tions on your iPhone would easily win against Deep Blue, and possibly against any
human. Young practitioners do not need to be trained by human counterparts any
more, but they can play against artificial opponents, at different level of skills and
with different tactical strategies. Artificial intelligence can also analyze the moves
they did, and provide assessments and advices on how to improve.26
The second factor, which is more related to our point, is that chess went online.
Each of us has pretty much all the matches ever played available to re-​play and to
study. In addition, the matches are accompanied by comments, videos, and ana-
lyses of important moves or alternative scenarios, all ways to broadly improve the
transmission of skills. Furthermore, when tired of contending with a machine, one
200 Cumulation

can play online with an endless number of human opponents. In sum, availability
and the improvement of transmission made possible by digital media may have
played a fundamental role in the recent evolution of chess skills.27

Junk culture

Most of us would agree that digital media provide availability and increase in
fidelity of transmission, but there may be more dispute about what the conse-
quences are. My bet is that the majority opinion today is that we are headed to-
ward a looming scenario in which hard-​to-​imagine amounts of information are
produced, forgotten, and abandoned in humanity’s new digital memory:  you
can call it “junk culture.” Together with historical chess matches and excellent
cooking recipes, there are out there millions of registrations of matches between
beginners, or bad advice on how to make pizza: accumulation without improve-
ment and ratcheting. Digital-​sceptic James Bridle uses the example of YouTube
videos and, in particular, videos for an audience of toddlers (as a father of a
three-​year old, I happen to possess a quite robust non-​academic knowledge of
the topic).28
Big hits for my daughter and, I suppose, her peers, are the likes of Wheels On
The Bus, Baby Shark, or Finger Family. The most viewed version of Wheels On The
Bus, when searching on YouTube, is from the channel “Little Baby Bum,” and it has
more than 2 billion views, making it the second most viewed YouTube video which
is not a music video. The first, among non music video, with more than 4 billion
views, is another episode of an animated series: the Russian Masha and the Bear.
Baby Shark, at 1.9 billion, scores as the third most viewed, while, for Finger Family,
the biggest hit has only around 950 million views, but, as we will see in a minute, is
interesting for other reasons.29
These numbers are impressive, but what is even more impressive is that, con-
sistently with what we are discussing, there are uncountable variants, copies with
small modifications, and remixes of different versions. When my daughter asks to
watch the Wheels On The Bus video, her favorite, she asks nowadays for the “blue”
or “green” bus, or for the one with “the robot” or “the spider,” among the many
versions that she knows, which no doubt represent a tiny percentage of the pos-
sible variants. Of course, there is no shortage of Baby Shark versions, and there are
plenty of videos in which Baby Shark lyrics are sung to the melody of Wheels On
The Bus and vice versa.
James Bridle focused, for good reasons, on the variants of the video Finger
Family. Finger Family is originally a nursery rhyme, where each of the hand’s fin-
gers is a character (daddy, mummy, brother, sister, and baby) and, for each, a simple
verse is repeated (“Daddy finger, daddy finger, where are you? Here I am, here I am,
how do you do?” and so on). Given this simple scheme, it is easy to add, with basic
 Junk culture  201

knowledge of computer graphics, various personalities to the fingers. In the most

“traditional” versions, they are indeed fatherly and motherly figures, followed by a
boy and a girl and, finally, a baby. Many, many, variants of this basic version already
exist. But why be limited to that? There are countless versions of mildly amusing
(yes, I am being sarcastic) Finger Family with animals, superheroes, or even means
of transport. Characters of other productions, such as Mickey Mouse or the pro-
tagonists of Frozen are also more or less copied-​and-​pasted on the basic template,
creating new variants.
But stranger versions exist. Some are just plain weird, for example featuring
computer-​graphic generated animals in unnatural colors (pink elephants, green
lions, etc.) fighting, while others are more troubling. One version features five
finger-​adapted Mickey Mouses shooting each other (the strand “normal-​cartoon-​
characters-​doing-​strange-​things” is huge by itself, including a fake Peppa Pig going
to a sadistic dentist that surfaced into worried mainstream media). In another ver-
sion of Finger Family the five fingers are Hitler’s faces. I have no doubt that it would
be possible to find many other examples. A journey into these derived versions of
children’s videos is, for an adult, a bizarre experience indeed.30
Why are these versions out there? It is not straightforward to calculate advertise-
ment earnings from YouTube, but it is sensible that, to make a decent amount of
money, videos need to rack up several thousands of views. However, there are two
factors that conjure in making variants of children’s videos a possibly rewarding
strategy. First, the original versions, as we just saw, have millions and millions of
views. Searching for Finger Family, a child can go through the various options
offered and end up viewing one of the derived versions. In addition, and more
importantly, the algorithm controlling the auto-​play feature in YouTube will pro-
pose, after the first ordinary Finger Family versions, all the other versions that are
around, possibly choosing among those that are most popular, unconcerned about
whether they are weird or not. Second, the cost of realizing derived versions is very
low: even if earnings are unlikely, one can flood YouTube with several algorith-
mically generated variants of Finger Family, with the hope that some of them, for
random reasons, will be successful.31
But why, then, the Hitlers and the gunned Mickey Mouses? If these derived
versions are unsupervised products of algorithms (their instructions being some-
thing like: (i) pick face-​like figures at random in some repository, (ii) paste them
on the basic Finger Family template, and (iii) publish on YouTube), they are just the
random output of the activity of the machine. This possibility is quite fascinating
and I cannot exclude it, but, from what we know, it is more likely that the creators of
the derived versions want sometime to have fun, and they throw in the bunch one,
let’s say controversial, version. I am ready to acknowledge that this conception of
fun activity is, at a minimum, debatable but, on the other side, the Hitler versions
of Finger Family represent an infinitesimal proportion of all the possible versions.
In addition, we do not know what effect these videos have on children. At this point
202 Cumulation

in the book, you may suspect I am quite skeptical of any important consequence
of occasional exposure to these stimuli, unless, of course, if a parent insisted on
showing the Hitler versions to their children, which would be another, certainly
worrying, problem.
The proliferation of YouTube video variants is an engrossing example of
digital junk culture, but the term may be misleading here: “junk” should not be
taken as an evaluative term (biologists talk about “junk DNA,” even though they
often now prefer to use the more neutral “noncoding DNA”). Junk culture is all
the pieces of information that are produced and left out there but nobody uses.
After all, the Hitler versions of Finger Family enjoy considerable cultural success.
James Bridle discussed them, and all his readers now know about them. I am
doing the same, and there is some probability that you will also remember them.
The weird Peppa Pig got covered by the media everywhere, as we mentioned, and
you may even tell your friends about it, and so on and on. While we may agree
that the world could have done without the Hitler versions of Finger Family, this
is not junk culture. False but successful news, as the piece about the morgue em-
ployee cremated by mistake while taking a nap, and being shared a million times,
is not junk culture.
However, one can search for Wheels on the Bus videos uploaded on YouTube
in the last hour, and, as I write, there are eight of them, all but one with no views.
I do not know whether this is representative or not, but it is reasonable to think
that there may be many thousands of versions of Wheels on the Bus, with just a
minority of them being successful. As we discussed earlier, online competition,
and cultural competition more generally. produces skewed distributions with just
a few very successful items while the majority of them are less lucky. In practice,
almost all YouTube videos have less than a hundred views. There are websites that
capitalize on this vast amount of material, providing streams of unseen YouTube
videos: a hypnotic journey—​which I’d advise—​into the long tail of online (un)
success.32
Of course, this is not just the case of YouTube videos: skewed distributions and,
as a consequence, junk culture, are everywhere. They are also present offline, but fi-
delity and reach/​availability, while providing the possibility of more accumulation,
exacerbate the problem of junk culture online. As we discussed at the beginning of
the book, small and scarcely connected populations create a natural bound to the
proliferation of cultural items, be they good or bad, useful or not. Unfaithful trans-
mission, on top of this, paradoxically provides a virtual sieve for culture. If remem-
bering and storing information or actual artefacts is costly, the unsuccessful ones
will be forgotten or transformed in the process of transmission.
How do we deal with junk culture in the digital age? How, in other words, can
we have access to what is worth having and can we discard what is not? Bentley and
O’Brien think that there is no other possibility except letting algorithms do this job
for us, and I tend to agree.
 Algorithms  203

Algorithms

If you follow 1,000 users on Twitter, and each user tweets three times a day, they
produce 3,000 tweets. Assuming that you need ten seconds to read a tweet, this
makes for more than eight hours of scrolling through your timeline every day so
as not to miss anything. Of course many of those tweets will not be very inter-
esting for you, so you may hope to pick up the good ones in the healthy half-​an-​
hour-​a-​day you devote to social media. In July 2013, Facebook estimated that,
on average, every time someone visits the News Feed they could be exposed to
around 1,500 stories from friends, pages that they follow, or sponsored content.
The problem scales up when you need to retrieve information from the internet.
A  search for the term algorithm on Google yields about 325,000,000 results.
Assuming again an optimistic ten seconds to read one result, you would need
37,615 days, or more than 100 years, to check each result (no free weekends, of
course, and you should never sleep). In sum, in this situation, there is simply no
way we can find anything useful without some form of curation. This is not a
new problem: if you remember from the beginning of this book, the diffusion of
the printing press generated a comparable panic, with intellectuals worried the
world would “fall into a state as barbarous as that of the centuries that followed
the fall of the Roman Empire.”33
There is a joke among pilots—​of course, I never heard any pilot actually telling
the joke, but I  found it online in plenty of commentaries about automation—​
according to which the best aeroplane crew is composed of a computer, a pilot,
and a dog. The computer flies the plane. The pilot feeds the dog. The dog’s task is
to bite the pilot whenever they try to interfere with the computer’s work. In other
words: let the machine do the job for you.
Algorithms today surely have a bad reputation. We cursorily examined algo-
rithms twice in this book. First, in the chapter about “Misinformation” we dis-
cussed how algorithmic selection of what we see in our social media feeds could
promote what we called “shallow engagement.” Shallow engagement favors content
that elicits quick reactions, in form of likes and comments, as opposed to content
that is in-​depth, truthful, or relevant. Then, in the previous chapter, algorithms
were considered as the general template of recipes, instructions manuals, and on-
line how-​to: sequences of instructions that aspire to be precise enough and, at the
same time, do not need pre-​existing knowledge, thus to be executed by a mindless
device.
Time to tie up loose ends:  the algorithms are sequences of instructions, un-
ambiguous and substrate-​neutral. When you divide 456 by 7 you are applying
an algorithm. You use an unambiguous sequence of instructions, :  all other
individuals—​and machines—​that apply the same algorithm will obtain the same
result./​You use a substrate-​neutral sequence of instructions: it does not matter if
you are dividing 456 candies among seven children, or cutting a 456-​centimeter
204 Cumulation

long piece of fabric to sew seven skirts. In this sense, algorithms are abstract
entities: being suspicious of “algorithms” as such is like being suspicious of divi-
sions or sorting procedures.
What is the scary part? Well, as you may have heard, machines control more
and more of our daily life. Algorithms are no longer a topic for philosophers and
engineers: machines decide what we see online, what drugs we should take given
our symptoms, and, if not now, then in a not-​so-​remote future, they will drive our
cars and choose whether we get a job or not. There are several lines of criticism
about how algorithms are employed today. I want to discuss at least three of them.
I believe these criticisms are correct but, I think, they are not decisive. A thorough
treatment of these topics would require another completely different book. Still,
I will spend some time on them, to make at least acceptable the idea that automatic
filtering of online information is necessary, and that some of these issues can be
framed in our cultural evolution perspective.

Biased algorithms

According to one line of criticism, algorithms reproduce the biases of the program-
mers who created them, or of the data used to train them. Automatic job screening
does look like a scary development. There are plenty of intuitive reasons that make
the idea of a machine having the power to take life-​changing decisions unpleasant,
especially if I am the subject of these decisions. Not surprisingly, accounts of “al-
gorithmic biases” in hiring are not missing. In October 2018, it was found that an
algorithm used by Amazon was discriminating against women. The algorithm was
fed with the previous decade of hiring decisions, and, since men were hired in the
majority of cases, it learned to give a negative weight to CVs whenever it found
mentions of women-​only colleges or membership of women-​only associations
(“women’s chess club”): in fact, to give negative weight to the bare presence of the
term “woman” or “women.”34
Algorithms are stupid—​I cannot resist mentioning an SMBC comic where The
Rise of the Machines fails as robots are fighting humans with spears and rocks, as
their machine-​learning algorithm was fed with data on historical battles, and “the
vast majority of battle-​winners used pre-​modern weaponry” (see Figure 8.4)—​but
that is indeed the point.
Biases will be present, but this is not a reason to throw the baby out with the
bathwater: it is a reason to abandon an overly naïve view for which algorithms, as
such, are foolproof procedures that always yield a perfect result (I am not sure who
would support this view, but I did find it often presented when criticizing algo-
rithms). The crux is not whether algorithms are biased (yes, they are!) but whether
we can recognize and correct biases in algorithms more or less efficiently than we
can in humans. This is, you will concede, an open question.35
 Biased algorithms  205

Figure 8.4  Rise of the Machines, from Saturday Morning Breakfast Cereal.

© Zach Weinersmith.

Let’s put it in another way. I am sure you can think of a committee composed of
humans that decided not to hire you for reasons that were, in your opinion, debat-
able, and perhaps not fair. Would an algorithm have been any better? I am a non-​
native English speaker and it is a reasonable assumption that this will disadvantage
me in an interview for a job. I am not talking about biases against my demographic
group (that is, interviewers being against non-​native English speakers as such), but
about the possibility that I may not be able to understand a subtlety in a question,
that my answers will look less brilliant because the breadth of my vocabulary is
more limited than that of a native speaker’s, or that I will appear not too fond of
informal verbal interactions as they may be more demanding for me than for a na-
tive speaker. Would an automatic process be preferable? Would I prefer to be inter-
viewed by a computer in front of a camera and to have my answer analyzed by an
algorithm, assuming that it would be able to understand what I am talking about?
Well, I’d say I am not against it.
If we think that the hypothetical bias against non-​native speakers is real, and it
is a problem, we could design the algorithm to rescale up the score for non-​native
speakers, so that a non-​native speaker getting a score slightly lower than a native
206 Cumulation

one would have the same final evaluation from the algorithm. Is that a desirable
outcome? Perhaps there are good reasons to think that a candidate who can under-
stand subtleties slightly better and who has a bigger vocabulary will do a better
job. And what if a non-​native speaker had a very good education in the English
language and speaks as a native with minimum effort? Should this candidate be
evaluated higher than a native English speaker who got the same score? This looks
unfair. But perhaps the fact that they mastered in depth a second language says
something interesting about the candidate? And, what about bilinguals?
In sum, subtle biases and their effects are very complicated to assess, both dir-
ectly by humans and indirectly by programming algorithms, but what about ob-
vious shortcomings, like the Amazon algorithm penalizing women? In a way, this
is a positive story: the bias was recognized and corrected by Amazon’s engineers.
Research is far from unanimous in showing shortcomings of algorithms and, when
procedures based on algorithms are explicitly compared to human activity, ma-
chines tend to do better than their fleshy counterpart. Selections of board directors
are one of the many possible cases of hiring decisions in which human biases can
have an undesirable influence. Researchers compared how a machine-​learning
algorithm would fare against real human decisions. They found that candidates
who were chosen by humans but who scored low according to the algorithm were
later unpopular with shareholders, and that candidates not chosen by humans but
who scored high according to the algorithm were successful with shareholders in
other companies. Why was that the case? Human decision tended to be biased by
features like gender, financial background, and previous experience with other
boards. The algorithm was less biased: “the algorithm is telling us exactly what in-
stitutional shareholders have been saying for a long time: that directors who are not
old friends of management and come from different backgrounds both do a better
job in monitoring management and are often overlooked.”36
However, programmers can also intentionally insert biases in the algorithms.
Here is a “random walk” through YouTube recommendations:  you can start by
choosing any video on YouTube and then click on the first “Up next” suggestion.
If you continue clicking on the “Up next,” you are effectively letting the algorithm
take all the decisions for you. Having a software performing more than 170,000
random walks through YouTube videos, researchers from the Pew Research Center
found that the algorithm pointed users to progressively longer contents (the first
randomly chosen videos were, on average, 9.3 minutes long, while, by the fourth
recommendation, the average was almost 15 minutes) and to more popular videos
(8.1 million views on average for the first video of the “walk,” and around 40 mil-
lion for the fourth).37
In general, all algorithms that select which information to show us online are
biased, in this explicit sense. As discussed in Chapter 6 “Misinformation,” algo-
rithms that select content that generated more engagement in social media (likes,
comments, etc.) can have the side effect of optimizing for shallow engagement.
 Opaque algorithms  207

The fact that the YouTube algorithm proposes longer and longer videos is also a
questionable practice. However, as mentioned above, there is simply no way to
find information without some form of selection. The crux, thus, is not that algo-
rithms should not be biased, as this is simply impossible and actually ineffective.
Unintentional biases should be monitored and taken care of, and the presence of
intentional biases that select for information should be transparent, and finalized to
the discovery of useful information instead of explicitly aiming at “hooking” users.

