For Contemporary Debates in Epistemology (3rd Edn.),
(eds.) B. Roeber, M. Steup, J. Turri & E. Sosa, (Oxford: Blackwell).
THERE CANNOT BE LUCKY KNOWLEDGE
DUNCAN PRITCHARD
University of California, Irvine
dhpritch@uci.edu
1. THE ANTI-LUCK INTUITION
The general idea that knowledge excludes luck¾the anti-luck intuition¾is widely endorsed in
epistemology. In order to understand the appeal of this claim, we first need to unpack what it
amounts to. Clearly it does not mean that knowledge is never acquired by luck. Consider
circumstantial epistemic luck, which is when it is a matter of luck that one is in epistemically propitious
conditions (such as being lucky to have the evidence that one does). Epistemic luck of this kind is
entirely compatible with knowledge. A lucky discovery is still knowledge, after all. For instance, it
can be a matter of luck that one happens to be looking out of one’s window when one did, and
hence was able to clearly see a murder occur across the street, but this doesn’t in itself prevent one
from knowing that this murder occurred. Or consider propositional epistemic luck, which is when the
proposition that one believes is only true as a matter of luck. This is also entirely compatible with
knowledge. It may just be a matter of luck that I wasn’t shot in the head by a sniper’s bullet a few
moments ago, but that doesn’t in itself prevent me from knowing that this is the case (e.g., by
witnessing the bullet fly by without harming me).
The anti-luck intuition thus concerns a specific kind of epistemic luck: that if one has
knowledge, then one’s belief, so formed, isn’t true as a matter of luck. This is known as veritic
epistemic luck.1 Consider the knowledge-undermining epistemic luck that is in play in Gettier-cases,
2
and which prevents the subject’s justified true belief from amounting to knowledge.2 A farmer
justifiably believes that there is a sheep in the field because of the sheep-shaped object that she can
see over in the field. Her belief is also true, in that there is a sheep in the field. Unfortunately, what
she is looking at is not a sheep at all, but a big hairy dog that is obscuring from view the genuine
sheep behind.3 The idea is that this isn’t knowledge because the subject’s belief, so formed, while
true and justified, is nonetheless only true as a matter of luck. It is, after all, simply happenstance
that there is a genuine sheep hidden from view behind the sheep-shaped object that our agent is
looking at (and which supports her justification for her belief). What Gettier-style scenarios
demonstrate is thus that the tripartite account of knowledge in terms of justified true belief is
inadequate to the task because it doesn’t exclude cases of veritic epistemic luck. Accordingly, this
debate displays the anti-luck intuition in action.
Gettier cases are not alone in this regard, though they are a particularly clear-cut instance of
the anti-luck intuition at work. For example, although the intuition is not quite as universal as the
Gettier intuition, most epistemologists agree that one cannot know that one’s lottery ticket is a
loser simply by reflecting on the astronomically long odds involved in winning.4 This is prima facie
puzzling, given that the likelihood that one’s belief is true given one’s evidence is extremely high,
and certainly much higher than lots of everyday beliefs which do qualify as knowledge. Indeed, it
seems as if finding out the lottery result by reading it in a national newspaper, even if one were
unaware of the astronomical odds involved, would suffice to enable one to know that one’s ticket
is a loser, even though intuitively it is more likely, given one’s evidence, that one’s belief is false.
The natural explanation for why the lottery belief doesn’t amount to knowledge is again the
presence of veritic epistemic luck. In particular, if one forms one’s belief that the ticket is a loser
simply by reflecting on the odds involved, then it seems to be merely a matter of luck that one’s
belief is true. After all, one could just as well have won the lottery, and if so one’s belief would be
false, even though one’s evidence for its truth would be just as strong. In contrast, in a case where
one has seen the result printed in a national newspaper it doesn’t appear to be merely a matter of
luck that one’s belief is true. Had one won then what is printed in the newspaper would have been
different, and hence one would have continued to have a true belief (albeit now the belief that one
won rather than lost). Again, then, we find that what is apparently motivating an epistemological
discussion is the anti-luck intuition that knowledge excludes veritic epistemic luck.
