EXTENDED COGNITION: GESTURES IN THE DIRECTION OF AN END
GAME
1. Introduction
Do mental processes occur purely inside the heads of creatures that instantiate them? Or are they
partly constituted by processes that occur in the bodies or even the wider environments of such
creatures. In recent years, the claims that (some) mental processes are embodied, and that (some)
are even environmentally extended, have become progressively more popular. The terms of the
debate between proponents and opponents of embodied and/or extended cognition have,
however, become curious. At first glance, one might suppose that the debate is one that will be
decided at the traditional level of philosophical, perhaps even conceptual, analysis. What sorts of
things are mental processes? Are they the sorts of things that are likely to extend beyond the
brain? If the debate were to occur at this level, then one might think that traditional philosophical
accounts of the mental might figure quite heavily. What is the mark of the mental? Is it
intentionality? Is it consciousness? And so on. This is not, however, the way things turned out.
Instead, the debate has become widely regarded as one that should be fought and decided at the
level of philosophy of cognitive science.1 The focus has largely been on sub-personal cognitive
processes – the vehicles of cognition. And the interest in the debate has largely been conceived
in term of its ramifications for the future development of cognitive science. How are the kinds of
this science to be individuated? What kinds of explanatory mechanisms will this science
1
Robert Rupert has, I think, played a significant role in pushing the debate in this direction. See his (2004) and
1
postulate? The possibility that the debate might, instead, be conducted at another – rather more
philosophical – level has scarcely been acknowledged.
In The New Science of the Mind (henceforth NSM) I attempted to relocate the debate to
this more abstract, philosophical, level. At the logical core of this attempt was an analysis of
intentionality. The theses of embodied and extended cognition are generally thought to be
recherché doctrines, at odds with common sense. I argued that this sense of oddity was grounded
in a particular, uncritically accepted, conception of intentional directedness – a conception that it
was the task of the book to unseat.
In this paper, I shall attempt to develop this case further. I shall argue that, given the
conception of intentionality defended in NSM, and rehearsed again in this paper – a conception
that, broadly speaking, has its roots in the phenomenological tradition – we are in a position to
see at the least the general outlines of an end game: the sort of game that is going to have to be
played in order to settle the dispute between proponents and opponents of the
extended/embodied view of mental processes.
2. What is Extended Cognition?
I shall use the expression ‘extended cognition’ (henceforth EC) is a slightly tendentious way.
Often, one finds extended cognition distinguished from embodied cognition. The concept of
extended cognition employed in this paper, following on from NSM, subsumes both views.
Cognitive processes are amalgamations of neural processes, bodily processes, and environmental
processes.
Amalgamations: There are two things wrong with the commonly used expression, ‘the extended
mind’. First, there is the word ‘mind’, which suggests that the issue concerns the nature of the
2
mind. It does not. The issue pertains to mental processes, usually cognitive processes. Secondly,
there is the word ‘extended’, which suggests that the issue pertains to the location of mental
items. It does not. From the idea of mental processes as amalgamations, it might be possible to
deduce various claims about where cognitive processes are not. But it is, in general, not possible
to deduce, with any useful level of precision, where they are. One of the more sensible
deductions is that cognitive processes often do not have any clear location.
Processes: According to the thesis of amalgamated cognition, cognitive processes can be, in part,
made up of the manipulation, transformation or exploitation of certain appropriate structures in
the environment (including the body). ‘In part’ is important – there is always a non-eliminable
neural component. But the whole process – the amalgamation – counts as cognitive, not just the
neural component. I do not identify external structures with cognitive states. So, I would deny,
for example, that the entries in Otto’s notebook count as beliefs. Otto’s manipulation of his
notebook can count as part of his process of remembering. But the sentences are not his beliefs. I
think there is good reason for denying this, but I won’t go into that here.
Existential Quantifier, not Universal: Some cognitive processes are amalgamations of neural,
bodily and environmental processes. Some are amalgamations of neural and bodily. Some are
purely neural. I do not want to claim that all cognitive processes are amalgamations. Only that
some are. Therefore, I’m quite comfortable with brains in vats, etc.