Opaque algorithms

The second broad criticism is that algorithms are opaque to users, and more
worryingly, because of their growing complexity and interconnectedness, to pro-
grammers themselves. Algorithms are opaque to users. Opacity is, after all, one
of the distinctive features of digital cultural transmission that has accompanied
us through the whole book. Our social media feeds, our Google searches, our
Amazon and Netflix suggestions, not to mention personalized advertisements in
the majority of websites, do not provide us with a random, unfiltered, selection of
information. This is inevitable: Amazon has a catalogue of hundreds of millions of
products. The probability of random suggestions having any relevance to us is in-
finitesimal. But, should I know why I had that exact book recommended to me this
morning instead of another one? Not necessarily.
Cultural evolution works by making available to us tools and ideas that we did
not have to invent by ourselves and, more to the point, of which we do not need
in many cases to understand the nitty-​gritty details. Imagine if we were to start
worrying about not knowing how our laptop works under the hood and, as a result,
we were to stop using it. Perhaps this looks somehow attractive, but the same logic
applies to almost all tools and ideas. How do a bike’s brakes work? Why do we cook
potatoes but eat lettuce raw? If we were really concerned with the opacity of how
things around us work, we would be paralyzed in our daily life. As explored earlier,
there are many reasons to “just do” as we have been told or demonstrated. Opacity,
in sum, is not a problem by itself, and we placidly live with opacity in many, if not
all, domains of life.
We are—​rightly—​suspicious of opacity when we fear that someone is mis-
leading us, or when things do not work as they should. If, when pulling the brakes,
my bike keeps going as though nothing happened, I have very good reason to want
to know more about the mechanism. Most likely, however, I will defer to experts in
the domain and let them repair the brakes, so that the mechanism will still remain
opaque to me. A problem arises if experts themselves do not know any more how
the brakes work. In a world of increasing complexity and increasing interconnect-
edness, the functioning of many pieces of technology becomes opaque to every-
body, experts included. It is perfectly plausible, and I do not see how it could be
208 Cumulation

different, that there is not a single person in the world who knows how a laptop, an
airplane, but also the Italian law system functions in its entirety.
Does anybody know how the algorithm that Google uses to present search
results to us works? This is an interesting question, but the answer may be more
complicated than what it seems. In a sense, no. As with all products of cultural
evolution, the Google algorithm stands on the shoulders of many giants and many
dwarves, big and small innovations accumulated through time. Luckily, there is no
need for programmers to know how the libraries (collections of pre-​written func-
tions that can be used as-​it-​is) they use work in the details, nor how high-​level
languages, such as Python or C++, are translated in low-​level, machine languages.
Complicated algorithms are created by teams of programmers, and a programmer
who worked on the graphical output will not know much, if anything, of the ma-
chine learning technique used to produce the results (and vice versa).
In addition, outputs like Google searches or Facebook newsfeeds are likely the
result of the optimizations of hundreds of parameters. Facebook may take into ac-
count your likes in the last day, week, or month, or possibly all of them, weighted
differently, the links you clicked, the likes of your friends (how many of them? how
long back in time?), the people you interacted with on Messenger, and who knows
what else. The weight of the parameters can be fine-​tuned according to their effects
on users. Simplifying a bit, one can imagine that engineers at Facebook test modi-
fications such as giving a 1 percent more influence to the likes of your last day in
respect to your last week likes. If, as a result, we spend a few seconds more on the
social media (multiply this for billion of users), the modification is kept. However,
some weight modifications will work better when accompanied by others. Perhaps
giving more influence to recent likes will have a positive effect—​positive, it goes
without saying, for Facebook—​only when the weight of your friends’ recent likes is
also increased. In truth, it will depend on how all other parameters are set. Perhaps
the success of the recent likes modification depends in a not-​obvious way on fa-
voring link clicks of pictures, or reading less news, or anything else you can im-
agine and that can be tracked. The point is, the algorithm is probably optimizing
hundreds if not thousands of parameters and it is not obvious to the programmers
themselves why exactly a 1 percent increase in the importance of recent likes, or
anything else, has an effect. Let us go back to our algorithm replacing the hiring
committee. Through an optimization such as the one just described, we find out
that it gives some importance to, say, wearing multiple rings. I use this example
both because I have two and because I cannot think about any way in which this
could be useful information for hiring a person. This is not a desirable outcome,
right? Again, this should caution against the naïve view that we are in full control
of algorithms. We are not: but again, the key is to compare the opacity of the algo-
rithm to the opacity of a human hiring committee. We will never find out whether
the president of the committee likes rings, or dark hair, or any other feature we are
pretty sure is not effective at all for a better choice. Humans are more opaque than
 Personalized algorithms  209

algorithms. Even if we do not know exactly why the algorithm produces this result,
as its procedure is opaque, we do know that it depends on the weight of the “ring”
parameter. We can go back to the drawing board and work out what is going on.

Personalized algorithms

The third, and last, criticism I consider is related to the possible extreme personal-
ization of algorithms. The case of Google is interesting: the original algorithm, the
well-​known PageRank, was a relatively simple ranking system that, at its core, clas-
sified websites based on the number of links they received. Website A, containing
the term “algorithm,” linked to by another ten websites, would have appeared first
in my search than website B, also containing the term “algorithm”, but linked to
by only five other websites. In fact, not just in my search: in all searches. In the fol-
lowing years, however, other algorithms have been used together with PageRank
and, by 2009, Google employees confirmed that “We’ve been telling people for a
long time that they shouldn’t focus on PageRank so much. Many site owners seem
to think it’s the most important metric for them to track, which is simply not true.”
These other algorithms were, in the majority of cases, providing personalized re-
sults, based on previous searches, localization, and other, known and unknown,
features. Today, if I search for pizza, my results are based on where I am, and where
I ate in the last few years. Even if I search ‘algorithm’ on Google, it is likely that my
results are different from yours.38
The original PageRank algorithm generated concerns because it could be
cheated: a way to climb up the ranks for a website was, for example, to create several
fake websites linking to it. Or, you may be old enough to remember the practice of
inserting in the text of websites keywords that would have appealed to the Google
web crawler, again, pushing the website up in the results. However, it is with per-
sonalization that the majority of criticisms arose. Personalization can be extremely
fine-​tuned today. Netflix, for example, does not simply provide you with a selected
subset of available movies and TV series based on your past viewing behavior and
ratings, but it presents them to you in a particular way. When a movie or TV series
is commercialized, the studios provide generic title images—​what would have
been, until a few years ago, the DVD cover or the movie poster—​and Netflix ini-
tially used these images to advertise them in your homepage. Then they discovered
it was more effective to personalize the image as well. Thus, if I am a fan of, say,
Morgan Freeman, Netflix will not only suggest that I watch Bruce Almighty (a 2003
movie starring Jim Carrey, where Freeman played the role of God), but, instead of
showing me in the homepage the official poster, where Freeman is not present, it
will present me with a screenshot from a scene where he is the main protagonist.39
This is a benign example, of course, but personalization is another feature of
automatic selection of information that worries many of us. Personalization is
210 Cumulation

useful, but how far can it go without being “creepy?” How fine-​tuned to my pre-
vious interests should the selection be, while still allowing me to find new in-
formation, without being stuck in the notorious filter bubble? As discussed in
Chapter  5  “Echo chambers,” there is little empirical evidence that warrants ex-
cessive worries about online segregation. The only study we considered that spe-
cifically looked into the effect of algorithmic selection of Facebook showed that
diversity, still remaining overall relatively high, was reduced more by users’ choices
than by the algorithmic filter. A recent research on news consumption suggests
that people who find news via search engines are exposed to a wider political spec-
trum when compared to people who affirm not to use search engines but go dir-
ectly to news outlets websites.40
Contrary to the prevailing common view, the majority of studies that analyzed
algorithmic personalization found little effect of informational segregation.41 In
one of the most comprehensive studies of Web searches personalization, from
2017, it was found that, on average, 11.7  percent of the results of Google Web
Search showed differences between individuals, that is around one out of ten.
Independently of whether you think this is a high or low proportion, the inter-
esting aspect is that the great majority of these differences are due to language and
geographical differences rather than users’ previous search behavior. If I search for
La Rochelle, a French town on the Atlantic coast, from Italy the first result is the
Italian Wikipedia page on the city, but it is the English Wikipedia page if I search
from Bristol in the UK, from where I am now writing. And, as above, my pizza
searches will be very localized. Searching from Bristol, the first non-​localized re-
sult, the English language Wikipedia page for pizza, is in the second page.42

The problem of finding an optimal balance between selecting relevant, personal-

ized, information and narrowing the variety down to the point that not enough
new, alternative, material is accessible is of course central for any algorithm that
selects information online. More generally, research on the various aspects of algo-
rithmic curation will be crucial in the coming years. My objective here was simply
to suggest that there is no viable alternative to algorithmic curation, and that the
best way forward is to try and understand how the specific algorithms work, and
what their benefits are and what their disadvantages. Hand waving against algo-
rithms may have the opposite effect, with social media users (that is, us all) be-
coming generally suspicious of any form of algorithmic curation.43 Algorithms are
biased and opaque, and personalization comes with a cost. We need to abandon
a naïve view of algorithms and be fully aware of their shortcomings. We should
be suspicious of bad algorithms just as much as we would be suspicious of a any
system that gives the wrong result when we divide two numbers, but it would be
unproductive being suspicious of divisions.
Given abundance, selection is a key mechanism for cumulative culture, and
algorithms that select information have now a decisive role when determining
 Personalized algorithms  211

which cultural traits can survive. In the digital age, availability and widespread fi-
delity may allow for cumulative culture in domains where it was not possible, or
it was limited, before. Along with useful traits, however, availability and fidelity
also allow for bare accumulation, and the proliferation of junk culture may follow.
A challenge for the coming years will be to design algorithms that help humans de-
cide what is junk culture and what is worth preserving. This is, not surprisingly, a
task that is far from easy.

Notes

1. Mesoudi and Thornton (2018), see also: Dean et al. (2014)

2. Whiten et al. (1999)
3. https://​en.wikipedia.org/​wiki/​List_​of_​programming_​languages, Petroski  (1994)
4. Lewis (2009), Lin et al. (2012), https://​www.statista.com/​statistics/​259477/​hours-​of-​video-
​uploaded-​to-​youtube-​every-​minute/​
5. Bartneck and Moltchanova (2018)
6. Petroski (1994), Mesoudi (2011)
7. https://​www.ssa.gov/​oact/​babynames/​
8. Galef (1992)
9. Boesch and Boesch (1982), Haslam et  al. (2017), for a different view on the potato
washing behavior, see: Schofield et al. (2018)
10. Tennie et al. (2009)
11. Tennie et al. (2009)
12. Kirby et al. (2008)
13. Lupyan and Dale (2010), but see also Greenhill et al. (2010)
14. Sobchuk (2018)
15. Shennan (2011), Derex et al. (2013), Vaesen et al. (2016)
16. Acerbi et al. (2017)
17. See e.g. Lewis and Laland (2012)
18. Bornmann and Mutz (2015), Bentley and O’Brien (2017)
19. OECD (2016)
20. Bloom et al. (2017)
21. Teng et al. (2012)
22. https://​w ww.newswhip.com/​2018/​11/​f ive-​lessons-​content-​pinterest/​, http://​press.
allrecipes.com
23. Ahn et al. (2011)
24. https://​en.wikipedia.org/​wiki/​Comparison_​of_​top_​chess_​players_​throughout_​his-
tory, http://​ratings.fide.com/​toplist.phtml?list=men, https://​www.mark-​weeks.com/​
chess/​ratings/​
25. Regan and Haworth (2011)
26. https://​en.wikipedia.org/​wiki/​Chessmaster
27. https:// ​ w ww.prospectmagazine.co.uk/​ m agazine/​ e scaping-​ c heckmate-​ w hy-
​human-​chess-​has-​survived-​the-​robot-​conquest
212 Cumulation

28. Bridle (2018)
29. https://​www.youtube.com/​watch?v=HP-​MbfHFUqs,
https://​www.youtube.com/​watch?v=KYniUCGPGLs,
https://​w ww.youtube.com/​w atch?v=XqZsoesa55w, https://​w ww.youtube.com/​
watch?v=YJyNoFkud6g,
https://​en.wikipedia.org/​wiki/​List_​of_​most-​viewed_​YouTube_​videos
30. https://​www.today.com/​parents/​moms-​warn-​disturbing-​video-​found-​youtube-​kids-​
please-​be-​careful-​t101552, https://​www.telegraph.co.uk/​news/​2017/​03/​27/​youtube-​
fire-​hosting-​disturbing-​videos-​aimed-​children-​tricking/​, https://​www.youtube.com/​
watch?v=463F4_​K3FGg
31. http:// ​ w ww.pewinternet.org/ ​ 2 018/ ​ 1 1/ ​ 0 7/ ​ m any- ​ turn- ​ to- ​ youtube- ​ for- ​ childrens-​
content-​news-​how-​to-​lessons/​#an-​analysis-​of-​random-​walks-​through-​the-​youtube-​
recommendation-​engine
32. http://​astronaut.io
33. https://​www.facebook.com/​business/​news/​News-​Feed-​FYI-​A-​Window-​Into-​News-​
Feed
34. https://​ w ww.reuters.com/​ a rticle/​ u s-​ a mazon-​ c om-​ j obs-​ a utomation-​ i nsight/​
amazon-​scraps-​secret-​ai-​recruiting-​tool-​that-​showed-​bias-​against-​women
35. https://​www.smbc-​comics.com/​comic/​rise-​of-​the-​machines
36. Erel et  al. (2018), https://​hbr.org/​2018/​04/​research-​could-​machine-​learning-​help-​
companies-​select-​better-​board-​directors
37. http:// ​ w ww.pewinternet.org/ ​ 2 018/ ​ 1 1/ ​ 0 7/ ​ m any- ​ turn- ​ to- ​ youtube- ​ for- ​ childrens-​
content-​news-​how-​to-​lessons
38. Lazer (2015), https://​en.wikipedia.org/​wiki/​PageRank
39. http://​nymag.com/​intelligencer/​2018/​08/​seo-​is-​back-​thank-​god.html, https://​www.
clickz.com/​machine-​learning-​choice-​paralysis-​netflix-​p ersonalizes-​title-​images/​
204354/​
40. Bakshy et al. (2015), Fletcher and Nielsen (2018)
41. Flaxman et al. (2016), Zuiderveen Borgesius et al. (2016), Möller et al. (2018)
42. Hannak et al. (2017), https://​twitter.com/​searchliaison/​status/​1070027261376491520
43. De Vito et al. (2017)
 Conclusion

It is up to us

As digital media pervades so many aspects of our lives, and they may do it even
more in the future, it would be foolish to pretend to have provided an exhaustive
treatment here. There are plenty of areas that this book did not consider. As
I write, for example, a few companies, such as Google, Facebook, and Amazon,
control disproportionate shares of our activities in the digital world. When Mark
Zuckerberg, in his 2018 Capitol Hill hearing, was asked if he considered the pos-
ition of Facebook as monopolistic, his rebuttal was: “Certainly doesn’t feel like it
to me.” The answer provoked one of the very few laughs in almost seven hours of
testimony. There are not-​easily dismissible arguments to make Internet Service
Providers, but possibly also web search engines, or even social media, publicly
owned utilities. The privacy of our personal data, be they web searches, social
media posts, private, or supposedly so, emails, is currently a hot topic. Even if
one were to trust the big companies that manage them, and there are some
good reasons for not doing so, how can we be protected from breaches from ill-​
intentioned third parties?1
Are we spending too much time in front of screens? Ten years after the first
iPhone, Apple introduced a feature called “Screen Time,” that provides a report of
the time spent on various applications, the number of notifications received, and
allows limits to be set on them. (If Apple is genuinely interested in the well-​being
of its customers or if it is just responding to the increasing concerns by adding,
ironically, another attractive feature, is of course another story.) The majority of re-
searchers are cautious about the off and on alarms about smartphone addiction, or
about the negative effects on well-​being, or even depression, linked to the usage of
mobile phones and social media. Nevertheless, it seems common sense that, even
if only because smartphones provide so many things we are interested to, we could
overuse them: trite as it is, you can have too much of a good thing. Remember the
story of Goldilocks and the three bears? In one bowl the porridge is too hot, in
another one it is too cold: only in the third is the porridge “just right.” As the tem-
perature of porridge does not correlate linearly with how “right” the porridge is,
engagement in digital activities is not linearly correlated with decreased mental
well-​being. Both too much and too little use of digital media are associated with
lower mental well-​being. In addition, the hypothesis that the usage of digital media
influences mental well-​being is at least as plausible as the hypothesis going in the
other direction, that is on average, people who are more anxious, stressed, or de-
pressed use digital media more (or less) than others.2

Cultural Evolution in the Digital Age, Alberto Acerbi. Oxford University Press (2020) © Oxford
University Press.
DOI: 10.1093/oso/9780198835943.001.0001
214 Conclusion

What about the billion or so individuals who do not have access to digital media,
or to internet? The International Telecommunication Union, the United Nations
agency for information and communication technologies, estimated in 2017 that
only around 20 percent of people living in African countries had had access to the
internet at least once in the previous year. This figure differs broadly depending
on the country. Nigeria has more than half of the population online, but for many
others, including Madagascar, Burundi, or Republic of the Congo, the figures are
below 10 percent. Eritrea, at the very bottom, is estimated to have just above 1 per-
cent of its population online. Add to this the fact that limited access is matched by
a gender gap, with the proportion of men using the internet being higher than the
proportion of women in two-​thirds of countries worldwide, in general the least
developed ones (the gender gap is negligible in Europe, and reversed in North and
South America, with slightly more women than men online).3
These issues, and possibly others, are no less important than the ones I have cov-
ered here. Perhaps some of them are more important. I dealt with themes that fit
naturally with the research done in cultural evolution and cognitive anthropology.
I focused on what we can call informational aspects of our digital and online lives.
How do we select information, and how does this information possibly influence
our behavior? How do digital media change cultural transmission, each time you
and I interact online, but also at the larger scale of cultural cumulation? My claim is
that it is possible and, I hope, useful to isolate and to dig into these topics, and that
this investigation could contribute to understanding the more general picture. If
you are worried, say, that all the history of your online activity today will be used in
twenty or thirty years to build a model of your personality by some evil company or
state, then you should be concerned with data privacy and read (also) about that.
Notwithstanding these limitations, our approach underscored that many of the
apprehensions that surround our digital and, especially, online activities are mis-
placed. I am persuaded that the negative effects of our daily interactions with digital
media may have been overestimated, and I do think that, for what we know now,
there are many positive effects that we just take for granted. The long view we took
here, based on a cultural evolutionary approach, suggests that we are wary learners,
resistant to various forms of persuasion. It suggests that many of the logics that
govern offline communication and cultural transmission can be also applied to on-
line communication and cultural transmission. It finally suggests that, if we will be
able to find ways to separate wheat from chaff. cultural evolution may be enhanced
by the availability, the reach, and the fidelity that digital and online media provide.
Being relatively optimistic is more difficult to sell, as we discussed when analyzing
how a negative emotional content is a good ingredient to make your ideas spread
with efficacy. The internet, and especially social media, have been identified as the
cause of almost everything bad that has happened in the last years. In some cases,
of course, “bad” depends on one’s point of view: social media has been blamed for
the rise of polarization and populist politics in western countries, the spread of the
 Informational inequality  215

anti-​vaxx movement, or teen depression and even an increase in teen suicide. More
picturesque targets are not missing, such as vandals ruining national parks with
graffiti to post pictures on social media or small, local, restaurants spoiled by too
many positive reviews.4
There is no doubt that someone will write their name on a tree and then share
a picture on Snapchat, but, as we saw, thinking that social media, or any form of
social influence, made them do it is inaccurate. It is up to us. I personally find this
message quite uplifting. Furthermore, the real danger of the narrative that charges
social media for almost everything bad happening is that we may end up not ad-
dressing the real problems. The rise of populist movements is a problem, that
parents decline to vaccinate their children is a problem, the increase (if real) of teen
depression is a problem. Attributing them to social media and smartphone is easy,
and mainly they become someone else’s problems. The risk is that, if the target is
mistaken, the problems will remain.