3
2. MODAL EPISTEMIC CONDITIONS
AND VERITIC EPISTEMIC LUCK
In order to further unpack the anti-luck intuition, we need to consider what kind of epistemic
principle captures it. One candidate in this regard is the requirement that knowledge entails sensitive
true belief, in the sense that it is a true belief formed on a basis such that, had what the agent
believed been false, she wouldn’t have believed it on that same basis.5 Conversely, if, in the closest
possible world in which the target proposition is false, the agent continues to believe this
proposition on the same basis, then her belief is insensitive, and doesn’t amount to knowledge.
We can see the plausibility of this principle at accommodating veritic epistemic luck by
considering how it deals with Gettier-style and lottery cases. While the farmer’s belief happens to
be true, in the closest possible world where what is believed is false—where the real sheep hidden
from view has wandered out of the field, but everything else stays the same—the farmer will
continue to believe that there is a sheep in the field regardless (as she will still be looking at the
sheep-shaped object). The insensitivity of the belief thus captures the veritic epistemic luck
involved.
Or consider the lottery case. If one forms one’s belief that one has lost the lottery by simply
considering the astronomical odds involved, then one’s true belief, so formed, is insensitive. In the
closest possible world where the believed proposition is false—where one wins the lottery, but
everything else stays the same¾our agent will continue to believe that the ticket is a loser on the
same basis. In contrast, reading about a lottery result in a national newspaper would generate a
sensitive belief, as in the closest possible world where one wins the lottery one would read a
different result, and hence form the belief that one has won instead.
One could thus plausibly contend that the sensitivity condition on knowledge captures the
anti-luck intuition by excluding veritic epistemic luck. Interestingly, however, there is another
modal principle that can accommodate these kinds of cases equally well. This is the requirement
that knowledge entails safe true belief, in that one has a true belief which could not have easily been
false.6 The safety principle is usually cashed-out as the claim that one has a true belief such that, in
close possible worlds, if one continues to form a belief on the same basis as in the actual world,
then one’s belief continues to be true.
Safety and sensitivity are closely entwined principles; indeed, if counterfactual conditions
contraposed (they don’t), then they would be equivalent. It shouldn’t be surprising then that safety
generates the same results as sensitivity for the cases under consideration (and many others
besides). The farmer’s belief is unsafe in that there is a close possible world where she continues to
form her belief on the same basis of seeing the sheep-shaped object and yet forms a false belief as
4
a result (i.e., the very same possible world that we considered above, where the sheep has
wandered out of the field, but everything else stays the same). Similarly, forming one’s belief that
one has lost the lottery by reflecting on the astronomical odds involved wouldn’t deliver a safe
belief, as there will be a close possible world where one continues to form a belief on this basis
and forms a false belief as a result (again, this is the very same close possible world that we
considered above, where the ticket happens to be a winner). In contrast, forming one’s true belief
that one has lost the lottery by reading the result in a national newspaper would generate a safe
belief, as in all close possible worlds where one continues to form a belief on this basis, one would
form a true belief. While lottery wins happen in close possible worlds (all it takes is that a few
coloured balls fall in a slightly different configuration), it is not a close possible world where a
national newspaper prints the wrong lottery result, as they have sophisticated checks in place to
ensure that such an embarrassing event doesn’t occur.
One could thus just as well unpack the notion of veritic epistemic luck in terms of the safety
principle as via the sensitivity principle. Is there a way to choose between them? This is usually
done by considering how the principles fare more generally. For example, safety seems to do
better than sensitivity when it comes to accounting for inductive knowledge, as even an excellent
inductive basis for a belief that p can be such that one would form a false belief on this basis in the
closest not-p world. For example, while away on holiday one has excellent inductive reasons for
believing that one’s home hasn’t been demolished. Nonetheless, in the closest possible world
where one’s house has been demolished—where a bizarre chain of events culminates in it being
demolished¾one would continue to believe that it is still standing regardless, given that one is
away on holiday. One’s belief is thus insensitive, even despite its excellent inductive basis. It
seems, then, that making sensitivity a requirement for knowledge would lead to us denying that
some inductive beliefs that have a solid epistemic pedigree count as knowledge.7
In contrast, notice that safety doesn’t have this problem with inductive knowledge. What is
causing the problem in the case just described is that sensitivity is taking us to the closest not-p
possible world, no matter how far out that world might be. In this particular scenario, the not-p
possible world is significantly removed from the actual world, as it describes an implausible chain
of events that would lead to something occurring that wouldn’t normally occur. That’s precisely
why it seems that the subject has inductive knowledge of the target proposition, even despite the
insensitivity of their belief, as far-fetched scenarios like this seem to be irrelevant to whether one
has knowledge. Appealing to safety doesn’t have this problem, since it is only concerned with what
is happening in close possible worlds. That’s why our protagonist’s belief that his house has not
been demolished is safe, as there is no close possible world where he continues to form a belief on
this basis and forms a false belief.