Contingent: The idea of cognitive processes as amalgamations is a contingent claim about the
way (some) cognitive processes, in fact, turned out, not a necessary claim about the way they
have to be. As a matter of contingent fact, some cognitive processes are amalgamations of
neural, bodily and environmental processes.
3
3. The Argument of The New Science of the Mind
In NSM I argued that cognition is revealing activity that conforms to the mark of the cognitive.
The argument is made up of two connected strands (a) the analysis of intentionality as revealing
activity, and (b) the criterion of cognition. The cornerstone of the entire book is the analysis of
intentionality. For reasons that are unclear to me, almost all commentators and reviewers have
focused on the mark of the cognitive and completely ignored the analysis of intentionality. This
is unfortunate. The mark of the cognitive cannot be understood independently of the analysis of
intentionality: without the analysis, the mark of the cognitive is (and was intended to be)
incomplete.
A. The Analysis of Intentionality
The analysis assumes a certain model of intentionality that has its roots in the phenomenological
tradition and, despite some lean years, is still widely enough accepted to be dubbed the standard
model. According to this, intentionality has a tripartite structure, comprising act, object, and
mode of presentation. The mode of presentation connects act and object. The general idea is the
act has a content, expressible in the form of a description, and the mode of presentation is that in
virtue of which the object satisfies that description.2
The core argument of NSM begins by showing that the idea of a mode of presentation is
ambiguous. In any intentional act, we find two different sorts of mode of presentation.
Empirical modes of presentation (aspects): Often, indeed typically, the notion of a mode of
presentation is understood as the way objects appear to subjects. If the tomato appears red and
2
For those who dislike description theories, there is a non‐description‐theoretic version of this model available (in
fact, there are several of them). This is irrelevant to current concerns, since these are unaffected by which version
of the standard model we adopt. I shall henceforth ignore this complication.
4
shiny, then redness and shininess is the mode of presentation of the tomato. In this sense, the
mode of presentation is an intentional object – it is the sort of thing of which I can become aware
if my attention is suitably engaged. I can attend not only to the tomato, but also to its redness and
shininess. An empirical mode of presentation is an intentional object. As such, it is identical with
what is sometimes called an aspect of an object.
Transcendental modes of presentation: The standard mode of intentionality has a clear, if
curiously overlooked, implication. In any intentional act, there must be more than an empirical
mode of presentation. There must also be a transcendental mode of presentation. The mode of
presentation is supposedly what fixes reference – determines the intentional object of a mental
act. So, whenever the object of an intentional act is an empirical mode of presentation, there
must be another mode of presentation – a transcendental mode of presentation – that fixes
reference to the empirical mode of presentation. The transcendental mode of presentation is that
component of the intentional act that permits the object to appear under empirical modes of
presentation (or aspects).
If we want to understand the intentionality – the directedness – of an act, we will look in
vain to the objects of this directedness. The directedness of an intentional act towards the world
consists in its transcendental mode of presentation. The transcendental mode of presentation is
the intentional core of an act. That is, the directedness of an intentional act consists in its
permitting objects to appear under aspects.
Therefore, in this sense, intentional directedness is a form of revealing or disclosing
activity. I distinguished two forms of disclosing activity: causal and constitutive. Constitutive
disclosure takes the form of a logically sufficient condition for the world to fall under an
5
empirical mode of presentation. What it is like to have an experience provides a logically
sufficient condition for the world to fall under an empirical mode of presentation. Causal
disclosure takes the form of a physically sufficient condition for the world to fall under an
empirical mode of presentation. Cognitive processes typically supply only a physically sufficient
condition for the world to fall under an empirical mode of presentation. For example, if correct,
David Marr’s account of the computational processes that progressively transform the retinal
image into a 3D object representation would provide a physically, but not logically, sufficient
condition for an object to appear a certain way. Constitutive disclosure is disclosure by way of
content. Causal disclosure is disclosure by way of vehicles of content. It is causal disclosure that
is relevant to extended cognition (since this is a thesis about cognitive vehicles).