Informational inequality

Of course we can and should use our knowledge to make things better. As men-
tioned earlier, a seemingly paradoxical reason why, for example, online misinfor-
mation appears as such a pressing problem is that it is, just like everything else,
more available than before. That almost a million people shared, liked, or com-
mented on a version of “Morgue employee cremated while taking a nap” looks
surprising only to the people who would not share it. As we have observed, misin-
formation on the internet exists not because online communication is ineffective
but, on the contrary, because it is very effective, and everybody ends up coming
across what they like. For some it is the alert of the new issue of the Journal of
Artificial Societies and Social Simulations (you can replace it with any specialized
and possibly obscure academic publication), for others it is the heated political
discussion, and for others again it is the last—​fake—​news about “Florida Man,” a
folkloristic internet figure associated with bizarre and unusual events. As I write, a
meta-​meme invites us to search in Google for our day of birth and “Florida Man”
and just see what news items appears. You can try it by yourself.5
Let me propose once again a food analogy (it is the last one). In many societies
around the world, the variety of food that is available is limited in respect to what
one might find in an average town in the industrialized world. Similarly, the variety
of food that one can find in an average town in the industrialized world today is
bigger than the variety of food that was available, say, one hundred years ago. The
variety of possible foodstuff I can find in Milano, in Italy, close to my hometown, is
beyond my imagination.
Of course, geographical and economic factors are key in determining the food
one has access to and the food one cannot consume. Some foodstuffs are simply too
216 Conclusion

costly (truffles, fois gras, Kobe beef) to be purchased regularly—​or at all—​by most
of us. People living in an urban area have access to more varieties of products: even
if Sichuan food is delicious and affordable, you may not be able to buy the ingre-
dients needed to prepare it if you live in a small village in the Italian Alps. We do
not need to be too precise to follow the analogy. You would agree that, the average
person, say again in Milano, has access to more variety of foodstuffs today than a
century ago.
Is this a positive change? Is the diet of the average Milanese better now than a
century ago? Nobody knows, but what we know for sure is that more variety creates
more differences, for better or worse. One could argue that a century ago every-
body had access to more or less the same foodstuff, produced mainly locally, with
recipes elaborated and tested through several past generations. Continuing with
this imaginary representation of the evolution of food’s variety, today, instead, we
can potentially have access to a far more diverse range of different foods. As a con-
sequence, diversity increases. One can cook the majority of their meals at home,
using local and fresh products, find the nuts they need for their healthy breakfasts
anytime in the year, and, to add some variety, learn Japanese cooking in the mean-
time. Another one can decide to be vegan, or practise any another specific form
of diet restriction: the great majority of recipes proposed by any vegan website, in
the western world, would have not been realizable until a couple of decades ago.
Someone else can buy mainly ready-​prepared meals, or eat a slice of pizza from a
fast food chain at every lunch.
In this analogy, foodstuff is, not surprisingly, information. Today, many people—​
not all, as we discussed above—​can access an enormous variety of information.
Does this easy, cheap, and immediate access create an “informational elite” that
uses this information in an efficient and productive way, while others consume the
equivalent of the ready-​prepared meals? Can we think of a continuum where, on
one end, we have educated, digital-​smart, people who use social media and search
engines wisely, do not waste too much of their time with useless information, and
find everything they need quickly, in a way that was impossible until a few years
ago? But, on the other end of the continuum, people use the internet mainly to
share quotations written on sunset pictures (not that bad) or to add angry com-
ments to the clickbait articles of the day (worse)?
In a regulated informational system, everybody has access to more or less the
same information (the classic books, compulsory school, and so on), which is a
disadvantage for the informational elite, but is possibly a benefit for the others. As
for foodstuffs, the distinctions do not need to be too rigid: healthy-​eating people
do pass by McDonalds from time to time. As much as the availability of a great
variety of foodstuff advantages those with biological and cultural capital, such as
fit people, high-​metabolism people, and good cooks, and harms others, the avail-
ability of a great variety of information can be a double-​edged sword.
 Culture, recast  217

Research from Oxford’s Reuters Institute pointed out, in 2018, how social in-
equalities, in the United Kingdom, are stronger for the consumption of online
than of offline news, or how it is less likely that people with a lower income access
news directly, as opposed to accessing them through social media. An analogous
association between social and information inequality is found in a larger sample
of western countries: “more educated and wealthier subjects follow on average a
larger number of sources. On average, a high-​income man with a graduate degree
follows approximately twice as many sources as a low-​income woman with no col-
lege degree.”6
The exposure and, to a greater degree, the diffusion of fake news is limited to
a restricted number of people: on Twitter, during the 2016 US presidential elec-
tions, 1 percent of individuals accounted for the 80 percent of fake news exposure,
and the 0.1 percent (1 in 1,000) of individuals accounted for 80 percent of sharing.
In the same period, but on Facebook, researchers found a strong demographic ef-
fect: “on average, users over 65 shared nearly seven times as many articles from fake
news domains as the youngest age group.”7
In the food domain, it would be strange to limit the availability or the variety
of products; still we would like everybody to have access to and consume healthy
products. How can we do this in the case of online information? On the other
hand, whereas we are reasonably sure that some foodstuff is bad for us, I discussed
at length how the effects of accessing low-​quality information have been generally
overestimated: we could perhaps embrace this variety.

Culture, recast

More generally, I have defended here a view according to which we should not
think of culture as an entity and, especially, as an entity with causal powers. As sup-
porters of cultural attraction theory like to say: “culture is a property, not a thing.”8
Representations, artefacts, be they internet memes, ways to build a hammer, word
meanings or anything you can think about, are all to a different degree part of long
chains of information that has been transmitted among many individuals. For hu-
mans, nothing is completely cultural, and nothing is not cultural at all. Some traits
belong to wide and robust chains of social transmission, such as major Hindu de-
ities, Facebook, and the pizza, others to shorter ones, such as the jokes between me
and my daughter. Some traits belong to chains in which content is transmitted with
high fidelity and practically recreated ex-​novo each time, such as the rules to play
chess, others are more idiosyncratic and their links with previous socially trans-
mitted behaviors or ideas seem less important, such as the dream I had last night.
I became increasingly convinced that the distinction between social and individual
learning is similar to the distinction between acquired and innate. Intuitively
218 Conclusion

sound, probably useful for some modelling purposes, and fundamentally wrong
when applied to specific behaviors or items.
It is not problematic to call these things “cultural traits” as long as this does
not make us think they are special in some way. As mentioned, the interesting
part is to explain why certain representations and artefacts are sufficiently stable
and widespread to be called cultural in the first place. The idea that we copy with
high-​fidelity what others do, using quasi-​automatic heuristics to choose when,
what, and from whom to copy is appealing exactly because it provides an answer
to this question. However, it happens to not be a satisfactory answer. As we have
discussed at length, we do copy others, we do get influenced by social cues -​until
we do not: epidemics of broccoli eating among schoolchildren are unheard of. In
this sense, we cannot explain the success of an idea or of an artefact with “cultural
influence”.
There is an assortment of reasons why cultural traits become stable and wide-
spread. Copying the majority or copying the rich and famous are important drives,
but they do not act in a vacuum. Orchestrated campaigns of online persuasion—​
not to mention the infamous social media bots—​also do not act in a vacuum and,
from what we know, their effect is limited by several other factors. The conse-
quences of propaganda or even subliminal advertising fascinate laypeople every-
where: why this is the case is an interesting question in itself. They also fascinate
social scientists and psychologists but, to the best of our current knowledge, they
are routinely overestimated.9
Thus: there is no recipe for how to go viral. In this book I have been partial to
general cognitive tendencies, explored especially in the chapter Misinformation,
that make some cultural traits appealing or memorable, such as negative content,
information about threats, social information, and so on. However, as discussed,
these forces are generally weak, easily overcame by other tendencies, and their ef-
fect can be discerned only zooming-​out at large scale. There is no point in using
them to predict if a single, say, internet meme or news item will become successful
or not. On the other hand, they differ from social influence forces because their
direction is constant: social-​influence driven fashions are, by definition, examples
of shifting forces: mustard color is so passé (I guess). In a system where there are
strong but shifting forces and weak but constant ones, the latter will leave their sig-
nature, at least in the long run.
Sure, sometimes we “just do it” and copy others using simple heuristics, and this
can make spread behaviors that are smarter than us. There are, sometimes, good
motivations to copy others quasi-​automatically, especially when the ideas and be-
haviors are not costly or are consistent with other beliefs we have. There are no
reasons not to copy the majority when everyone uses clothes made of caribou skins
as opposed to bear skins, to use an example from the first chapters of this book, or
not to copy the way my grandmother makes lasagne. In other cases, transmission
needs to be supported by institutions: pupils do not automatically copy teachers’
 Culture, recast  219

skills, nor do they want to, even though these skills are useful and everybody re-
inforces the belief that they are important.
High-​fidelity transfer of new information by virtue of the mechanisms involved
in the process of cultural transmission does happen. It produces ideas, behaviors
or artefacts that would be statistically impossible for a single individual to invent
alone. This is essential to make human culture cumulative. However, high-​fidelity
transmission requires motivation and, in many cases, external “fidelity amplifiers,”
as we called them in the chapter Transmitting and sharing, to be effective.

I have been torn about whether to conclude with the following considerations—​in
fact, I was convinced not to—​but talking with many people has made me change
my mind. I have been surprised that many habits that I thought were common
sense, look puzzling to other people. My smartphone only goes into my bedroom
as an alarm clock, which I use infrequently. My wife is happy, at least I hope so, to
do the same with hers. Neither does it stay on our table when eating, and when the
classic “Google it!” situation happens in a conversation (“What was the name of the
spouse of Frida Kahlo?”; “Are there more types of cheese in France or in Italy?”)
I do my best to try to elaborate answers without picking it up. Of course, I do not
always succeed.
My three-​year old daughter has access to YouTube, but she is not left alone in
front of it for hours with auto play on. If anything, I discovered myself a few gems
in long days on the couch waiting for her last flu to pass, such as Walt Disney’s
Silly Symphonies. Yes, I also saw repetitive, algorithmically suggested, computer-​
curated animations. Skip them, and say “no” if your children want to watch them.
They will soon prefer something else.
Twitter is a great tool for staying updated with the research in my area and it
makes it possible to spread my work to a larger audience. It is often a good place to
explore scientific issues, to have interesting conversations, to exchange ideas. I have
been lucky enough to ask questions on Twitter, and to have had, within a couple of
hours, answers from the specialists in the fields I was enquiring about. But many
parts of this book have been thought out and elaborated on in a stroll, or simply in
a screen-​free train ride, and especially in offline, face-​to-​face, conversations with
friends and colleagues. There is some research backing up these common-​sense
pieces of advice. You may want to check it on Google Scholar. But then take a walk,
if you can, and see what your friends think about it.

Notes

1. https://​www.nytimes.com/​2018/​04/​10/​business/​dealbook/​mark-​zuckerberg-​congress-​
hearing.html
2. Przybylski and Weinstein (2017), Orben and Przybylski (2019)
220 Conclusion

3. https://​www.itu.int/​en/​ITU-​D/​Statistics/​Pages/​stat/​default.aspx, https://​www.itu.int/​
en/​ITU-​D/​Statistics/​Documents/​facts/​ICTFactsFigures2017.pdf
4. https://​www.nytimes.com/​2013/​06/​05/​us/​as-​vandals-​take-​to-​national-​parks-​some-​
point-​to-​social-​media.html, https://​www.thrillist.com/​eat/​portland/​stanichs-​closed-
​will-​it-​reopen-​burger-​quest
5. https://​en.wikipedia.org/​wiki/​Florida_​Man
6. Kalogeropoulos and Nielsen (2018), Kennedy and Prat (2017)
7. Grinberg et al. (2019), Guess et al. (2019)
8. Scott-​Phillips et al. (2018)
9. Mercier (2017)
 References

Acerbi, A., 2019. Cognitive attraction and online misinformation. Palgrave Communications,
5, p. 15.
Acerbi, A., Ghirlanda, S., and Enquist, M., 2014. Regulatory traits: cultural influences on cul-
tural evolution. In Stefano Cagnoni, Marco Mirolli, Marco Villani, Evolution, Complexity
and Artificial Life (pp. 135–​147). Springer, Berlin, Heidelberg.
Acerbi, A., Kendal, J., and Tehrani, J.J., 2017. Cultural complexity and demography: The case
of folktales. Evolution and Human Behavior, 38(4), pp. 474–​480.
Acerbi, A. and Mesoudi, A., 2015. If we are all cultural Darwinians what’s the fuss about?
Clarifying recent disagreements in the field of cultural evolution. Biology and Philosophy,
30(4), pp. 481–​503.
Acerbi, A. and Tehrani, J.J., 2018. Did Einstein really say that? Testing content versus context in
the cultural selection of quotations. Journal of Cognition and Culture, 18(3–​4), pp. 293–​311.
Acerbi, A., Tennie, C., and Mesoudi, A., 2016. Social learning solves the problem of narrow-​
peaked search landscapes: experimental evidence in humans. Royal Society Open Science,
3(9), p. 160215.
Acerbi, A., Tennie, C., and Nunn, C.L., 2011. Modeling imitation and emulation in con-
strained search spaces. Learning and Behavior, 39(2), pp. 104–​114.
Adamic, L.A., Lento, T.M., Adar, E., and Ng, P.C., 2016. Information evolution in social net-
works. In Proceedings of the ninth ACM international conference on web search and data
mining (pp. 473–​482). ACM, New York, NY.
Ahn, Y.Y., Ahnert, S.E., Bagrow, J.P., and Barabási, A.L., 2011. Flavor network and the prin-
ciples of food pairing. Scientific Reports, 1, p. 196.
Allcott, H. and Gentzkow, M., 2017. Social media and fake news in the 2016 election. Journal
of Economic Perspectives, 31(2), pp. 211–​236.
Allport, G.W. and Postman, L., 1946. An analysis of rumor. Public Opinion Quarterly, 10(4),
pp. 501–​517.
Amos, C., Holmes, G., and Strutton, D., 2008. Exploring the relationship between celeb-
rity endorser effects and advertising effectiveness: A quantitative synthesis of effect size.
International Journal of Advertising, 27(2), pp. 209–​234.
Anthony, D., Smith, S.W., and Williamson, T., 2009. Reputation and reliability in collective
goods: The case of the online encyclopedia Wikipedia. Rationality and Society, 21(3),
pp. 283–​306.
Asch, S.E., 1955. Opinions and social pressure. Scientific American, 193(5), pp. 31–​35.
Asch, S.E., 1956. Studies of independence and conformity: I. A minority of one against a
unanimous majority. Psychological monographs: General and applied, 70(9), p. 1–​70.
Atkisson, C., O’Brien, M.J., and Mesoudi, A., 2012. Adult learners in a novel environment
use prestige-​biased social learning. Evolutionary Psychology, 10(3), pp. 519–​537.
Aunger, R., 2000. Darwinizing culture. The status of memetics as a science. Oxford: Oxford
University Press.
Bakshy, E., Hofman, J.M., Mason, W.A., and Watts, D.J., 2011. Everyone’s an influ-
encer: Quantifying influence on Twitter. In Proceedings of the fourth ACM international
conference on Web search and data mining (pp. 65–​74). ACM. New York, NY.
222 References