5
That’s a count in favour of safety over sensitivity, and of course if the former is a more
plausible principle than the latter, and they can both equally accommodate our intuitions about
veritic epistemic luck, then it would make more sense to opt for safety as the anti-luck condition
on knowledge.
3. ANTI-LUCK/RISK EPISTEMOLOGY
AND VERITIC EPISTEMIC LUCK
One way of adjudicating between safety and sensitivity is thus to see how each principle fares in
terms of accommodating a range of cases. There is also a second way of proceeding on this score,
however, which I think is more helpful at understanding why we should prefer the one principle
over the other when it comes to capturing the anti-luck intuition. This approach involves
determining what is meant by luck and then applying to it our thinking about the anti-luck
intuition. I call this methodology anti-luck epistemology.8 As we will see, this manner of tackling the
issue also brings with it reflections on the nature of risk¾and thus epistemic risk¾too, given the
close connections between the notions of luck and risk.9
We’ve already noted that the anti-luck intuition is most plausibly captured by two modal
principles about knowledge, safety and sensitivity. The modal aspect of these principles should not
be surprising given that luck itself seems to be a modal notion. Consider a paradigm case of a
lucky event, such as a lottery win. This luckiness consists in the fact that there are close possible
worlds where the relevant initial conditions for that event remain fixed (e.g., one continues to
purchase a lottery ticket, the lottery is run in the same manner, and so on), where one does not
win the lottery.10 In contrast, events that aren’t lucky, such as the sun rising this morning, don’t
satisfy such a condition; indeed, in this case, the target event obtains in all close possible worlds.
More generally, we can measure the extent to which an event is lucky by just how close the
relevant possible world is in which the target event doesn’t obtain. For example, I am very lucky to
survive the sniper’s bullet if the relevant possible world where I am fatally hit is especially close,
but less lucky if it is further off. We can thus account for why nearly being hit by a sniper’s bullet
from a few inches away is, ceteris paribus, luckier than nearly being hit by a sniper’s bullet from a few
feet away.
This is the so-called modal account of luck.11 Note that this proposal doesn’t distinguish
between the valence of a lucky event—i.e., whether it is good or bad luck. So, for example, losing a
lottery is also a lucky event, albeit bad luck rather than good luck. After all, there are also relevant
close possible worlds where one wins a lottery (we will come back to this feature of lotteries in a
6
moment).