Revealing activity often – not always, certainly not necessarily – straddles neural
processes, bodily processes, and processes of manipulating or transforming environmental
structures. This is why cognitive processes are often amalgamations of all three.
Intentionality is revealing activity. Cognition is a specific form of intentional activity –
revealing activity that conforms to the mark of the cognitive. This brings us to the second strand
of the argument in NSM: the criterion of cognition.
B. The Mark of the Cognitive
A process P is a cognitive process if:
(1) P involves information processing—the manipulation and transformation of informationbearing structures.
(2) This information processing has the proper function of making available either to the subject
or to subsequent processing operations information that was (or would have been) prior to (or
6
without) this processing, unavailable.
(3) This information is made available by way of the production, in the subject of P, of a
representational state.
(4) P is a process that belongs to the subject of that representational state.
I argued that (1)-(3) are fairly straightforward, and are derived from an examination of
paradigmatically internalist examples of cognitive scientific practice (I employed Marr’s theory
of vision as a canonical example of internalist cognitive science).
It is crucial to understand that the criterion of cognition cannot be understood in isolation
from the analysis of intentionality. The criterion and the analysis are linked through condition
(4). The ownership of cognitive processes is ultimately a matter of the fact that they are
disclosing activity; and disclosure is always disclosure to someone (broadly construed). The
ownership of a cognitive process is, therefore, understood in terms of disclosure.
4. Aside: On the Development of my Mark of the Cognitive
In a recent paper, Fred Adams has been spinning the contribution that he and his colleague Ken
Aizawa have made to the development of my thought.3 (AS is traditional, I shall henceforth refer
to the collective as A&A). The spin goes something like this: thanks to our valiant efforts,
Rowlands has finally come around to accept the need for a mark of the cognitive.
This spin, however, seems to overlook the fact that I was always working with a mark of the
cognitive – certainly as far back as 1999 (maybe 1995). In The Body in Mind (henceforth BIM)
or example, this was the criterion of cognition I employed:
3
Adams (2010). I am not the only one to whose development they have contributed in this respect. Adams also
mentions other benighted souls such as Mike Wheeler.
7
1. Cognitive task: defined by ostension.
2. Cognitive process: information processing operations (understood as the manipulation
and transformation of information bearing structures)
3. Where this makes previously unavailable information available.
4. And this information is relevant to the solution of the cognitive task.
The principal argument of BIM was that processes extending outside the skin counted as
cognitive because they satisfied this criterion of the cognitive. I no longer believe this criterion is
adequate – this is what has prompted my return to this subject in recent years. But an inadequate
criterion of the cognitive is, nevertheless, a criterion of the cognitive.
Adams (2010) also mentions a conference (Extended Mind II: Just When You Thought It
Was Safe to Go Back in The Head) held at the University of Hertfordshire in 2006. He laments:
To our great surprise, almost all of the proponents of extended mind vehemently objected
to the call for a mark of the cognitive by Adams & Aizawa. Even Rob Rupert was
warming slowly, if at all to the need. There seemed to be a wealth of opinion that one
doesn’t need a mark of the medical to practice medicine, or a mark of the biological to do
biology, or a mark of the psychological to do psychology, etc. The list of skeptics about
the need included: Clark, Chalmers, Keijzer, Menary, Hurley, Gallagher, Ross,
Rowlands, Shapiro, Sutton, among others.4
On a brighter note, he adds in a footnote:
4
Fred Adams (2010), p. 325.
8
I’m happy to report that Rowlands, Shapiro, and Wheeler (at least) have now come
around and see the need for a mark of the cognitive in connection with this dispute.5
I’m a little puzzled by this, and I suspect Fred must be misremembering. At the 2006 conference
in question, I presented a paper that outlined and defended a criterion of the cognitive. This was
an earlier version of my (2009) version, and looked like this:
P is a cognitive task if:
1. P involves information processing – the manipulation and transformation of information
bearing structures
2. This manipulation and transformation makes available information – to a subject or subpersonal processing operations – that was previously unavailable.