Bakshy, E., Messing, S., and Adamic, L.A., 2015. Exposure to ideologically diverse news and
opinion on Facebook. Science, 348(6239), pp. 1130–​1132.
Banerjee, K., Haque, O.S., and Spelke, E.S., 2013. Melting lizards and crying mail-
boxes: Children’s preferential recall of minimally counterintuitive concepts. Cognitive sci-
ence, 37(7), pp. 1251–​1289.
Barberá, P., 2014. How social media reduces mass political polarization:  Evidence from
Germany, Spain, and the US. Job Market Paper, 46, New York University.
Barberá, P., Jost, J.T., Nagler, J., Tucker, J.A., and Bonneau, R., 2015. Tweeting from left to
right:  Is online political communication more than an echo chamber? Psychological
Science, 26(10), pp. 1531–​1542.
Barkow, J.H., O’Gorman, R., and Rendell, L., 2012. Are the new mass media subverting cul-
tural transmission? Review of General Psychology, 16(2), pp. 121–​133.
Baron, R.S., Vandello, J.A., and Brunsman, B., 1996. The forgotten variable in conformity
research: Impact of task importance on social influence. Journal of Personality and Social
Psychology, 71(5), p. 915–​927.
Bartneck, C. and Moltchanova, E., 2018. LEGO products have become more complex. PloS
One, 13(1), p. e0190651.
Baylis, P., Obradovich, N., Kryvasheyeu, Y., Chen, H., Coviello, L., Moro, E., Cebrian, M., and
Fowler, J.H., 2018. Weather impacts expressed sentiment. PloS One, 13(4), p. e0195750.
Bebbington, K., MacLeod, C., Ellison, T.M., and Fay, N., 2017. The sky is falling: Evidence
of a negativity bias in the social transmission of information. Evolution and Human
Behavior, 38(1), pp. 92–​101.
Bentley, R.A., Hahn, M.W., and Shennan, S.J., 2004. Random drift and culture change.
Proceedings of the Royal Society of London B: Biological Sciences, 271(1547), pp. 1443–​1450.
Bentley, R.A. and O’Brien, M.J., 2017. The acceleration of cultural change: From ancestors to
algorithms. Cambridge, MA: MIT Press.
Bentley, R.A. and Shennan, S.J., 2003. Cultural transmission and stochastic network growth.
American Antiquity, 68(3), pp. 459–​485.
Berger, J. and Milkman, K.L., 2012. What makes online content viral? Journal of Marketing
Research, 49(2), pp. 192–​205.
Bernstein, M.S., Monroy-​Hernández, A., Harry, D., André, P., Panovich, K., and Vargas,
G.G., 2011. 4chan and/​b: An Analysis of Anonymity and Ephemerality in a Large Online
Community. In Proceedings of the Fifth International AAAI Conference on Weblogs and
Social Media (pp. 50–​57). AAAI Press. Menlo Park, CA.
Bessi, A., Zollo, F., Del Vicario, M., Puliga, M., Scala, A., Caldarelli, G., Uzzi, B., and
Quattrociocchi, W., 2016. Users polarization on Facebook and YouTube. PloS One, 11(8),
p. e0159641.
Betsch, C., Brewer, N.T., Brocard, P., Davies, P., Gaissmaier, W., Haase, N., Leask, J.,
Renkewitz, F., Renner, B., Reyna, V.F., and Rossmann, C., 2012. Opportunities and chal-
lenges of Web 2.0 for vaccination decisions. Vaccine, 30(25), pp. 3727–​3733.
Blackmore, S., 1999. The meme machine, Oxford: Oxford University Press.
Blaine, T. and Boyer, P., 2018. Origins of sinister rumors: A preference for threat-​related ma-
terial in the supply and demand of information. Evolution and Human Behavior, 39(1),
pp.  67–​75.
Blancke, S., Van Breusegem, F., De Jaeger, G., Braeckman, J., and Van Montagu, M., 2015.
Fatal attraction: The intuitive appeal of GMO opposition. Trends in Plant Science, 20(7),
pp. 414–​418.
Blair, A., 2003. Reading strategies for coping with information overload ca. 1550–​1700.
Journal of the History of Ideas, 64(1), pp. 11–​28.
 References  223

Bloom, N., Jones, C.I., Van Reenen, J., and Webb, M., 2017. Are ideas getting harder to find?
(No. w23782). National Bureau of Economic Research.
Boesch, C. and Boesch, H., 1982. Optimisation of nut-​cracking with natural hammers by
wild chimpanzees. Behaviour, 83(3-​4), pp. 265–​286.
Bornmann, L. and Mutz, R., 2015. Growth rates of modern science: A bibliometric analysis
based on the number of publications and cited references. Journal of the Association for
Information Science and Technology, 66(11), pp. 2215–​2222.
Boudry, M., 2018, Replicate after reading: on the extraction and evocation of cultural infor-
mation. Biology and Philosophy, 33, p. 27.
Boxell, L., Gentzkow, M., and Shapiro, J.M., 2017. Greater Internet use is not associated with
faster growth in political polarization among US demographic groups. Proceedings of the
National Academy of Sciences, p. 201706588.
Boyd, R. and Richerson, P.J., 1988. Culture and the evolutionary process. Chicago,
IL: University of Chicago Press.
Boyd, R. and Richerson, P.J., 2000. Memes:  Universal acid or a better mousetrap. In:  R.
Aunger, 2000, Darwinizing culture: The status of memetics as a science (Oxford: Oxford
University Press).
Boyd, R., Richerson, P.J., and Henrich, J., 2011. The cultural niche: Why social learning is
essential for human adaptation. Proceedings of the National Academy of Sciences, 108, pp.
10918–​10925.
Boyer, P., 1994. The naturalness of religious ideas: A cognitive theory of religion. Berkeley,
CA: University of California Press.
Boyer, P., 2018. Minds make societies: How cognition explains the world humans create. New
Haven, CT: Yale University Press.
Boyer, P. and Parren, N., 2015. Threat-​related information suggests competence: A possible
factor in the spread of rumors. PloS One, 10(6), p. e0128421.
Bridges, J. and Vásquez, C., 2018. If nearly all Airbnb reviews are positive, does that make
them meaningless?. Current Issues in Tourism, 21(18), pp. 2057–​2075.
Bridle, J., 2018. New dark age:  Technology and the end of the future. New  York,
NY: Verso Books.
Brodie, R., 1996. Virus of the mind: The new science of the meme. Carlsbad, CA: Hay House.
Burman, J.T., 2012. The misunderstanding of memes: Biography of an unscientific object,
1976–​1999. Perspectives on Science, 20(1), pp. 75–​104.
Burton-​Chellew, M.N. and Dunbar, R.I., 2015. Romance and reproduction are socially
costly. Evolutionary Behavioral Sciences, 9(4), p. 229–​241
Bush, A.J., Martin, C.A., and Bush, V.D., 2004. Sports celebrity influence on the behavioral
intentions of generation Y. Journal of Advertising Research, 44(1), pp. 108–​118.
Carmody, R.N. and Wrangham, R.W., 2009. The energetic significance of cooking. Journal
of Human Evolution, 57(4), pp. 379–​391.
Cerridwen, A. and Simonton, D.K., 2009. Sex doesn’t sell—​nor impress! Content, box office,
critics, and awards in mainstream cinema. Psychology of Aesthetics, Creativity, and the
Arts, 3(4), pp. 200–​210.
Cha, M., Haddadi, H., Benevenuto, F., and Gummadi, P.K., 2010. Measuring user
influence in twitter: The million follower fallacy. In Proceedings of the Fourth International
AAAI Conference on Weblogs and Social Media (pp. 10–​ 17). AAAI Press. Menlo
Park, CA.
Charbonneau, M., 2018. Understanding cultural fidelity. The British Journal for the
Philosophy of Science, advance article:  https://​academic.oup.com/​bjps/​advance-​article-​
abstract/​doi/​10.1093/​bjps/​axy052/​5065471
224 References

Chetty, R., Hendren, N., Kline, P., and Saez, E., 2014. Where is the land of opportunity? The
geography of intergenerational mobility in the United States. The Quarterly Journal of
Economics, 129(4), pp. 1553–​1623.
Chua, A.Y. and Banerjee, S., 2013. Reliability of reviews on the Internet:  The case of
Tripadvisor. In S.I. Ao, W.S. Grundfest, J. Burgstone, World Congress on Engineering and
Computer Science (pp. 453–​457). Newswood Limited. Hong Kong.
Chudek, M., Heller, S., Birch, S., and Henrich, J., 2012. Prestige-​ biased cultural
learning:  Bystander’s differential attention to potential models influences children’s
learning. Evolution and Human Behavior, 33(1), pp. 46–​56.
Claidière, N., Bowler, M., Brookes, S., Brown, R., and Whiten, A., 2014. Frequency of be-
havior witnessed and conformity in an everyday social context. PloS One, 9(6), p. e99874.
Claidière, N., Bowler, M., and Whiten, A., 2012. Evidence for weak or linear conformity
but not for hyper-​conformity in an everyday social learning context. PLoS One, 7(2), p.
e30970.
Claidière, N., Scott-​Phillips, T.C., and Sperber, D., 2014. How Darwinian is cultural evo-
lution? Philosophical Transactions of the Royal Society B: Biological Sciences, 369(1642),
p. 20130368.
Clasen, M., 2012. Attention, predation, counterintuition:  Why Dracula won’t die. Style,
46(3-​4), pp. 378–​398.
Colman, I., Kingsbury, M., Weeks, M., Ataullahjan, A., Bélair, M.A., Dykxhoorn, J.,
Hynes, K., Loro, A., Martin, M.S., Naicker, K., and Pollock, N., 2014. CARTOONS
KILL: Casualties in animated recreational theater in an objective observational new study
of kids’ introduction to loss of life. BMJ, 349, p. g7184.
Conover, M., Ratkiewicz, J., Francisco, M.R., Gonçalves, B., Menczer, F., and Flammini, A.,
2011. Political polarization on Twitter. Icwsm, 133, pp. 89–​96.
Coultas, J.C., 2004. When in Rome . . . an evolutionary perspective on conformity. Group
Processes and Intergroup Relations, 7(4), pp. 317–​331.
Craik, H. ed., 1916, English Prose. New York, NY: The Macmillan Company.
Danchin, E., Nöbel, S., Pocheville, A., Dagaeff, A.C., Demay, L., Alphand, M., Ranty-​
Roby, S., van Renssen, L., Monier, M., Gazagne, E., and Allain, M., 2018. Cultural
flies: Conformist social learning in fruitflies predicts long-​lasting mate-​choice traditions.
Science, 362(6418), pp. 1025–​1030.
Davis, C.A., Ciampaglia, G.L., Aiello, L.M., Chung, K., Conover, M.D., Ferrara, E.,
Flammini, A., Fox, G.C., Gao, X., Gonçalves, B., and Grabowicz, P.A., 2016. OSoMe: The
IUNI observatory on social media. PeerJ Computer Science, 2, p. e87.
Davison, W.P., 1983. The third-​person effect in communication. Public Opinion Quarterly,
47(1), pp. 1–​15.
Dawkins, R., 1976. The selfish gene, Oxford: Oxford University Press.
Dawkins, R., 1999. Foreword to The Meme Machine by Susan Blackmore, Oxford: Oxford
University Press.
Dean, L.G., Vale, G.L., Laland, K.N., Flynn, E., and Kendal, R.L., 2014. Human cumulative
culture: A comparative perspective. Biological Reviews, 89(2), pp. 284–​301.
De Barra, M., 2017. Reporting bias inflates the reputation of medical treatments: A com-
parison of outcomes in clinical trials and online product reviews. Social Science and
Medicine, 177, pp. 248–​255.
Del Vicario, M., Bessi, A., Zollo, F., Petroni, F., Scala, A., Caldarelli, G., Stanley, H.E., and
Quattrociocchi, W., 2016. The spreading of misinformation online. Proceedings of the
National Academy of Sciences, 113(3), pp. 554–​559.
 References  225

Derex, M., Beugin, M.P., Godelle, B., and Raymond, M., 2013. Experimental evidence for
the influence of group size on cultural complexity. Nature, 503(7476), pp. 389–​391.
Derex, M., Bonnefon, J.F., Boyd, R., Mesoudi, A., and Exeter, P.T., 2019. Causal under-
standing is not necessary for the improvement of culturally evolving technology. Nature
Human Behaviour, 3, pp. 446–​452.
Derex, M., Feron, R., Godelle, B., and Raymond, M., 2015. Social learning and the replica-
tion process: an experimental investigation. Proc. R. Soc. B, 282(1808), p. 20150719.
DeVito, M.A., Gergle, D., and Birnholtz, J., 2017. Algorithms ruin everything: #RIPTwitter,
folk theories, and resistance to algorithmic change in social media. In: Proceedings of the
2017 CHI Conference on Human Factors in Computing Systems (pp. 3163–​3174). ACM,
New York, NY.
Dingemanse, M., Roberts, S.G., Baranova, J., Blythe, J., Drew, P., Floyd, S., Gisladottir, R.S.,
Kendrick, K.H., Levinson, S.C., Manrique, E., and Rossi, G., 2015. Universal principles in
the repair of communication problems. PloS One, 10(9), p. e0136100.
Djafarova, E. and Rushworth, C., 2017. Exploring the credibility of online celebri-
ties’ Instagram profiles in influencing the purchase decisions of young female users.
Computers in Human Behavior 68, pp. 1–​7.
Dubois, E. and Blank, G., 2018. The echo chamber is overstated: The moderating effect of
political interest and diverse media. Information, Communication and Society, 21(5), pp.
729–​745.
Dunbar, R.I., 1993. Coevolution of neocortical size, group size and language in humans.
Behavioral and Brain Sciences, 16(4), pp. 681–​694.
Dunbar, R.I., 1998. The social brain hypothesis. Evolutionary Anthropology, 6(5), pp.
178–​190.
Dunbar, R.I., 2016. Do online social media cut through the constraints that limit the size of
offline social networks? Royal Society Open Science, 3(1), p. 150292.
Dunbar, R.I., 2018. The anatomy of friendship. Trends in Cognitive Sciences, 22(1), pp.  32–​51.
Dunbar, R.I., Arnaboldi, V., Conti, M., and Passarella, A., 2015. The structure of online so-
cial networks mirrors those in the offline world. Social Networks, 43, pp. 39–​47.
Dunbar, R.I. and Sosis, R., 2018. Optimising human community sizes. Evolution and Human
Behavior, 39(1), pp.106–​111.
Efferson, C., Lalive, R., Richerson, P.J., McElreath, R., and Lubell, M., 2008. Conformists and
mavericks:  the empirics of frequency-​dependent cultural transmission. Evolution and
Human Behavior, 29(1), pp. 56–​64.
Efferson, C., Richerson, P.J., McElreath, R., Lubell, M., Edsten, E., Waring, T.M., Paciotti, B.,
and Baum, W., 2007. Learning, productivity, and noise: An experimental study of cultural
transmission on the Bolivian Altiplano. Evolution and Human Behavior, 28(1), pp. 1–​17.
Eisenstein, E., 1979. The printing press as an agent of change, volume I:  Communications
and cultural transformations in early-​ modern Europe, Cambridge:  Cambridge
University Press.
Enquist, M., Eriksson, K., and Ghirlanda, S., 2007. Critical social learning: a solution to
Rogers’s paradox of nonadaptive culture. American Anthropologist, 109(4), pp. 727–​734.
Erdogan, B.Z., 1999. Celebrity endorsement:  A literature review. Journal of Marketing
Management, 15(4), pp. 291–​314.
Erel, I., Stern, L.H., Tan, C., and Weisbach, M.S., 2018. Selecting directors using machine
learning (No. w24435). National Bureau of Economic Research.
Eriksson, K. and Coultas, J.C., 2009. Are people really conformist-​biased? An empirical test
and a new mathematical model. Journal of Evolutionary Psychology, 7(1), pp. 5–​21.
226 References