Importing the modal account of luck into our notion of veritic epistemic luck lends support
to safety over sensitivity. Our focus is now the event of forming a true belief, and we are
interested in whether there are relevant close possible worlds where one ends up forming a false
belief. Knowledge is thus meant to exclude veritic epistemic luck in just this sense. But that’s just
to say that when one knows there is not a close possible world where one’s belief, so formed,
results in a false belief. And that’s just safety.12
Note too that by focussing on the modal account of luck we have an explanation of why our
modal principles are basis-relative, in the sense that one needs to keep the actual basis fixed when
making assessments of whether a belief is safe of sensitive. This aspect of these principles is
independently motivated by appeal to cases.13 For example, we saw above that whether a belief
that one’s lottery ticket is a loser is formed on the basis of reflecting on the odds involved or by
reading the results in a national newspaper is relevant to whether it counts as being subject to
veritic epistemic luck, and thus either safe or sensitive. Insofar as we understand veritic epistemic
luck in terms of the modal account of luck, then we should expect such basis-relativity. This is
because it’s important to a luck assessment to keep the relevant initial conditions for the target
event fixed, and in the epistemic case that means keeping the basis fixed.14
By bringing the modal account of luck into the discussion we can also better explain what is
going on in the lottery case. How can a true belief that is overwhelming likely to be true, given
one’s evidence, fail to amount to knowledge, even though a corresponding true belief, which isn’t
overwhelmingly likely to be true given one’s evidence, does amount to knowledge? The answer lies
in an important feature of modal closeness, which is that low probability events can occur in close
possible worlds. That’s why it can both be a matter of (good) luck that one wins the lottery and
also a matter of (bad) luck that one loses the lottery. In the latter case, even though the probability
that one wins the lottery is astronomically low (and thus the probability that one loses is
astronomically high), there is nonetheless a close possible world where one wins the lottery.15
Once we grant, in line with the anti-luck intuition, that knowledge excludes veritic epistemic luck,
and thus demands safety, then it becomes unsurprising that one fails to know that one’s lottery
ticket is a loser simply by reflecting on the odds involved. This is because even despite the very
high likelihood that one’s belief is true, given one’s evidence, it is nonetheless the case that
forming a belief on this basis will lead to a false belief in close possible worlds (i.e., the close
possible world in which one wins the lottery). In contrast, where one forms one’s true belief that
one’s ticket is a loser by reading the result in a national newspaper, then one’s belief can amount to
knowledge, as there won’t be a close possible world where a belief formed on that basis will be
false. The crux of the matter is that what’s important to whether one’s belief is subject to veritic
7
epistemic luck is not the probabilistic likelihood that one’s belief is true, given one’s evidence, but
whether one’s basis for belief ensures that one forms a true belief across relevant close possible
worlds.
That the combination of the modal account of luck and veritic epistemic luck leads to safety
also explains why safety is better able to deal with inductive knowledge than sensitivity. For all that
matters in excluding veritic epistemic luck is that one’s basis for belief doesn’t lead to false belief
in close possible worlds; whether one’s basis for belief leads to false belief in far-off possible
worlds is irrelevant to an assessment of veritic epistemic luck. This is what the sensitivity principle
gets wrong, and why in particular it struggles with inductive knowledge, since it makes what one
believes on the same basis in the closest not-p possible world relevant to whether one has
knowledge, even if that not-p world is far-off. In contrast, safety doesn’t have this difficulty
precisely because it is only concerned with what the subject believes on the same basis in the
modal neighbourhood.
Importing the modal account of luck into veritic epistemic luck doesn’t just favour safety
over sensitivity, but it also tells us how to understand the safety condition. For example, just how
intolerant is safety to false belief formed on the same basis in close possible worlds? All of them,
or only most of them? (And in the latter case, how exactly is one meant to unpack that claim?)
Recall, however, that the modal account of luck captures the idea that luck comes in degrees in
terms of the closeness of the non-obtaining of the target event in relevant possible worlds.
Applied to the epistemic case, we can thus see how there can be a gradual increase in tolerance to
epistemic luck as one moves further out into the modal neighbourhood and beyond. To maintain
that knowledge is incompatible with veritic epistemic risk is to treat it as completely incompatible
with false belief on the same basis in very close possible worlds and compatible with false belief on
the same basis in non-close possible worlds, with a continuum of gradual compatibility as one
moves from the former to the latter. As one might put the point, not every false belief in the
modal neighbourhood carries the same weight, as it also depends how close the possible world is
too. With this in mind, we can accommodate the thought that there might be some tolerance to
veritic epistemic luck when the possible worlds where false beliefs are formed on the same basis
are fairly close but not very close. In such cases we would expect our intuitions to be mixed (as I
think they often are).
We noted earlier that combining the modal account of luck with veritic epistemic luck
explains the basis-relativity of a principle like safety, but it also suggests a particular way that this
basis-relativity should be understood. We might antecedently be tempted by a version of safety
where we only consider close possible worlds where the subject believes the very same proposition on
the same basis. That’s the most straightforward way to read the claim that one has a true belief
8
that couldn’t have easily been false, even though nothing in that claim entails this particular
reading. Such an interpretation is problematic, however, as becomes clear once we consider
propositions that are true across all close possible worlds, such as necessary propositions. For such
propositions there is trivially no close possible world where one falsely believes them on the same
basis as the actual world, since there is no close possible world where they are false. On such a
reading one’s belief in such a proposition would thus be automatically safe. Intuitively, however,
this is not what we are trying to capture when we formulate the safety principle.