3. It does this by the production in the subject of a representational state.
Ron Chrisley, in conversation, convinced me that (2) needed to be amended to incorporate the
idea of proper function. And a useful objection by Richard Samuels – it was actually a version of
the bloat objection – convinced me that I needed a further condition. This became condition (4)
of my criterion I defended in NSM.6 My thanks to Chrisley and Samuels. But I suspect the
significance of A&A in the development of my thought might have been somewhat
overestimated.
5. Adams and Aizawa’s Objections to my Mark of the Cognitive
Adams (2010) raises several objections to my mark of the cognitive. Here is the first:
5
Adams (2010), p 325.
And also earlier my (2009) version (which was scarred by an ‘iff’ when there should only have been an ‘if’. His was
a typo. I’m a sloppy proofreader).
6
9
Now, of course, all of the action is in (4). If “belongs to the subject” means cognitive
subject, then the account is circular and helps not at all. If it does not mean cognitive
subject, then my computer satisfies all four conditions and my computer is not a
cognitive agent—believe me…it is not! So his conditions are hardly sufficient for
cognition.7
In a similar vein, Aizawa has claimed that my criterion of the cognitive is satisfied by
Chatterbots, Look up Tables, and CD Players.8
This objection is an obvious example of the fallacy of false dilemma (Aizawa grasps one
horn of this false dilemma). Underlying this fallacy is a failure to understand condition (4) and
the analysis of intentionality that underwrites it.
The project is, of course, not to explain cognition in terms of cognition, but to explain it
in terms of intentionality.9 Cognition is explained as a specific form of intentional activity.
Intentional activity is understood in terms of the idea of a transcendental mode of presentation.
This, in turn, is understood in terms of the idea of revealing or disclosing activity. Disclosure is
primitive, although different forms of this are distinguishable. You may not like the analysis of
intentionality. You may positively hate it. But you can hardly accuse me of trying to explain the
cognitive in terms of the cognitive, thus ruling out the first horn of the dilemma. Moreover, this
account has one clear implication: cognition can occur only in subjects of intentional states
(subjects to whom there is disclosure), thus ruling out the second horn.
7
Adams (2010),p. 330.
See, for example, his blog, The Bounds of Cognition.
http://theboundsofcognition.blogspot.com/search/label/Rowlands
th
See the entry dated Saturday, April 10 , ‘Problems for Rowlands MoTC’)
9
To be fair to Fred, however, he was critiquing the (2009) version of the mark of the cognitive that did not come
with attached analysis of intentionality. I did, however, make more than a few noises in that paper about the
notion of ownership being both complicated and crucial, which might perhaps have given him pause.
8
10
Adams continues with his objections to my mark of the cognitive:
Rowlands’ condition (1) is just an information processing condition. Everyone and her
mother thinks that cognition involves information processing. So Rowlands puts that in.
But the important issue is the kind of information processing. It has to be processing of
semantic content.
Trees and plants process information (in the mathematical or
communication theoretic sense of information), but no one thinks they are cognitive. So
(1) is surely not sufficient and, in my view, not specific about whether it is mere
information or semantic content that is being processed and only that latter rises to the
level of the cognitive.
Yes, that must be it. Everyone and her mother thinks that cognition involves information
processing: that’s why I put it in. More seriously, the first point to note is that is that the content
processed certainly doesn’t have to be semantic in the sense of truth-evaluable. Mental models
don’t contain semantic content in this sense. Most cognitive processing is not processing of
semantic content. Semantic content is bound up, for familiar Davidsonian reasons, with the
logical connectives. But mental models – cognitive maps, etc – do not stand in the same sort of
relation to the connectives. The negation of a sentence is another specific sentence; but the
‘negation’ of a map is not another map. The disjunction of two sentences is another sentence. But
the ‘disjunction’ of two maps is not another map.