Eriksson, K. and Coultas, J.C., 2014. Corpses, maggots, poodles and rats:  Emotional se-
lection operating in three phases of cultural transmission of urban legends. Journal of
Cognition and Culture, 14(1-​2), pp.  1–​26.
Eriksson, K., Coultas, J.C., and De Barra, M., 2016. Cross-​cultural differences in emotional
selection on transmission of information. Journal of Cognition and Culture, 16(1-​2), pp.
122–​143.
Evans, D.G., Barwell, J., Eccles, D.M., Collins, A., Izatt, L., Jacobs, C., Donaldson, A., Brady,
A.F., Cuthbert, A., Harrison, R., and Thomas, S., 2014. The Angelina Jolie effect: How
high celebrity profile can have a major impact on provision of cancer related services.
Breast Cancer Research, 16(5), p. 442.
Evans, D.G., Wisely, J., Clancy, T., Lalloo, F., Wilson, M., Johnson, R., Duncan, J., Barr,
L., Gandhi, A., and Howell, A., 2015. Longer term effects of the Angelina Jolie ef-
fect:  Increased risk-​reducing mastectomy rates in BRCA carriers and other high-​risk
women. Breast Cancer Research, 17(1), p. 143.
Faulkes, Z., 2014. The vacuum shouts back: Postpublication peer review on social media.
Neuron, 82(2), pp. 258–​260.
Feld, S.L., 1991. Why your friends have more friends than you do. American Journal of
Sociology, 96(6), pp. 1464–​1477.
Fessler, D.M., Pisor, A.C., and Navarrete, C.D., 2014. Negatively-​biased credulity and the
cultural evolution of beliefs. PloS One, 9(4), e95167.
Ferris, K.O., 2010. The next big thing: Local celebrity. Society, 47(5), pp. 392–​395.
Flaxman, S., Goel, S., and Rao, J.M., 2016. Filter bubbles, echo chambers, and online news
consumption. Public Opinion Quarterly, 80(S1), pp. 298–​320.
Fletcher, R., Cornia, A., Graves, L., and Nielsen, R.K., 2018. Measuring the reach of “fake
news” and online disinformation in Europe. Reuters Institute Factsheet.
Fletcher, R. and Nielsen, R.K., 2017. Are news audiences increasingly fragmented? A cross‐
national comparative analysis of cross‐platform news audience fragmentation and dupli-
cation. Journal of Communication, 67(4), pp. 476–​498.
Fletcher, R. and Nielsen, R.K., 2018. Are people incidentally exposed to news on social
media? A comparative analysis. New Media and Society, 20(7), pp. 2450–​2468.
Fletcher, R., Radcliffe, D., Levy, D., Nielsen, R.K., and Newman, N., 2015. Reuters Institute
digital news report 2015: Supplementary report. https://​reutersinstitute.politics.ox.ac.uk/​
sites/​default/​files/​2017-​06/​Supplementary%20Digital%20News%20Report%202015.pdf
Fischer, F., Böttinger, K., Xiao, H., Stransky, C., Acar, Y., Backes, M., and Fahl, S., 2017, May.
Stack overflow considered harmful? The impact of copy&paste on android application
security. In Security and Privacy (SP), 2017 IEEE Symposium on (pp. 121–​136). IEEE
Computer Society. Los Alamitos, CA.
Fuchs, B., Sornette, D., and Thurner, S., 2014. Fractal multi-​level organisation of human
groups in a virtual world. Scientific Reports, 4, p. 6526.
Galef, B.G., 1992. The question of animal culture. Human Nature, 3(2), pp. 157–​178.
Garfield, Z.H., Garfield, M.J., and Hewlett, B.S., 2016. A cross-​cultural analysis of hunter-​
gatherer social learning. In Hideaki Terashima, Barry. S. Hewlett, Social learning and in-
novation in contemporary hunter-​gatherers (pp. 19–​34). Springer, Tokyo.
Garimella, K. and Weber, I., 2017. A long-​term analysis of polarization on Twitter. arXiv pre-
print arXiv:1703.02769.
Gawande, A., 2009. The checklist manifesto:  How to get things right. New  York, NY:
Metropolitan Books.
Gazzaley, A. and Rosen, L.D., 2016. The distracted mind: Ancient brains in a high-​tech world.
Cambridge, MA: MIT Press.
 References  227

Gaumont, N., Panahi, M., and Chavalarias, D., 2017. Methods for the reconstruction of
the socio-​semantic dynamics of political activist Twitter networks. PloS One, 13(9),
p. e0201879.
Gentzkow, M. and Shapiro, J.M., 2011. Ideological segregation online and offline. The
Quarterly Journal of Economics, 126(4), pp. 1799–​1839.
Gergely, G. and Csibra, G., 2006. Sylvia’s recipe: The role of imitation and pedagogy in the
transmission of cultural knowledge. Roots of Human Sociality: Culture, Cognition, and
Human Interaction, pp. 229–​255.
Gjesfjeld, E., Chang, J., Silvestro, D., Kelty, C. and Alfaro, M., 2016. Competition and
extinction explain the evolution of diversity in American automobiles. Palgrave
Communications, 2, p. 16019.
Ghirlanda, S., Acerbi, A., and Herzog, H., 2014. Dog movie stars and dog breed popu-
larity: A case study in media influence on choice. PLoS One, 9(9), p. e106565.
Glowacki, L. and Molleman, L., 2017. Subsistence styles shape human social learning strat-
egies. Nature Human Behaviour, 1, p. 0098.
Goldenberg, J., Han, S., Lehmann, D.R., and Hong, J.W., 2009. The role of hubs in the adop-
tion process. Journal of Marketing, 73(2), pp. 1–​13.
Goldenberg, J. and Levy, M., 2009. Distance is not dead: Social interaction and geographical
distance in the internet era. arXiv preprint arXiv:0906.3202.
Golder, S.A. and Macy, M.W., 2011. Diurnal and seasonal mood vary with work, sleep, and
daylength across diverse cultures. Science, 333(6051), pp. 1878–​1881.
Gonçalves, B., Perra, N. and Vespignani, A., 2011. Modeling users’ activity on twitter net-
works: Validation of Dunbar’s number. PloS One, 6(8), p. e22656.
Goody, J., 1977. The domestication of the savage mind. Cambridge: Cambridge University
Press.
Greenhill, S.J., Hua, X., Welsh, C.F., Schneemann, H., and Bromham, L., 2018. Population
size and the rate of language evolution: A test across Indo-​European, Austronesian and
Bantu languages. Frontiers in Psychology, 9, p. 576.
Grinberg, N., Joseph, K., Friedland, L., Swire-​Thompson, B., and Lazer, D., 2019. Fake
news on Twitter during the 2016 US presidential election. Science, 363(6425),
pp. 374–​378.
Guess, A., Nagler, J., and Tucker, J., 2019. Less than you think: Prevalence and predictors of
fake news dissemination on Facebook. Science Advances, 5(1), p. eaau4586.
Guess, A., Nyhan, B. and Reifler, J., 2018. “Selective exposure to misinformation: Evidence
from the consumption of fake news during the 2016 US presidential campaign.” European
Research Council.
Hackforth, R. (ed.), 1972. Plato: Phaedrus. Cambridge: Cambridge University Press.
Hagen, E.H. and Hammerstein, P., 2006. Game theory and human evolution: A critique of
some recent interpretations of experimental games. Theoretical Population Biology, 69(3),
pp. 339–​348.
Haines, R., Hough, J., Cao, L., and Haines, D., 2014. Anonymity in computer-​mediated com-
munication: More contrarian ideas with less influence. Group Decision and Negotiation,
23(4), pp. 765–​786.
Hamilton, M.J., Milne, B.T., Walker, R.S., Burger, O., and Brown, J.H., 2007. The complex
structure of hunter-​gatherer social networks. Proceedings of the Royal Society of London
B: Biological Sciences, 274(1622), pp. 2195–​2203.
Hamari, J., Koivisto, J., and Sarsa, H., 2014, Does gamification work? A literature review of
empirical studies on gamification. In 2014 47th Hawaii international conference on system
sciences (HICSS) (pp. 3025–​3034). IEEE Computer Society. Los Alamitos, CA.
228 References

Haslam, N., Loughnan, S., and Perry, G., 2014. Meta-​Milgram: An empirical synthesis of the
obedience experiments. PloS One, 9(4), p. e93927.
Haslam, M., Hernandez-​Aguilar, R.A., Proffitt, T., Arroyo, A., Falótico, T., Fragaszy, D.,
Gumert, M., Harris, J.W., Huffman, M.A., Kalan, A.K., and Malaivijitnond, S., 2017.
Primate archaeology evolves. Nature, Ecology and Evolution, 1(10), pp. 1431–​1437.
Henrich, J., 2004. Demography and cultural evolution:  how adaptive cultural processes
can produce maladaptive losses—​the Tasmanian case. American Antiquity, 69(2), pp.
197–​214.
Heath C., Bell C., and Sternberg E., 2001. Emotional selection in memes: The case of urban
legends. Journal of Personality and Social Psychology, 81(6), pp. 1028–​1041.
Henrich, J., 2015. The secret of our success: how culture is driving human evolution, domesticating
our species, and making us smarter. Princeton, NJ: Princeton University Press.
Henrich, J. and Boyd, R., 1998. The evolution of conformist transmission and the emergence
of between-​group differences. Evolution and Human Behavior, 19(4), pp. 215–​241.
Henrich, J. and Boyd, R., 2002. On modeling cognition and culture: Why cultural evolution
does not require replication of representations. Journal of Cognition and Culture, 2(2), pp.
87–​112.
Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H., and McElreath, R., 2001.
In search of homo economicus:  Behavioral experiments in 15 small-​scale societies.
American Economic Review, 91(2), pp. 73–​78.
Henrich, J., Boyd, R., Derex, M., Kline, M.A., Mesoudi, A., Muthukrishna, M., Powell, A.T.,
Shennan, S.J., and Thomas, M.G., 2016. Understanding cumulative cultural evolution.
Proceedings of the National Academy of Sciences, 113(44), pp. E6724–​E6725.
Henrich, J. and Broesch, J., 2011. On the nature of cultural transmission networks: Evidence
from Fijian villages for adaptive learning biases. Philosophical Transactions of the Royal
Society of London B: Biological Sciences, 366(1567), pp. 1139–​1148.
Henrich, J. and Gil-​White, F.J., 2001. The evolution of prestige: Freely conferred deference as
a mechanism for enhancing the benefits of cultural transmission. Evolution and Human
Behavior, 22(3), pp. 165–​96.
Henrich, J. and McElreath, R., 2003. The evolution of cultural evolution. Evolutionary
Anthropology: Issues, News, and Reviews, 12(3), pp. 123–​135.
Herzog, H.A., Bentley, R.A., and Hahn, M.W., 2004. Random drift and large shifts in
popularity of dog breeds. Proceedings of the Royal Society of London. Series B: Biological
Sciences, 271(suppl_​5), pp. S353–​S356.
Hester, J.B. and Gibson, R., 2003. The economy and second-​level agenda setting: A time-​
series analysis of economic news and public opinion about the economy. Journalism and
Mass Communication Quarterly, 80(1), pp. 73–​90.
Hewlett, B.S. and Cavalli‐Sforza, L.L., 1986. Cultural transmission among Aka pygmies.
American Anthropologist, 88(4), pp. 922–​934.
Heyes, C., 2012. What’s social about social learning? Journal of Comparative Psychology,
126(2), p. 193–​202.
Heyes, C., 2016. Blackboxing: social learning strategies and cultural evolution. Philosophical
Transactions of the Royal Society B: Biological Sciences, 371(1693), p. 20150369.
Hodas, N.O. and Lerman, K., 2014. The simple rules of social contagion. Scientific reports,
4, p. 4343.
Hodges, B.H. and Geyer, A.L., 2006. A nonconformist account of the Asch experi-
ments:  Values, pragmatics, and moral dilemmas. Personality and Social Psychology
Review, 10(1), pp. 2–​19.
 References  229

Hollander, M.M. and Turowetz, J., 2017. Normalizing trust: Participants’ immediately post‐
hoc explanations of behaviour in Milgram’s ‘obedience’ experiments. British Journal of
Social Psychology, 56(4), pp. 655–​674.
Hosokawa, S., 1984. The Walkman effect. Popular Music, 4, pp. 165–​180.
Hu, N., Zhang, J., and Pavlou, P.A., 2009. Overcoming the J-​shaped distribution of product
reviews. Communications of the ACM, 52(10), pp. 144–​147.
Huberman, B.A. and Adamic, L.A., 1999. The nature of markets in the World Wide Web
(No. 521). Society for Computational Economics.
Hull, D.L., 1988. Interactors versus vehicles. In: The role of behavior in evolution, (Cambridge,
MA: MIT Press).
Jacquet, P.O., Safra, L., Wyart, V., Baumard, N., and Chevallier, C., 2019. The ecological roots
of human susceptibility to social influence: a pre-​registered study investigating the im-
pact of early-​life adversity. Royal Society Open Science, 6(1), p. 180454.
Jimenéz, A.V. and Mesoudi, A., 2019. Prestige biased social learning: Current evidence and
outstanding questions, Palgrave Communications, 5, p. 20.
Jin, S.A.A. and Phua, J., 2014. Following celebrities’ tweets about brands:  The impact of
twitter-​based electronic word-​of-​mouth on consumers’ source credibility perception,
buying intention, and social identification with celebrities. Journal of Advertising, 43(2),
pp. 181–​195.
Jo, H.H., Saramäki, J., Dunbar, R.I., and Kaski, K., 2014. Spatial patterns of close relation-
ships across the lifespan. Scientific Reports, 4, p. 6988.
Kalogeropoulos, A. and Nielsen, R.K., 2018. Social inequalities in news consumption.
Factsheet, News Media Digital Media, 461, p. 475.
Keil, F.C., 1992. Concepts, kinds, and cognitive development. Cambridge, MA: MIT Press.
Kelly, K., 2017. The inevitable: understanding the 12 technological forces that will shape our
future. London, Penguin.
Kendal, R.L., Boogert, N.J., Rendell, L., Laland, K.N., Webster, M., and Jones, P.L., 2018.
Social learning strategies: Bridge-​building between fields. Trends in Cognitive Sciences,
22(7), pp. 651–​665.
Kirby, S., Cornish, H., and Smith, K., 2008. Cumulative cultural evolution in the la-
boratory:  An experimental approach to the origins of structure in human language.
Proceedings of the National Academy of Sciences. 105(31), pp. 10681–​10686.
Kirschenbaum, M.G., 2016. Track changes. Harvard, MA: Harvard University Press.
Kline, M.A. and Boyd, R., 2010. Population size predicts technological complexity in
Oceania. Proceedings of the Royal Society of London B: Biological Sciences, 277(1693), pp.
2559–​2564.
Knoll, J. and Matthes, J., 2017. The effectiveness of celebrity endorsements: A meta-​analysis.
Journal of the Academy of Marketing Science, 45(1), pp. 55–​75.
Kohler, T.A., VanBuskirk, S., and Ruscavage-​Barz, S., 2004. Vessels and villages: Evidence
for conformist transmission in early village aggregations on the Pajarito Plateau, New
Mexico. Journal of Anthropological Archaeology, 23(1), pp. 100–​118.
Laland, K.N., 2004. Social learning strategies. Animal Learning and Behavior, 32(1),
pp.  4–​14.
Laland, K.N., 2018. Darwin’s unfinished symphony:  how culture made the human mind.
Princeton, NJ: Princeton University Press.
Lazer, D., 2015. The rise of the social algorithm. Science, 348(6239), pp. 1090–​1091.
Lerique, S. and Roth, C., 2018. The Semantic Drift of Quotations in Blogspace: A Case Study
in Short‐Term Cultural Evolution. Cognitive Science, 42(1), pp. 188–​219.
230 References

Levy, A., Salamon, A., Tucci, M., Limebeer, C.L., Parker, L.A., and Leri, F., 2013. Co‐sensi-
tivity to the incentive properties of palatable food and cocaine in rats; Implications for
co‐morbid addictions. Addiction Biology, 18(5), pp. 763–​773.
Lewis, M.P., 2009. Ethnologue: Languages of the world. SIL international.
Lieberson, S., 2000. A matter of taste: How names, fashions, and culture change. New Heaven,
CT: Yale University Press.
Lin, Y., Michel, J.B., Aiden, E.L., Orwant, J., Brockman, W., and Petrov, S., 2012. Syntactic
annotations for the Google Books Ngram Corpus. In Proceedings of the ACL 2012
System Demonstrations (pp. 169–​ 174). Association for Computational Linguistics,
Stroudsburg, PA.
Logan, F.A., 1999. Errors in copy typewriting. Journal of Experimental Psychology: Human
Perception and Performance, 25(6), p. 1760.
Lynch, A., 1996. Thought contagion: How belief spreads through society: The new science of
memes. New York, NY: Basic Books.
Lynch, M.P., 2016. The internet of us: Knowing more and understanding less in the age of big
data. New York, NY: WW Norton & Company.
Luca, M. and Zervas, G., 2016. Fake it till you make it: Reputation, competition, and Yelp
review fraud. Management Science, 62(12), pp. 3412–​3427.
Lupyan, G. and Dale, R., 2010. Language structure is partly determined by social structure.
PloS One, 5(1), p. e8559.
Madison, G. and Schiölde, G., 2017. Repeated listening increases the liking for music re-
gardless of its complexity:  Implications for the appreciation and aesthetics of music.
Frontiers in Neuroscience, 11, p. 147.
Mawdsley, S.E., 2016. ‘Salk Hops’: Teen health activism and the fight against polio, 1955–​
1960. Cultural and Social History, 13(2), pp. 249–​265.
Mayr, E., 1982. The growth of biological thought:  Diversity, evolution, and inheritance.
Cambridge, MA: Harvard University Press.
Mayzlin, D., Dover, Y., and Chevalier, J., 2014. Promotional reviews: An empirical investiga-
tion of online review manipulation. American Economic Review, 104(8), pp. 2421–​2455.
McClure, C.C., Cataldi, J.R., and O’Leary, S.T., 2017. Vaccine hesitancy: Where we are and
where we are going. Clinical Therapeutics, 39(8), pp. 1550–​1562.
McCormick, K., 2016. Celebrity endorsements: Influence of a product-​endorser match on
Millennials attitudes and purchase intentions. Journal of Retailing and Consumer Services,
32, pp. 39–​45.
McElreath, R., Bell, A.V., Efferson, C., Lubell, M., Richerson, P.J., and Waring, T., 2008.
Beyond existence and aiming outside the laboratory: estimating frequency-​dependent
and pay-​off-​biased social learning strategies. Philosophical Transactions of the Royal
Society of London B: Biological Sciences, 363(1509), pp. 3515–​3528.
McElreath, R., Lubell, M., Richerson, P.J., Waring, T.M., Baum, W., Edsten, E., Efferson, C.,
and Paciotti, B., 2005. Applying evolutionary models to the laboratory study of social
learning. Evolution and Human Behavior, 26(6), pp. 483–​508.
McGee, J., Caverlee, J.A., and Cheng, Z., 2011. A geographic study of tie strength in social
media. In Proceedings of the 20th ACM international conference on information and know-
ledge management (pp. 2333–​2336). ACM. Bettina Berendt, Arjen de Vries, Wenfei Fan,
Craig Macdonald, Iadh Ounis, and Ian Rutven, ACM. New York, NY.
McGuigan, N., Makinson, J., and Whiten, A., 2011. From over‐imitation to super‐
copying: Adults imitate causally irrelevant aspects of tool use with higher fidelity than
young children. British Journal of Psychology, 102(1), pp. 1–​18.
 References  231

McLuhan, M., 1964, Understanding media:  The extensions of man. Cambridge,

MA: MIT Press.
Menon, M.K., Boone, L.E., and Rogers, H.P., 2001. Celebrity advertising: An assessment of
its relative effectiveness. In Proceedings of the Society for Marketing Advances Conference,
New Orleans, Louisiana.
Mercier, H., 2017. How gullible are we? A review of the evidence from psychology and social
science. Review of General Psychology, 21(2), pp. 103–​122.
Mercier, H., Majima, Y., and Miton, H., 2018. Willingness to transmit and the spread of
pseudoscientific beliefs. Applied Cognitive Psychology, 32(4), pp. 499–​505.
Mercier, H. and Sperber, D., 2011. Why do humans reason? Arguments for an argumenta-
tive theory. Behavioral and Brain Sciences, 34(2), pp. 57–​74.
Mercier, H. and Sperber, D., 2017. The enigma of reason. Cambridge, MA:  Harvard
University Press.
Meredith, J. and Stokoe, E., 2014. Repair: Comparing Facebook “chat” with spoken inter-
action. Discourse and Communication, 8(2), pp. 181–​207.
Mesoudi, A., 2008. An experimental simulation of the “copy-​successful-​individuals” cul-
tural learning strategy: adaptive landscapes, producer–​scrounger dynamics, and infor-
mational access costs. Evolution and Human Behavior, 29(5), pp. 350–​363.
Mesoudi, A., 2009. The cultural dynamics of copycat suicide. PLoS One, 4(9), p. e7252.
Mesoudi, A., 2011a. Cultural evolution: How Darwinian theory can explain human culture
and synthesize the social sciences. Chicago, IL: University of Chicago Press.
Mesoudi, A., 2011b. An experimental comparison of human social learning strat-
egies:  payoff-​biased social learning is adaptive but underused. Evolution and Human
Behavior, 32(5), pp. 334–​342.
Mesoudi, A., 2014. Experimental studies of modern human social and individual learning
in an archaeological context: People behave adaptively, but within limits. In Akazawa,
T., Nishkiaki, Y., Aoki, K., Dynamics of learning in neanderthals and modern humans
Volume 2 (pp. 65–​76). Springer, Tokyo.
Mesoudi, A., Chang, L., Dall, S.R., and Thornton, A., 2016. The evolution of individual and
cultural variation in social learning. Trends in Ecology and Evolution, 31(3), pp. 215–​225.
Mesoudi, A., Chang, L., Murray, K., and Lu, H.J., 2015. Higher frequency of social learning
in China than in the West shows cultural variation in the dynamics of cultural evolution.
Proceedings of the Royal Society B: Biological Sciences, 282(1798), p. 20142209.
Mesoudi, A. and Lycett, S.J., 2009. Random copying, frequency-​dependent copying and cul-
ture change. Evolution and Human Behavior, 30(1), pp. 41–​48.
Mesoudi, A. and Thornton, A., 2018. What is cumulative cultural evolution? Proceedings of
the Royal Society B, 285(1880), p. 20180712.
Mesoudi, A., Whiten, A., and Dunbar, R., 2006. A bias for social information in human cul-
tural transmission. British Journal of Psychology, 97(3), pp. 405–​423.
Milgram, S., 1963. Behavioral study of obedience. The Journal of Abnormal and Social
Psychology, 67(4), p. 371.
Miton, H. and Mercier, H., 2015. Cognitive obstacles to pro-​vaccination beliefs. Trends in
Cognitive Sciences, 19(11), pp. 633–​636.
Molleman, L. and Gächter, S., 2018. Societal background influences social learning in co-
operative decision making. Evolution and Human Behavior, 39(5), pp. 547–​555.
Möller, J., Trilling, D., Helberger, N. and van Es, B., 2018. Do not blame it on the algo-
rithm: An empirical assessment of multiple recommender systems and their impact on
content diversity. Information, Communication and Society, 21(7), pp. 959–​977.
232 References

Morgan, T.J.H., Rendell, L.E., Ehn, M., Hoppitt, W., and Laland, K.N., 2012. The evolu-
tionary basis of human social learning. Proceedings of the Royal Society B:  Biological
Sciences, 279(1729), pp. 653–​662.
Morin, O., 2013. How portraits turned their eyes upon us: Visual preferences and demo-
graphic change in cultural evolution. Evolution and Human Behavior, 34(3), pp. 222–​229.
Morin, O., 2016. How traditions live and die. Oxford: Oxford University Press.
Morin, O., 2018. Spontaneous emergence of legibility in writing systems: The case of orien-
tation anisotropy. Cognitive Science, 42(2), pp. 664–​677.
Morin, O. and Acerbi, A., 2017. Birth of the cool: A two-​centuries decline in emotional ex-
pression in Anglophone fiction. Cognition and Emotion, 31(8), pp. 1663–​1675.
Morin, O. and Mercier, H., 2019. Majority rules: how good are we at aggreagating conver-
gent opinions?. Evolutionary Human Sciences, 1, p. E6.
Morin, O., Kelly, P., and Winters, J., 2018. Writing, Graphic Codes, and Asynchronous
Communication. Topics in Cognitive Science, epub, doi:10.1111/​tops.12386
Muthukrishna, M., Morgan, T.J., and Henrich, J., 2016. The when and who of social learning
and conformist transmission. Evolution and Human Behavior, 37(1), pp. 10–​20.
Myers, D.G. and Kaplan, M.F., 1976. Group-​induced polarization in simulated juries.
Personality and Social Psychology Bulletin, 2(1), pp. 63–​66.
Neiman, F.D., 1995. Stylistic variation in evolutionary perspective: inferences from decora-
tive diversity and interassemblage distance in Illinois Woodland ceramic assemblages.
American Antiquity, 60(1), pp. 7–​36.
Nichols, S., 2002. On the genealogy of norms: A case for the role of emotion in cultural evo-
lution. Philosophy of Science, 69(2), pp. 234–​255.
Nickerson, R.S., 1998. Confirmation bias:  A ubiquitous phenomenon in many guises.
Review of General Psychology, 2(2), p. 175.
Niemelä, P.T. and Dingemanse, N.J., 2017. Trustworthiness of online beer ratings as a source
of social information. Behavioral Ecology and Sociobiology, 71(1), p. 24.
Niven, D., 2001. Bias in the news: Partisanship and negativity in media coverage of presi-
dents George Bush and Bill Clinton. Harvard International Journal of Press/​Politics, 6(3),
pp.  31–​46.
Norenzayan, A., Atran, S., Faulkner, J., and Schaller, M., 2006. Memory and mystery: The
cultural selection of minimally counterintuitive narratives. Cognitive Science, 30(3), pp.
531–​553.
Novaes Tump, A., Wolf, M., Krause, J., and Kurvers, R.H., 2018. Individuals fail to reap the
collective benefits of diversity because of over-​reliance on personal information. Journal
of the Royal Society Interface, 15(142), p. 20180155.
OECD, 2016, OECD science, technology and innovation outlook 2016, OECD Publishing,
Paris; Available at: https://​doi.org/​10.1787/​sti_​in_​outlook-​2016-​en.
Olson, R.S. and Neal, Z.P., 2015. Navigating the massive world of Reddit: Using backbone
networks to map user interests in social media. PeerJ Computer Science, 1, p. e4.
Orben, A. and Przybylski, A.K., 2019. The association between adolescent well-​being and
digital technology use. Nature Human Behaviour, 3, pp. 173–​182.
Orne, M.T. and Holland, C.H., 1968. On the ecological validity of laboratory deceptions.
International Journal of Psychiatry, 6(4), pp. 282–​293.
Pareto, V. 2014, Manual of Political Economy, Oxford:  Oxford University Press (original
edition: 1906).
Park, P.S., Blumenstock, J.E., and Macy, M.W., 2018. The strength of long-​range ties in
population-​scale social networks. Science, 362(6421), pp. 1410–​1413.
 References  233

Paul, B., Salwen, M.B., and Dupagne, M., 2000. The third-​person effect: A meta-​analysis of
the perceptual hypothesis. Mass Communication and Society, 3(1), pp. 57–​85.
Pennock, D.M., Flake, G.W., Lawrence, S., Glover, E.J., and Giles, C.L., 2002. Winners don’t
take all: Characterizing the competition for links on the web. Proceedings of the National
Academy of Sciences, 99(8), pp. 5207–​5211.
Petrarca, F., 1951. Rime, trionfi, e poesie latine: a cura di F. Neri [et al.] (Vol. 6). R. Ricciardi.
Petroski, H., 1994. The evolution of useful things. New York, NY: Vintage.
Petty, R.E., Wegener, D.T., and Fabrigar, L.R., 1997. Attitudes and attitude change. Annual
Review of Psychology, 48(1), pp. 609–​647.
Powell, A., Shennan, S., and Thomas, M.G., 2009. Late Pleistocene demography and the ap-
pearance of modern human behavior. Science, 324(5932), pp. 1298–​1301.
Pridmore, S., Auchincloss, S., Soh, N.L., and Walter, G.J., 2013. Four centuries of suicide in
opera. Medical Journal of Australia, 199(11), pp. 783–​786.
Probst, F., Grosswiele, L., and Pfleger, R., 2013. Who will lead and who will follow: Identifying
influential users in online social networks. Business and Information Systems Engineering,
5(3), pp. 179–​193.
Przybylski, A.K. and Weinstein, N., 2017. A large-​scale test of the Goldilocks hypoth-
esis: Quantifying the relations between digital-​screen use and the mental well-​being of
adolescents. Psychological Science, 28(2), pp. 204–​215.
Qiu, X., Oliveira, D.F., Shirazi, A.S., Flammini, A., and Menczer, F., 2017. Limited individual
attention and online virality of low-​quality information. Nature Human Behaviour, 1,
p. 0132.
Quattrociocchi, W., Scala, A. and Sunstein, C.R., 2016. Echo chambers on Facebook.
Available at SSRN: https://​ssrn.com/​abstract=2795110
Reicher, S.D., Haslam, S.A., and Smith, J.R., 2012. Working toward the experi-
menter: Reconceptualizing obedience within the Milgram paradigm as identification-​
based followership. Perspectives on Psychological Science, 7(4), pp. 315–​324.
Ren, Y., Kraut, R., Kiesler, S., and Resnick, P., 2012. Encouraging commitment in online
communities. Building successful online communities:  Evidence-​based social design, pp.
77–​124.
Reyes-​Garcia, V., Molina, J.L., Broesch, J., Calvet, L., Huanca, T., Saus, J., Tanner, S., Leonard,
W.R., McDade, T.W., and TAPS Bolivian Study Team, 2008. Do the aged and knowledge-
able men enjoy more prestige? A test of predictions from the prestige-​bias model of cul-
tural transmission. Evolution and Human Behavior, 29(4), pp. 275–​281.
Rendell, L., Boyd, R., Cownden, D., Enquist, M., Eriksson, K., Feldman, M.W., Fogarty, L.,
Ghirlanda, S., Lillicrap, T., and Laland, K.N., 2010. Why copy others? Insights from the
social learning strategies tournament. Science, 328(5975), pp. 208–​213.
Richerson, P.J. and Boyd, R., 2008. Not by genes alone: How culture transformed human evo-
lution. Chicago, IL: University of Chicago Press.
Rogers, A.R., 1988. Does biology constrain culture? American Anthropologist, 90(4), pp.
819–​831.
Ross, L., Greene, D., and House, P., 1977. The “false consensus effect”: An egocentric bias in
social perception and attribution processes. Journal of Experimental Social Psychology,
13(3), pp. 279–​301.
Rozin, P. and Royzman, E.B., 2001. Negativity bias, negativity dominance, and contagion.
Personality and Social Psychology Review, 5(4), pp. 296–​320.
Rubin, D.C., 1995. Memory in oral traditions: The cognitive psychology of epic, ballads, and
counting-​out rhymes. Oxford: Oxford University Press.
234 References

Salganik, M.J., Dodds, P.S., and Watts, D.J., 2006. Experimental study of inequality and un-
predictability in an artificial cultural market. Science, 311(5762), pp. 854–​856.
Salganik, M.J. and Watts, D.J., 2008. Leading the herd astray: An experimental study of self-​
fulfilling prophecies in an artificial cultural market. Social Psychology Quarterly, 71(4),
pp. 338–​355.
Schofield, D.P., McGrew, W.C., Takahashi, A., and Hirata, S., 2018. Cumulative culture in
nonhumans: Overlooked findings from Japanese monkeys? Primates, 59(2), pp. 113–​122.
Schüll, N.D., 2012. Addiction by design:  Machine gambling in Las Vegas. Princeton,
NJ: Princeton University Press.
Scott-​Phillips, T.C., 2017. A (simple) experimental demonstration that cultural evolution
is not replicative, but reconstructive—​and an explanation of why this difference matters.
Journal of Cognition and Culture, 17(1-​2), pp.  1–​11.
Scott‐Phillips, T., Blancke, S., and Heintz, C., 2018. Four misunderstandings about cultural
attraction. Evolutionary Anthropology, 27(4), pp. 162–​173.
Shennan, S., 2001. Demography and cultural innovation: A model and its implications for the
emergence of modern human culture. Cambridge Archaeological Journal, 11(1), pp. 5–​16.
Shennan, S.J. and Wilkinson, J.R., 2001. Ceramic style change and neutral evolution: a case
study from Neolithic Europe. American Antiquity, 66(4), pp. 577–​593.
Sherry, D.F. and Galef, B.G., 1984. Cultural transmission without imitation:  Milk bottle
opening by birds. Animal Behaviour.
Shifman, L., 2014. Memes in digital culture. Cambridge, MA: MIT Press.
Shore, J., Baek, J. and Dellarocas, C., 2018. Network structure and patterns of information
diversity on Twitter. MIS Quarterly, 42(3), pp. 849–​972.
Sobchuk, O., 2018. Charting artistic evolution:  An essay in theory. Tartu:  University of
Tartu Press.
Solove, D.J., 2007. The future of reputation: Gossip, rumor, and privacy on the Internet. New
Haven, CT: Yale University Press.
Sperber, D., 1985. Anthropology and psychology: Towards an epidemiology of representa-
tions. Man, 20(1), pp. 73–​89.
Sperber, D., 1996. Explaining culture: A naturalistic approach. Oxford: Blackwell.
Sperber, D., 2000. An objection to the memetic approach to culture. In: R. Aungar, 2000,
Darwinizing culture: The status of memetics as a science (Oxford: Oxford University Press).
Sperber, D., Clément, F., Heintz, C., Mascaro, O., Mercier, H., Origgi, G. and Wilson, D.,
2010. Epistemic vigilance. Mind and Language, 25(4), pp. 359–​393.
Sperber, D. and Hirschfeld, L.A., 2004. The cognitive foundations of cultural stability and
diversity. Trends in Cognitive Sciences, 8(1), pp. 40–​46.
Spry, A., Pappu, R., and Bettina Cornwell, T., 2011. Celebrity endorsement, brand cred-
ibility, and brand equity. European Journal of Marketing, 45(6), pp. 882–​909.
Sterelny, K., 2006. The evolution and evolvability of culture. Mind and language, 21(2), pp. 137–​165.
Stibbard-​Hawkes, D.N., Attenborough, R.D., and Marlowe, F.W., 2018. A noisy signal: To
what extent are Hadza hunting reputations predictive of actual hunting skills? Evolution
and Human Behavior, 39(6), pp. 639–​651.
Stubbersfield, J.M., Flynn, E.G. and Tehrani, J.J., 2017. Cognitive evolution and the trans-
mission of popular narratives:  A literature review and application to urban legends.
Evolutionary Studies in Imaginative Culture, 1(1), pp. 121–​136.
Stubbersfield, J. and Tehrani, J.J., 2012. Expect the unexpected? Testing for minimally coun-
terintuitive (MCI) bias in the transmission of contemporary legends: A computational
phylogenetic approach. Social Science Computer Review, 31(1), pp. 90–​102.
 References  235