Most importantly, however, this objection is the result of a failure to understand the way
in which (1)-(3) relate to each other. Information processing – the manipulation and
transformation of information bearing structures – counts as information processing only because
it has the proper function of making available information that was previously unavailable. It
11
achieves this function by way of the production in a subject of a representational state. So, the
sort of information processed will be the sort of information contained in the representational
state. That is an entailment of (1)-(3) taken together. And, as I say when developing the criterion,
pick any sort of representation you like. Knock yourself out – EC still follows. Adams continues:
Rowlands’ second condition brings up the matter of proper function … The problem here
as well is that it is not just any functions that make up a mind, but the cognitive
teleological functions. So, unless we know which functions are at issue, this won’t much
help as part of a set of sufficient conditions for cognition.
If the previous objection was the result of a failure to understand the relations between (1)–(3),
this objection seems to result from a failure to even properly read condition (2). For, there, I say
exactly what function is involved: the function of making available, either to a subject or to
subsequent processing operations, information that was previously unavailable. Adams accuses
me of presupposing the notion of a cognitive function. I find this charge mystifying. Conditions
(1)-(3) collectively specify what a cognitive function is: the function of making available
previously unavailable information through the production in a subject of a representational state
(where subject is understood, along the lines of the analysis of intentionality, as an intentional
agent.
Adams continues:
Condition (3) also is not going to fill the bill. Of course there have to be representations
in a cognitive system, but as I pointed out above, representations exist in the circulatory
system, in the digestive system, in the control of blood sugar levels and thermoregulation
in the brain. So, just as referring to mere information rather than semantic content,
12
talking about representations rather than representations that can be true or false is the
wrong level to be part of a set of sufficient conditions for cognition.
First of all, once again, the restriction to truth and falsity would rule out a staple of cognitive
scientific practice – postulation of cognitive maps, mental models and the like. These cannot be
true or false, only accurate or inaccurate. But, far more importantly, as I say when developing the
criterion, you can help yourself to whatever conception of representation you want. Insert in
condition (3) any restrictions on representation you prefer, EC still follows.
Adams concludes with the final condition:
Condition (4) as I pointed out earlier, just begs the whole question. If we have a true
subject, a person, then we already have a cognitive agent. This can’t be a necessary part
of a set of sufficient conditions for a cognitive system…it would be sufficient all by
itself.
For this, I need only refer you to the response to the first objection. The project is to explain
cognition as a form of intentional activity, not to explain cognition in terms of itself.
6. Adams’ Mark of the Cognitive
Adams (2010) presents his own mark of the cognitive; one made up, purportedly, of four
conditions.
1. Cognitive processes involve states that are semantically evaluable
Again, I should note that if semantic evaluation is understood in its customary sense. Models and
maps are neither true nor false. Nothing much turns on this; we can broaden the notion to that of
13
accuracy-conditional, rather than truth-conditional. More importantly, note that this entailed by
my condition (3) – especially since I am willing to allow my opponent to understand the notion
of representation in whatever way he wants. Far from denying it, I endorse this claim. There is
nothing in, however, that precludes EC.
2. The contents carried by cognitive systems do not depend for their content on other minds
This is also compatible with EC. It is entailed by condition (3) of my criterion of the cognitive.
3. Cognitive contents can be false or even empty, and hence are detached from the actual
environmental causes
Here, Adams is running two things that are typically distinguished – misrepresentation and
decouplability. I actually think there is a deeper connection between the two than is commonly
acknowledged, and so applaud his instincts if not his clarity. But this is also entailed by condition
(3) of my mark of the cognitive. For any case of representation, you have to have the possibility
of misrepresentation. If you want decouplability too, that’s fine with me. Let’s build it into
condition (3) that the representation it invokes must be decouplable.
Finally these representational states must:
4. Cause and explain in virtue of their representational content
Yes, this is fine too. Let’s build this into condition (3) of my mark of the cognitive.
I have presented a mark of the cognitive that, I have argued is compatible with EC.
Adams claims his criterion is incompatible with EC. But his criterion is entailed by mine – or,
more accurately, by an interpretation of mine that I am more than willing to accept. Clearly, one
of us is going to be very disappointed.