Stubbersfield, J.M., Tehrani, J.J., and Flynn, E.G., 2015. Serial killers, spiders and
cybersex:  Social and survival information bias in the transmission of urban legends.
British Journal of Psychology, 106(2), pp. 288–​307.
Stubbersfield, J., Tehrani, J., and Flynn, E., 2018. Faking the news: Intentional guided vari-
ation reflects cognitive biases in transmission chains without recall. Cultural Science
Journal, 10(1).
Sunstein, C.R., 2002. The law of group polarization. Journal of Political Philosophy, 10(2),
pp. 175–​195.
Sunstein, C.R., 2018. # Republic: Divided democracy in the age of social media. Princeton,
NJ: Princeton University Press.
Tehrani, J.J., 2013. The phylogeny of Little Red Riding Hood. PloS One, 8(11), p. e78871.
Teng, C.Y., Lin, Y.R., and Adamic, L.A., 2012. Recipe recommendation using ingredient
networks. In: Proceedings of the 4th Annual ACM Web Science Conference (pp. 298–​307).
ACM, New York, NY.
Tennie, C., Call, J., and Tomasello, M., 2006. Push or pull: Imitation vs. emulation in great
apes and human children. Ethology, 112(12), pp. 1159–​1169.
Tennie, C., Call, J., and Tomasello, M., 2009. Ratcheting up the ratchet: On the evolution of
cumulative culture. Philosophical Transactions of the Royal Society of London B: Biological
Sciences, 364(1528), pp. 2405–​2415.
Toelch, U., Bruce, M.J., Newson, L., Richerson, P.J., and Reader, S.M., 2014. Individual con-
sistency and flexibility in human social information use. Proceedings of the Royal Society
B: Biological Sciences, 281(1776), p. 20132864.
Toelch, U. and Dolan, R.J., 2015. Informational and normative influences in conformity from
a neurocomputational perspective. Trends in Cognitive Sciences, 19(10), pp. 579–​589.
Toelch, U., van Delft, M.J., Bruce, M.J., Donders, R., Meeus, M.T., and Reader, S.M., 2009.
Decreased environmental variability induces a bias for social information use in humans.
Evolution and Human Behavior, 30(1), pp. 32–​40.
Tooby, J. and Cosmides, L., 1992. The psychological foundations of culture. In: John Tooby,
Leda Cosmides, Jerome H. Barkow, 1992, The adapted mind: Evolutionary psychology and
the generation of culture (Oxford: Oxford University Press). Oxford, New York.
Toyokawa, W., Saito, Y., and Kameda, T., 2017. Individual differences in learning behaviours
in humans: Asocial exploration tendency does not predict reliance on social learning.
Evolution and Human Behavior, 38(3), pp. 325–​333.
Toyokawa, W., Whalen, A., and Laland, K.N., 2019. Social learning strategies regulate the
wisdom and madness of interactive crowds. Nature Human Behaviour, 3, pp. 183–​193.
Trouche, E., Sander, E., and Mercier, H. 2014. Arguments, more than confidence, explain
the good performance of reasoning groups. Journal of Experimental Psychology: General,
143(5), pp. 1958–​1971.
Turchin, P., 2016. Ages of discord. Chaplin, CT: Beresta Books.
Vaesen, K., Collard, M., Cosgrove, R., and Roebroeks, W., 2016. Population size does not ex-
plain past changes in cultural complexity. Proceedings of the National Academy of Sciences,
113(16), pp. E2241–​E2247.
van der Meer, T.G., Kroon, A.C., Verhoeven, P., and Jonkman, J., 2018. Mediatization and
the disproportionate attention to negative news: The case of airplane crashes. Journalism
Studies, 20(6), pp. 783–​803.
Van Leeuwen, E.J., Cronin, K.A., Schütte, S., Call, J., and Haun, D.B., 2013. Chimpanzees
(Pan troglodytes) flexibly adjust their behaviour in order to maximize payoffs, not to con-
form to majorities. PLoS One, 8(11), p. e80945.
236 References

Van Norel, N.D., Kommers, P.A., Van Hoof, J.J., and Verhoeven, J.W., 2014. Damaged cor-
porate reputation: Can celebrity Tweets repair it? Computers in Human Behavior, 36, pp.
308–​315.
Verpooten, J. and Dewitte, S., 2017. The conundrum of modern art. Human Nature, 28(1),
pp.  16–​38.
Viégas, F.B., Wattenberg, M., and Dave, K., 2004. Studying cooperation and conflict be-
tween authors with history flow visualizations. In Proceedings of the SIGCHI conference
on Human factors in computing systems (pp. 575–​582). ACM Elizabeth Dykstra-​Erickson,
Manfred Tscheligi. ACM. New York, NY.
Vosoughi, S., Roy, D., and Aral, S., 2018. The spread of true and false news online. Science,
359(6380), pp. 1146–​1151.
Watts, D.J. and Dodds, P.S., 2007. Influentials, networks, and public opinion formation.
Journal of Consumer Research, 34(4), pp. 441–​458.
Whiten, A., Goodall, J., McGrew, W.C., Nishida, T., Reynolds, V., Sugiyama, Y., Tutin, C.E.,
Wrangham, R.W., and Boesch, C., 1999. Cultures in chimpanzees. Nature, 399(6737), pp.
682–​685.
Wilson, C., Sala, A., Puttaswamy, K.P., and Zhao, B.Y., 2012. Beyond social graphs: User
interactions in online social networks and their implications. ACM Transactions on the
Web (TWEB), 6(4), p. 17.
Winking, J. and Mizer, N., 2013. Natural-​field dictator game shows no altruistic giving.
Evolution and Human Behavior, 34(4), pp. 288–​293.
Wirtz, J.G., Sparks, J.V., and Zimbres, T.M., 2018. The effect of exposure to sexual appeals
in advertisements on memory, attitude, and purchase intention: A meta-​analytic review.
International Journal of Advertising, 37(2), pp. 168–​198.
Wojcieszak, M., 2008. False consensus goes online: Impact of ideologically homogeneous
groups on false consensus. Public Opinion Quarterly, 72(4), pp. 781–​791.
Wood, L.A., Kendal, R.L., and Flynn, E.G., 2013. Whom do children copy? Model-​based
biases in social learning. Developmental Review, 33(4), pp. 341–​356.
Yaqub, O., Castle-​Clarke, S., Sevdalis, N., and Chataway, J., 2014. Attitudes to vaccination: a
critical review. Social Science and Medicine, 112, pp. 1–​11.
Zervas, G., Proserpio, D., and Byers, J., 2015. A first look at online reputation on Airbnb,
where every stay is above average. Available at SSRN:  https://​ dx.doi.org/​10.2139/​
ssrn.2554500.
Zhou, W.X., Sornette, D., Hill, R.A. and Dunbar, R.I., 2005. Discrete hierarchical organiza-
tion of social group sizes. Proceedings of the Royal Society of London B: Biological Sciences,
272(1561), pp. 439–​444.
Zuiderveen Borgesius, F., Trilling, D., Moeller, J., Bodó, B., de Vreese, C.H., and Helberger,
N., 2016. Should we worry about filter bubbles? Internet Policy Review, 5(1), publication
online.
 Index

Figures are indicated by f following the page number

For the benefit of digital users, indexed terms that span two pages (e.g., 52–​53) may, on
occasion, appear on only one of those pages.

Numbers Apple
4chan 126 iPhone xv
“Screen Time” feature  213
accessibility  65, 66, 214 Arab Spring  96
accumulation 187f,  187–​88 art, evolution of  193–​94
Adamic, Lada  174–​76 art appreciation, effect of
advertisements prestige  61–​62
celebrity endorsements  60–​61, 67 “artificial music market”
sexual content  146 experiments  85–​87
agentic theory  37 flipped ranking version  87–​88
Airbnb reviews  91 Asch, Salomon, conformity
aircraft, technological advances  xv experiment  34–​36, 35f, 76, 83
Aka pygmies, social learning Asimov, Isaac  59
strategies  52–​53 Atran, Scott  147–​48
Alda, Alan  59 attractors 162
Alefantis, James  125 authority, Milgram’s obedience
algorithms  126–​27, 152, 166, 203–​4 experiment  36–​38
bias  204–​7 automobile industry, technological
click-​bait model  127 advance xv
opacity  207–​9 availability xviii, 75–​76
personalization  209–​10 hyper-​availability  96
role in cultural cumulation  210–​11 informational inequality  215–​17
Allcott, Hunt  132–​33
alliteration 167 Baby Shark, YouTube videos  200
Allrecipes 198 Bach, Johann Sebastian, popular
Amazon 12 compositions 73f, 73
product reviews  83–​84,  88 Baillet, Adrian  xvii–​xviii
J-​shaped distribution  91–​92,  92f Bakshy, Eytan  65–​66
Angelina effect  62–​65 Banerjee, Konika  146–​47
animated films, depiction of death  139–​40 Barberá, Pablo  108–​9,  113–​14
anonymous contributions  115, 129 Bartlett, Frederic  150
and misinformation  125–​26 Beckham, Odell  49
self-​regulation  126 Benecol, reviews of  91
Wikipedia 14 benevolence  68, 69
anti-​conformity (rarity) bias  32 Bentley, Alex  74–​75, 196, 202
anti-​vaccination movement  26, 145, 214–​15 Berger, Jonah  182
false consensus effect  123–​24 biased algorithms  204–​7
238 Index

biases Collins, Wilkie, The Moonstone  193–​94

confirmation bias  99 communication 44
conformist bias  76 epistemic vigilance  45
content-​based  134–​36 communities
myside bias  99–​100,  123 layers of grouping  3–​5, 4f
negativity bias  138–​41 virtual networks  5–​7
evolutionary advantages  141–​42 community sizes  2–​3
purchasing bias  92 competence  68, 69
reporting bias  91, 92 competition  188–​89
self-​similarity  98–​99,  100 complex contagion  31
big data  21–​22 confirmation bias  99
Blackmore, Susan  156–​57 confirmatory information  122–​23
Bloody Mary urban legend  148 conformist bias  76, 158
Boris, fake news publication  118–​19 conformity  76–​79,  93
Boxell, Levi  111 dangers of  83
Boyd, Robert  30, 32–​33, 53–​54, 55–​56, experiments on  80–​83
70n9, 77–​78,  157–​58 Asch’s experiment  34–​36, 35f, 76
Boyer, Pascal  122, 124–​25, 147, 162 impact on individual learning  77–​78,  83
Brexit referendum  111 informational and normative  83
Bridle, James  200–​1 maintenance of cultural differences  78
Brodie, Richard  156–​57 production of skewed
Bunny Ears 167 distribution  78–​79
Butler, Oobah  89–​90 s-​shaped curve  77f, 77
Buzzfeed, top fake news stories list  134 conformity bias  32
connections, number of as measure of
Caleb Williams, William Godwin  193–​94 influence 65
Cavalli-​Sforza, Luigi Luca  52–​53 consensus, false consensus effect  122–​24
celebrity influences see prestige conspiracy theories  101–​3,  133
celebrity-​related fake news  151 “Pizzagate” 125
Charbonneau, Mathieu  161 see also misinformation
Charlie bit my finger meme  178 content-​based biases  134–​36
chess, Elo rating trends  198–​200, 199f content factors  69
children’s songs  167 context collapse  9–​10
children’s videos  200 context heuristics  32
derived versions  200–​2 cooking  29–​30
chimpanzees see also recipes
cultural diversity  187 copiers 26
nut cracking  190 copying errors  169–​70
social learning strategies  33 impact, relationship to population
Chinese Whispers  xix, 138 size  195–​96
see also transmission chain experiments see also transmission chain experiments
Cinderella  161–​62,  164 Cosmides, Leda  171–​72
Claidière, Nicolas  81–​82 costly information hypothesis  59–​60
click-​bait model  127 Coultas, Julie  80–​81, 82, 143–​44, 180–​81
climate change denial  26 couple relationships, implications for
Clooney, George  61, 66 friendships  4–​5
clothing 30 critical social learning  27–​28
clown craze  178–​79 Cruise, Tom  64–​65
cognition, evolution of  24–​25 Csibra, Gergely  29–​30
cognitive mechanisms  51–​52 cultural attraction  136–​38,  159–​62
 Index  239

Cultural Cognition Project, Yale deaths, depiction in animated films  139–​40

University  111–​12 de Barra, Micheál  90–​91, 144
cultural differences, role of conformity  78 deception 44
cultural diversity  187–​88 deliberating enclaves  103–​5
cultural evolution  xiii–​xiv Dennet, Daniel  156
importance of study  21–​22 density of a network  110–​11
misinformation  151–​53 depolarization  112–​14
opaque customs  29–​31 Derex, Maxime  30–​31, 43
social learning  25–​29 detective novels, "invention" of  193–​94
social learning strategies  31–​34 diameter of a network  110
see also cumulative cultural evolution Dictator Game  15–​17
cultural hitchhiking  56–​57, 68, 69 digital cumulation  195–​200
cultural traits  23–​24,  217–​18 digital storage  169
cultural transmission  xxii, 218–​19 digital tools, technological advance  xv–​xvii
digital  xix–​xx,  168–​71 digital transmission  168
effective cultural population size  17–​19 fidelity amplifiers  168–​71
evolution without replication  158–​59 Dingemanse, Mark  170
features of  xviii–​xix DISCOUNTMACHINE strategy  39–​40
fidelity  160–​61 disgust  143–​45,  180–​81
continuum of  162–​64 Distracted Mind, The, Adam Gazzaley and
fidelity amplifiers  164–​68 Larry Rosen  24–​25
jukebox analogy  171–​72 distributions  71–​73
memes  156–​58 see also online popularity distributions;
internet memes  172–​79 skewed distributions
online sharing  179–​84 diversity  187–​88
role of social media  9–​11 diversity exposure  113–​14
of stories  161–​62 provision by media  115
three phases of  180 Dodds, Peter  67–​68,  85–​87
Xerox machine analogy  172 dog breeds  79
culture, nature of  217–​18 influence of movies  64
cumulative cultural evolution  xxii, 186–​87, skewed distribution  74–​75
187f, 191 Downfall (Hitler Reacts) meme  176–​77
accumulation  187–​88 Dracula 164
of art  193–​94 Drake 49
chess skills  198–​200, 199f Dunbar, Robin  4–​5, 6, 10
digital cumulation  195–​200 Dunbar’s number  2–​3
effect of population size  192–​93
importance of fidelity  195–​96 echo chambers  xxi, 96, 114–​16
improvement  188–​89 and deliberating enclaves  103–​5
junk culture  200–​2 group polarization  96–​97
of language  191–​93, 192f myside bias  99–​100
online recipes  197–​98 old and new media  110–​12
ratcheting  190–​91 online and offline  105–​7
role of algorithms  210–​11 polarization in social media  107–​10
role of science  196–​97 science and conspiracy theories  101–​3
self-​similarity bias  98–​99
Darwin, Charles  22–​23 economic games, cooperation  15–​17
data use  21–​22 edge homogeneity  101–​2
date format experiment  80–​81 effective cultural population size  17–​18
Dawkins, Richard  156 impact of digital technology  18–​19
240 Index

Einstein, Albert  22, 57–​58 consumption of, relationship to

Elo chess rating system, temporal Facebook usage  131–​32
trend  198–​200,  199f diffusion on Twitter  128–​29
email sharing  183 disgust reactions  143–​45
emotional content, sharing of  182–​83 exposure to  217
emotional expression, trends in printed minimally counterintuitive
books 139 concepts  148–​49
encode-​and-​retrieve phase  180–​82 negativity bias  140–​41
engagement rate  66–​67 sexual content  146
epidemiology of representations (cultural social content  150–​51
attraction theory)  136–​38 threat-​related  142–​43
epistemic vigilance  45 see also misinformation
Equality versus equity meme  173f, 173 fake news websites  130–​31
modifications and variations  173–​74 fake restaurant story (“The Shed at
Eriksson, Kimmo  82, 143–​44, 180–​81 Dulwich”) 89
essentialist thinking  149 fake reviews
ethnicity filters for  90
differences in disgust reactions  144 impact of local competition  90
self-​similarity bias  98–​99 false consensus effect  122–​24
ethnomedicinal knowledge  54–​55 false positives and false negatives  142
etiquette norms  143 feedback 167
Expedia reviews  90, 91 reputation points  13
explicitness  xix, 51 fidelity  xix, 160–​61, 184, 219
explorers 26 continuum of  162–​64,  171–​72
exposure 31 and cultural cumulation  195–​96
extremism  96–​97 internet memes  173
see also polarization fidelity amplifiers  164–​68
eyewitness accuracy  35–​36 digital transmission  168–​71
Finger Family, YouTube versions  200–​2
Facebook  xv, 1–​2, 5, 213 “first-​order” cues  53
algorithm use  127 flexibility 33
context collapse  9–​10 “Florida Man”  215
data use, mood change folktales
investigation  21–​22 complexity 195
friends transmission of  161–​62, 164
geographical distribution  8 foraging problems  24–​25
number of  7 frequency-​dependent biases  32
like button and counters of shares  84 friends
networks  5–​6,  7 Dunbar’s number  2–​3
newsfeed  208–​9 Facebook  1–​2,  7
polarization  107–​8 geographical distribution  8–​9
science and conspiracy theories  101–​3 impact of digital technology  5
usage, relationship to fake news layers of grouping  3–​5, 4f
consumption  131–​32 time investment  4–​5
volume of content  203 friendship paradox  109–​10
Facebook groups  96 Froehle, Craig  173–​74
faces, pareidolia  135–​36
fake news  xxi, 118–​19, 179 Gabriel, Peter  37
availability 215 gamification 13
celebrity content  151 Gangnam Style  178, 179
 Index  241

Gazzaley, Adam  24–​25 improvement 187f,  188–​89

Gelman, Andrew  119–​20 indirect bias  70n9
gender differences in internet usage  214 individual learning  42, 43, 77–​78,
gender roles  98 83, 217–​18
genetically modified organisms  149 advantages  26–​28
genetic testing, Angelina effect  62–​64 computer tournament
Gentzkow, Matthew  106–​7,  132–​33 strategies  39–​40
geographical distribution of contacts  8–​9 “virtual arrowhead” experiments  40–​41
Gergely, György  29–​30 influencers 52, 65–​67
Ghirlanda, Stefano  64 local (micro-​influencers)  66–​67
GitHub  10–​11 informational asymmetry  141–​42
global village  8–​9 informational conformity  83
Godwin, Mike  172–​73 informational inequality  215–​17
Godwin, William, Caleb Williams  193–​94 information donors  12–​14
Godwin’s law  102 motivation  14–​17
“gone wrong” videos  50–​51 information overload  xvii–​xviii
Goody, Jack  xi–​xii information storage  25–​26
Google information transmission, social nature
algorithm opacity  207–​9 of  24–​25
algorithm use  126–​27 Instagram 5
Page Rank algorithm  209 celebrities 52
personalization of searches  210 internet access  169, 214
volume of information  203 internet memes  172–​73
gossip  150–​51 clown craze  178–​79
Grass 11 definition of  174–​75
Grimm Brothers folktales, counterintuitive Downfall (Hitler Reacts)  176–​77
elements  147–​48 Equality versus equity 173f,  173–​74
group polarization  96–​97,  111–​12 fidelity 173
deliberating enclaves  103–​5 Grumpy Cat 177
science and conspiracy theories  101–​3 Harlem Shake 178
see also polarization modifications and
Grumpy Cat meme  177 variations  173–​76,  177
mimicry 178
Harlem Shake meme  178 remix 177
Hawking, Stephen  60 mutation rate  175–​76
health campaigns real world counterparts  178–​79
celebrity endorsements  62–​64, 67 internet use  xvii
conspiracy theories  101 Inuit people, caribou skin clothing  30
see also anti-​vaccination movement Io Vivo a Roma  130–​31
height distribution  71–​72 iPhone xv
Henrich, Joe  25–​26, 30, 32, 53, 77–​78, 98, 158 isolation index  106–​7
Herzog, Hal  64
Hewlett, Barry  52–​53 Jacquet, Pierre  41
high-​fidelity transmission  160–​61, 162, Japanese macaques (“snow monkeys”)  190
173, 184, 195, 219 job-​screening algorithms  204, 205–​6,  208–​9
see also fidelity amplifiers Jolie, Angelina  62–​64
Hitler Reacts meme  176–​77 Jordan, Michael  60, 61
hunting skills  53, 54 jukebox analogy, cultural
Hutterites 2 transmission  171–​72
hyper-​availability  96 junk culture  200–​2
242 Index

Kelly, Kevin  12 Milkman, Katherine  182

Kendal, Jeremy  195 Miller, George (“Joji”)  178
Kennedy, John F, conspiracy theories  133 mimicry 178
keyboard covering experiment  80 minimally counterintuitive
Kiki challenge  49–​51, 50f concepts  146–​49,  161–​62
Kirby, Simon  191–​92 misinformation  xxi–​xxii, 117–​22,  184
Knapp, Justin  13–​14 “artificial music market”
experiment  87–​88
Laland, Kevin  39–​40 availability 215
language diversity  188 cognitive appeal  134–​38
language evolution  191–​93, 192f cultural evolution viewpoint  151–​53
laughing  162–​63 diffusion on Twitter  128–​29
Lerique, Sébastien  176 disgust reactions  143–​45
letters, analysis of  163 effects  131–​34
Lieberson, Stanley  63 fake news  118–​19
lies 117 false consensus effect  122–​24
see also misinformation false positives and false negatives  142
Lincoln, Abraham  57 list of top fake news stories  134
linear conformity  81–​82 minimally counterintuitive
Little Red Riding Hood 195 concepts  146–​49
local influencers (micro-​influencers)  66–​67, 69 motivation 153
locked-​room plot device, evolution negativity bias  138–​41
of  193–​94 evolutionary advantages  141–​42
LOLcats 177 New York Sun moon hoax series  117–​18
Lycett, Stephen  78–​79 Oreo cookies study  120–​22
Lynch, Aaron  156–​57 “Pizzagate” 125
Lynch, Michael Patrick  24 prevalence  129–​31
reputational costs and
“Macedonian teenagers,” fake news anonymity  124–​26
publication  118–​19 role of reach  117–​18
Makinson, Jenny  165, 166f sexual content  145–​46
Martian meteorite bacterium paper  119 social interactions and gossip  150–​51
Masha and the Bear 200 threat-​related  141–​43
Mayr, Ernst  22–​23 see also online sharing
McCarthy, Jenny  64–​65 missing-​target problem  56, 68, 69
McElreath, Richard  32 model-​based biases  32
McGuigan, Nicola  165, 166f mood, influencing factors  21–​22
measles vaccination  123–​24 mood signature of comments  102
see also anti-​vaccination movement moon landings, conspiracy theories  101
medical treatments, reviews of  90–​91 Moonstone, The, Wilkie Collins  193–​94
memes  156–​57 MOORPGs  6–​7
clown craze  178–​79 Morgan, Tom  33
criticisms of  157–​59 Morin, Olivier  134–​35, 163, 170, 181
internet  172–​79 movies
Mercier, Hugo  99–​100 influence on dog breeds  64
Mesoudi, Alex  40–​42, 53, 78–​79,  150 sexual content  145–​46
micro-​influencers  66–​67,  69 Murders in the Rue Morgue, The, Edgar
Milgram, Stanley, obedience Allan Poe  193–​94
experiment  36–​38 mutual reviewing  91
 Index  243

myside bias  99–​100,  123 ostensive communication  44

overimitation experiment  165, 166f
names 189
celebrity influences  63 Page Rank algorithm  209
skewed distribution  74f, 74, 75–​76 Paltrow, Gwyneth  64–​65
National Report  152–​53 Pardus  6–​7
negative content, sharing of  182 pareidolia  135–​36
negativity bias  138–​41 parents, learning from  23, 33, 52–​53
evolutionary advantages  141–​42 pay-​off (success-​biased) social learning  53
neocortex ratio  2 Pearl Harbour, conspiracy theories  133
Netflix, personalized suggestions  209 Peppa Pig, derived version  201, 202
networks  5–​7 personalized algorithms  209–​10
news coverage, negativity bias  140 Petroski, Henry  188
newspapers, echo chambers  110–​12 Pinterest 198
New York Sun, moon hoax series  117–​18 “Pizzagate” 125
Nichols, Shaun  143 Plato xi
non-​human animals Podesta, John  125
social learning  164–​65 Poe, Edgar Allan, The Murders in the Rue
see also chimpanzees; Japanese acaques Morgue  193–​94
non-​rival goods  169 polarization  114–​16,  214–​15
Norenzayan, Ara  147–​48 age differences  111–​13
normal distribution  72 deliberating enclaves  103–​5
normative conformity  83 effects of diversity exposure  113–​14
Not by Genes Alone, Robert Boyd & Pete effects of old and new media  110–​12
Richerson  32–​33 group polarization  96–​97
nut cracking, chimpanzees  190 isolation index  106–​7
reduction of  112–​14
obedience, Milgram’s experiment  36–​38 science and conspiracy theories  101–​3
O’Brien, Mike  196, 202 in social media  107–​10
offline interactions, echo chambers  105–​7 polio vaccination campaign, role of
old media, echo chambers  110–​12 Elvis  62, 63f
online popularity distributions  79 political ideology
online sharing effects of diversity exposure  113–​14
by email  183 Facebook friends  107–​8
emotional content  182–​83 isolation index  106–​7
recipes  197–​98 polarization trends  111
success of misinformation  184 Twitter users  108–​9
threat-​related misinformation  181–​82 political misinformation  129–​30, 131–​34,
transmission chains  179–​81, 180f 151, 152
opacity  xviii–​xix, 100, 115, 207–​8 Polynesians, tattoos  55–​56
opaque algorithms  207–​9 popularity  xx–​xxi, 50–​52,  93–​94
opaque customs  29–​31 advantages of  71
opera, suicide depiction  139–​40 assessment of  71
oral conversations, "repair" requests  170 and availability  75–​76
oral traditions  167 conformity  76–​79
Orchestral Suite No. 3, Johann Sebastian experiments on  80–​83
Bach 73 explicit cues  83–​88
Oreo cookies study  120–​22 “artificial music market”
Osbourne, Ozzy  60 experiments  85–​88
244 Index

popularity (cont.) random copying (unbiased transmission)

fake restaurant story  89–​90 models 75
fake reviews  90 rarity (anti-​conformity) bias  32
online popularity distributions  79 ratcheting 187f,  190–​91
reporting bias  91 evolution of the detective story  193–​94
skewed distributions  72–​75 reach  xviii, 71, 117–​18, 125
population size Real News Right Now  152–​53
effect on cumulative cultural recipes  165–​68
evolution  194–​95 online sharing  197–​98
and impact of copying errors  195–​96 opaque customs  29–​30
and language evolution  192–​93 Reddit 11
relationship to folktale complexity  195 Redfield, Rosie  119
population thinking  22–​24 religious beliefs  147
populism  214–​15 remix 177
portrait traditions  134–​35 repair mechanisms  170–​71
positive content, sharing of  182–​83 replicators 156
positive test strategy  99–​100 evolution without replication  158–​59
post-​truth  117 see also memes
potato washing behaviour, macaques  190 reporting bias  91, 92
Presley, Elvis  62, 63f reputational costs of
prestige misinformation  124–​26
Angelina effect  62–​65 reputation points  13
Kiki challenge  49, 50–​51 research, diversity of  xiii
online influencers  52, 65–​67 Retenews24 130
in perspective  67–​69 reviews  83–​84,  92–​94
strength of celebrity influence  59–​62 “artificial music market”
and success of quotations  58–​59 experiments  85–​88
prestige bias  32 fake
prestige-​biased social learning  53–​55 fake restaurant story  89–​90
cultural hitchhiking  56–​57 filters for  90
missing-​target problem  56–​57 impact of local competition  90
runaway processes  55–​56 mutual reviewing  91
presumption of good design  25, 68 reporting bias  90–​91
primitive technology, video reward systems  13–​14
tutorials 171 Reyes-​Garcia, Victoria  55
printing press  xvii–​xviii,  168–​69 rhymes 167
privacy  213, 214 Richerson, Pete  30, 32–​33, 53–​54, 55–​56,
product reviews see reviews 70n9, 157–​58
programming Rise of the Machines, The  204, 205f
esoteric languages  11 rival goods  169
sources of code  10–​11 Rogers, Alan, Roger’s paradox  27f,  27–​28
purchasing bias  92 Ronaldo, Cristiano  52, 66
Roosevelt, Franklin, conspiracy theories  133
Quattrociocchi, Walter  101–​2 Rosen, Larry  24–​25
quotes Roth, Camille  176
changes to  176 Royzman, Edward  141
choice of, effect of popularity Rozin, Paul  141
cues  84–​85 Rubin, David  167
misattributed  57–​59 runaway processes  55–​56,  68
 Index  245

Salganik, Matthew  85–​88 slot machine design  136–​37

Santé+  130–​31 smartphone use  219
satirical news  140–​41, 148–​49,  152–​53 Snapchat 5
satirical versions of memes  174 Snoop Dog  60
Schüll, Natasha Dow  136–​37 Snopes 148
Science, Martian meteorite bacterium Sobchuk, Oleg  193–​94
paper 119 social brain hypothesis  1–​2
science, role in cultural cumulation  196–​97 social grouping
science theories  101–​3 community sizes  2–​3
scientific papers layers of organization  3–​5, 4f
discussion in social media  119–​20 social inequalities  217
Oreo cookies study  120–​22 social influence
Scientology  64–​65,  70n31 Asch’s conformity experiment  35f
Scott-​Phillips, Thom  163 eyewitness accuracy  35–​36
Screen It! website  145–​46 forest fire metaphor  67–​68
screen time  213 power of authority  36–​38
search engines see Google third-​person effect  45–​46
“second-​order” strategies and cues  53 see also influencers
Selfish Gene, The, Richard Dawkins  156 social information, sharing of  181–​82
self-​similarity  98–​99 social learning  24–​25,  217–​18
self-​similarity bias  100 Asch’s conformity experiment  34–​36
serial reproduction, method critical  27–​28
of 138f,  138–​39 cultural variability  41–​42
see also transmission chain experiments deception and vigilance  43–​47
Sessions, Roger  57 environmental influences  41
sexual content, role in misinformation  145–​46 experiments on  40, 41
Shaggy Dog, The (Disney, 1959)  64 “virtual arrowhead” task  40
shallow engagement  126–​29, 152, 203 information storage  25–​26
Shapiro, Jesse  106–​7 limitations  26–​28
sharing non-​human animals  164–​65
by email  183 opaque customs  29–​31
emotional content  182–​83 overestimation of  42–​43, 46
success of misinformation  184 Roger’s paradox  27f
threat-​related misinformation  181–​82 underuse 42
transmission chains  179–​81, 180f social learning strategies  28–​29, 31–​32,
see also online sharing 33–​34,  46
“Shed at Dulwich, The”  89–​90 computer simulation  39–​40
Shennan, Stephen  74–​75 conflicting interpretations  38
Shifman, Limor  177–​78 flexibility 33
Shiggy 49 prestige biased learning  67–​69
similarity bias  32 self-​similarity bias  98–​99
Simpson, O.J.  61 transmission biases  32–​33
“six degrees of separation”  8–​9 social media
skewed distributions  72–​75, 93 cultural transmission  9–​11
junk culture  202 geographical distribution of
names 189 contacts  8–​9
online popularity distributions  79 implications for friendships  5
reviews  91–​92,  92f information donors  12–​14
role of conformity  78–​79 motivation  14–​17
246 Index

social media (cont.) simulation of online sharing  179–​81, 180f

virtual networks  5–​7 social information  150
see also Facebook star/​scribbles case  159–​61, 160f, 163
social stories  150–​51 TripAdvisor
Socrates  xi–​xii,  170 fake restaurant story  89–​90
Spears, Britney  61, 67 reviews  88–​89, 90, 91
specialization 103 triplet experiment  99–​100
deliberating enclaves  103–​5 true information sharing  119–​20
Spencer, Herbert  22–​23 Trump, Donald  111, 182–​83
Sperber, Dan  23–​24, 45, 99–​100, 136, 159–​ election, effects of
61, 162, 195–​96 misinformation  131–​33
Spotify  86–​87,  88 fake news concerning  118
stability of cultural traits  161 “Truth or Trash” experiment  139
Stack Overflow  10–​11 Twain, Mark  57
reputation points  13 Twitter 13
star/​scribbles experiment  159–​61, 160f, algorithm use  126–​27
163,  195–​96 core/​periphery structure  109–​10
Sterelny, Kim  43 data use, mood change
Stoker, Bram  164 investigation  21–​22
stories see folktales diversity exposure  113–​14
Stubbersfield, Joe  143–​44, 148, false and true news diffusion study  128
151, 181–​82 follower numbers  10
success bias  32 polarization  109–​10
success-​biased (pay-​off) social learning  53 retweeting behaviour  104–​5,  108–​9
suicide, depiction in opera  139–​40 spatial distribution of contacts  9
Sunstein, Cass  xviii, 96–​97, 104, 105 study of influencers  65–​66
superstitious beliefs  147 use of  219
Swift, Jonathan  117 volume of tweets  203
“Sylvia’s recipe”  29–​30 typewriting errors  170
sympathy groups  3–​4,  6–​7
unbiased transmission (random copying)
tastes 135 models 75
tattoos  55–​56 university homepages, popularity
technological variation  188 distribution 79
Tehrani, Jamie  148, 195 unnecessary actions, copying
television, echo chambers  110–​12 of  164–​65,  166f
Tennie, Claudio  164–​65 Upworthy 127
third-​person effect  45–​46 urban legends  143–​44
threat-​related misinformation  141–​43,  153 minimally counterintuitive concepts  148
sharing of  181–​82
tie range  8–​9 vaccine hesitancy  26, 145, 214–​15 see anti-​
Tomasello, Michael  190–​91 vaccination movement
Tooby, John  171–​72 false consensus effect  123–​24
tool use, chimpanzees  187, 190 vandalism  214–​15
Top de l’Humour, Le  130–​31 video tutorials  171
transmission biases  28, 32–​33, 46, 158 viral spread  179, 218
transmission chain experiments  138f, Kiki challenge  49–​51
138–​39,  142 “virtual arrowhead” task  40–​41, 43, 54
disgust-​evoking stories  143–​44 cultural variability  41–​42
 Index  247

virtual networks  5–​7 winner-​take-​all (skewed)

virtual replication  169 distributions  72–​75
word repetition  167
Wales, Jimmy  13–​14 writing  163–​64
Walkman effect  xii–​xiii effects of  xi–​xii
wary learners (flexible imitators)  46–​47, impact of digital technology  xvi–​xvii
83, 124, 214–​15
Watts, Duncan  67–​68,  85–​88 Xerox machine analogy, cultural
weak ties  113 transmission 172
wealth distribution  72
weather, influence on mood  21–​22 Yelp, filtering for fake reviews  90
West, Kanye  84 YouTube 5, 13–​14
WhatsApp 5 children’s use of  219
Wheels on the Bus, YouTube videos  200, 202 popularity of videos  15
Whiten, Andrew  165, 166f random walk through
Whitespace 11 recommendations 206
Wikipedia videos for toddlers  200
anonymous contributions  14, 125–​26 derived versions  200–​2
motivation  15, 17 video tutorials  171
reward systems  13–​14
Wilde, Oscar  59 Zuckerberg, Mark  213