14
Why does Adams think his criterion is incompatible with EC? The reason is that he has
smuggled in a fifth, tacit, condition:
Yes there are states that satisfy these conditions within the body and brain of the
“coupled” agent. But no, there are no good reasons yet to think there is cognition going
on everywhere along the causal chain. There are as yet no good reasons to think that
cognitive processing is occurring anywhere outside the head of the agent coupled to the
environment.10
That is:
5. Conditions (1)-(4) must apply everywhere in the process for the process to count as
cognitive.
And at last the general contours of the end game become visible.
7. End Game
I have advanced a view according to which cognitive processes are amalgamations of neural,
bodily, and environmental processes. Adams argues that states which satisfy his criterion of the
cognitive never occur outside the skull. But, it is the contribution of condition (2) that is the
decisive. The entries on Otto’s notebook are semantically evaluable (1), can be false or empty
and detached (3), and play a role in explaining Otto’s behavior (4). It is (2) and only (2) that
precludes the sentence from counting as cognitive according to Adams’ criterion. So, it all comes
down to non-derived content. A&A argue that non-derived representation never extends beyond
10
Adams (2010), p. 328.
15
the neural. Therefore, everything outside that is scaffolding – it facilitates but does not constitute
cognitive processes.
My first entry into the game will look something like this. Have you forgotten the
arguments I developed in my 2006 book, Body Language? I argued that non-derived content
does extend out into the world: the operations that we perform on the world when we manipulate
information bearing structures are representational – in the sense that they satisfy the usual
naturalistic criteria of representation – and where this representational status is non-derived. So, I
would simply deny A&A’s central claim. We do, in fact, find things with non-derived
representational status out in the world when we engage in the manipulation and transformation
of external information bearing structures.
A&A will, I suspect, have a likely response: this is not enough. This non-derived
representationality, if it is indeed real, attaches only to processes of manipulation, and not to the
structures manipulated. But that is what you need to satisfy our conditions: the structures
manipulated must have non-derived content.
My response will be obvious enough: why? I do not accept this response, since I do not
see that it has been grounded in any argument they have given. If parts of the external process
can have non-derived representational status, then why must the states or structures implicated
have this too? But I don’t want to stop playing the game yet. So, I henceforth propose to ignore
my own work. My follow up will be along entirely more familiar lines: a charge of questionbegging.
Would A&A be so confident in their claim when it applies to a neural process? Suppose,
for example, that we identify a neural process that involves the manipulation of non-derived
16
representations. Further on downstream, we identify another process of the same sort. But in
between, causally connecting these two processes, is a section of processing that does not
involve representations. We might think of it as a non-representational portion of the processing
stream. Would we really be so confident in the assertion that this portion of the stream is,
therefore, non-cognitive? It is causally shaped by the initial processing stream, and goes on to
causally shape, perhaps in important ways, the later processing stream. But it does not involve
representations and is, therefore, non-cognitive. I can see no basis for confidence in this
assessment. Certainly there is nothing in cognitive scientific practice that would yield any degree
of confidence.
Yes there is, A&A might respond: the rules and representations approach (which we have
assumed without argument and which therefore might leave us open to another charge of
question-begging), mandates that anything that is cognitive must be an operation performed on a
representation. And our criterion of cognition mandates that this representation must be nonderived.
This, I think is an interesting stage of the game. As far as I can see, the rules and
representations approach mandates nothing of the sort. In fact, the rules and representations
approach has the opposite implication: not every structure operated on by rules can be a
representation – and so the issue of whether it is derived or non-derived does not apply. This
follows from the intimate relation between rules and representations.
The retinal image, for example, is not a representation. It is not a representation because
there is no difference between what causes a pattern of light intensity values distributed over the
retina and what should cause it. The image is simply caused by whatever causes it. The image,
17
therefore, makes no normative claim on the world; and normativity is essential to representation.
Nevertheless processes operating on the image are, in classical (i.e. rules and representations)
approaches regarded as cognitive.
For example, to use the sort of example familiar from the work of Fodor, the same pattern
of intensity values can be caused by (a) a horse, (b) a donkey-in-the-distance. There is no sense
that can be given the claim that the retinal image should have been caused by a horse. We can’t
say, for example, that the retinal image should have been caused by a horse, because a horse is,
in fact, there? If so, what retinal image should the donkey have caused? The horse and the
donkey produce the same retinal image: that is why it is difficult to distinguish them.
Normativity does not enter the picture at this stage. There is no sense that we can give to the
claim that the horse should have produced one retinal image and the donkey another. The retinal
image is not a representation because there is no possibility of its misrepresenting – and without
the possibility of misrepresentation there is no representation. Normativity, hence the possibility
of misrepresentation, comes in only downstream.
But, I can imagine A&A saying, (because the second A did, in fact say this to me),11 the
retinal image is so too a representation. I am, in effect, looking for normativity in all the wrong
places. The retinal image can make a normative claim with respect to simpler quantities? For
example: Shape. Color. Brightness. Suppose, to take Aizawa’s example, there is something
wrong with the eye and as a result levels of illumination that would normally produce activation
levels of quantity Q only produce activation levels of 0.5Q. (In the visual subject, the world
11
See our altogether‐far‐too‐friendly‐to‐be‐really‐entertaining‐debate on Philosophy TV.
http://www.philostv.com/kenneth‐aizawa‐and‐mark‐rowlands/
18
appears dim). Can we not say that the eye should have registered activation level Q and not
0.5Q?
My response is that this is exactly the same situation as that of the horse/donkey.
Normativity can only emerge downstream. What we have here is only the illusion of normativity.
We might think there is normativity here because we have some sort of idea of the proper
function of the eye: it is supposed to register Q in the presence of ambient light of a certain level
of illumination. However, its proper function in this regard cannot be disentangled from the
nature of the downstream processing. If this were suitably different – for example, if there were
an augmentation function applied to the input values at a give stage of processing – then the eye
might have the proper function of registering 0.5Q in these circumstances, and not Q. The moral
is the exactly the same: the retinal image is caused by whatever causes it. Normativity emerges
only later, after processing has begun.
Normativity enters the picture in this way. After the causal impingements responsible for
the retinal image, the brain makes various ‘guesses’ about the sort of thing that might have
produced the image. On the Marrian account, for example, these guesses take the form of the
application of principles such as good continuation, closure, common fate, and so on. If the
world turns out contrary to these guesses, then the brain has guessed wrong. Now we have
normativity and, consequently, the possibility of misrepresentation. Therefore, the retinal image
is not a representation. In itself, it makes no normative claim on the world. Normativity arises
only through subsequent processing operations.
There is a more general point that underlies this example, more valuable than the
example itself. It is the application of transformational rules to a structure that transforms this
19
structure from one that merely carries information to one that represents. This is because the
application of rules transforms the structure from one that does not make a normative claim to
one that does. This connection between representation and rules is crucial. It is the application of
rules that creates representations. Therefore, these rules must, initially, operate on things that are
not representations.
Therefore, the claim that cognitive processes must operate only on non-derived
representations cannot be sustained. Cognitive processes operate on representations, but they
also operate on things that are not representations. To put the same point another way: the
architecture of cognition must consist in more than vehicles of content.
At this point in the game, there are, it seems two options available to A&A. The first
option is to try an identify something that the retinal image shares with genuine representations,
in virtue of which we could (i) legitimately regard operations formed on both as cognitive, and
(ii) rule our EC. Seemingly the only candidate here is non-derived information. Both the retinal
image and representations carry non-derived information. The recommended modification would
be this: cognitive processes must act only on structures that carry non-derived information.
This, however, would concede the case to EC. It might rule out cases of EC like Otto’s
notebook, but it concedes many other cases of EC. For example, manipulation of the optic array
in order to make available invariant information is compatible with this condition – for the optic
array carries non-derived information. So, too is the probing of visual structures more generally,
etc – the sorts of things that enactivists get excited about. This relaxation of the condition permits
much of what passes for extended cognition. In fact, since I understand EC in terms of the
20
existential claim explained earlier, you’ve already given me everything I want. We are no longer
arguing about whether EC is true, but simply about its scope.
The second option is to deny that the processing operations performed on the retinal
image are cognitive processes. Now, this is where a charge of question-begging would hit really
hard. The idea that the processes that transform the retinal image into the raw primal sketch – on
the Marrian way of looking at things – are cognitive processes is a deeply engrained component
of the classical rules and representations tradition. Cognitive processing starts with operations
performed on the retinal image. It does not start with operations performed on the raw primal
sketch. A significant part of the A&A case against EC turns on an appeal to what cognitive
scientists do – they employ representations with non-derived content, etc, etc. Adams’ conditions
(1)-(4) are based on an examination of cognitive scientific practice.
It is worth noting that appeals to cognitive scientific practice are skirting with questionbegging at the best of times. In the dispute between EC and classical approaches, the nature of
cognitive scientific practice is precisely what is in dispute. That, to some extent at least, is what
is up for grabs. But imagine how much this problem would be compounded if it was felt we
could ignore the appeal to cognitive scientific practice whenever it yielded conclusions we didn’t
like. This, I strongly suspect, would make the A&A position untenable.
8. The Real End Game?
I agree with A&A that understanding the nature of cognition is an important aspect of assessing
the claims of EC and its detractors. However, it is extraordinarily doubtful that we can
understand what cognition is in the absence of an account of a suitably worked out account of the
mental. In particular, cognitive scientific practice and the philosophy of that practice will not be
21
able to tell us whether manipulation of non-derived representations counts as cognition. This is
something that can be decided only by digging deeper. And ‘digging deeper’ involves providing
an account of what makes something mental.
Thus, the real source of the idea that cognition must involve non-derived representation
lies not in a theory of cognition, but in a theory of the mental. It is non-derived intentionality that
is the hallmark of the mental. To this extent, I suspect A&A would agree with me. However,
they have chosen to interpret this idea in a way unjustified by any argument they have given and
which seems to contradict the cognitive-scientific practice in which they claim to place such
great store. From the claim, which I accept, that:
(i)
Unless you have non-derived intentionality, you don’t have anything mental.
Non-derived intentionality is the hallmark of the mental.
They have inferred:
(i)
Every part of a mental process must consist in operations performed on structures
that are non-derived representations.
The intuition expressed in (i) is a good one. Its interpretation as (ii) is a non sequitur.
But how can we even assess these claims unless we know what intentionality is? In NSM
I argued for an account of intentionality as revealing activity. EC – extended cognition – follows
from this as an obvious, almost banal, consequence. As revealing activity, non-derived
intentionality permeates our dealings with the world, including those dealings we classify as
cognitive. Otto’s sentences may not have non-derived intentionality, but his manipulations of
these sentences certainly do. My analysis may be wrong. But I suspect that it is that this level
22
that the debate is going to have to be fought – that the end game really starts to be played. This
level is that of traditional philosophy of mind – which, for me at least, comes with a liberal
sprinkling of phenomenological elements – and not philosophy of cognitive science.
References
Adams, Fred (2010) ‘Why we still need a mark of the cognitive’, Cognitive Systems Research
11, pp. 324-31.
Adams, Fred and Aizawa, Ken (2008) The Bounds of Cognition (New York: Wiley-Blackwell).
Rowlands, Mark (1999) The Body in Mind: Understanding Cognitive Processes (Cambridge:
Cambridge University Press)
Rowlands, Mark (2006) Body Language: Representation in Action (Cambridge, Mass: MIT
Press).
Rowlands, Mark (2009) ‘Extended cognition and the mark of the cognitive’, Philosophical
Psychology 22,pp. 1-20.
Rowlands, Mark (2010) The New Science of the Mind: From Extended Mind to Embodied
Phenomenology (Cambridge, Mass: MIT Press).
Rupert, Robert, (2004) ‘Challenges to the hypothesis of extended cognition’, Journal of
Philosophy, 101, pp. 389-428.
23
Rupert, Robert (2009) Cognitive Systems and the Extended Mind (New York: Oxford University
Press).
24