The problem is the artificial restriction to the actual proposition believed. With the modal
account of luck in mind, what we are really after is rather a basis for belief such that there is no
close possible world where it generates false belief. That claim, however, doesn’t keep the actual
proposition believed fixed in the relevant modal assessment at all, as one might on the same basis
believe a different proposition in close possible worlds. So, for example, consider a case where
someone believes that ‘2 + 2 = 4’ as a result of flipping a coin. While on the restrictive reading this
epistemically dubious basis generates a belief that is automatically safe, this is not the case on the
less restrictive reading that we just proposed. For while there are obviously no close possible
worlds where the actually believed proposition is false, there are close possible worlds where this
basis for belief generates false beliefs, such as the belief that ‘2 + 2 = 5’.
The modal account of luck thus offers a way of spelling out the safety principle, and thereby
the notion of veritic epistemic luck. Notice too that this approach also enables us to accommodate
some important intuitions that are relevant to this debate concerning epistemic risk. The reason
for this is that risk is a closely related notion to luck, and indeed for that reason amenable to a
similar modal analysis. In particular, we can think of risk assessments as primarily differing from
luck assessments in terms of the perspective that one takes on a particular event, especially in
terms of whether one adopts a forwards-looking or backwards-looking assessment. So, for
example, I’m lucky to have avoided the sniper’s bullet (backwards-looking) because there is a
relevant close possible world where the bullet hits me. Conversely, I was at a high risk of being
shot by the sniper (forwards-looking) because there is a close possible world where the bullet hits
me. Relatedly, as with luck, it is important to risk to capture the sense in which even unlikely
events can be risky. So, for example, playing Russian Roulette with only one bullet is still highly
risky even though the odds are significantly in one’s favour, as there is a relevant possible world
where you are shot that is very close.16
Given the close connections between luck and risk, it is unsurprising that we are just as
strongly inclined to treat knowledge as incompatible with high levels of veritic epistemic risk as we
are high levels of veritic epistemic luck, as they essentially amount to the same thing—viz., an
aversion to forming beliefs on a basis that could easily lead to error (false belief). Anti-luck
9
epistemology and anti-risk epistemology thus go hand-in-hand. Indeed, one might even go so far
as to suggest that risk has the conceptual ‘whip-hand’ here, in the sense that the reason why we
care about eliminating veritic epistemic luck is because we care about eliminating veritic epistemic
risk rather than vice versa. That is, we are averse to the idea of lucky knowledge in the relevant sense
precisely because we are averse to the idea of risky knowledge.17
Finally, note that in construing the anti-luck intuition as safety we are only claiming that
safety is a necessary condition on knowledge. I think we can recognise fairly quickly that this must
be so. Safety captures a certain kind of modal profile that a belief should have it is to be
knowledge, but there can be ways that such a modal profile might be exhibited that are
epistemically problematic. For example, imagine a subject who is forming her beliefs in
epistemically dubious ways, but where an external agent is ensuring that any beliefs, so formed, are
guaranteed to be true. There would be no epistemic luck (or epistemic risk for that matter)
involved in the agent forming true beliefs in this way, but clearly such beliefs do not amount to
knowledge, as the safe cognitive success on display doesn’t in any significant way reflect the
cognitive performance of the subject (but is rather simply due to the intervention of the other
agent). There is thus more to knowledge than simply getting to the truth in a non-lucky (or nonrisky) way. In particular, it is important that the non-lucky (safe) cognitive success is attributable in
some significant way to the cognitive agency of the subject rather than to other factors.18
4. PUTATIVE COUNTEREXAMPLES
Several putative counterexamples have been put forward to the idea that knowledge excludes
veritic epistemic luck, and hence entails safety. One kind of counterexample to safety trades on the
notion of circumstantial epistemic luck that we noted above, and which we saw is compatible with
knowledge. Consider the following case that is offered by Peter Baumann (2014). Baumann begins
by noting that one cannot gain knowledge of the time by looking at a broken watch, even if one
happens on the right time, due to the epistemic luck involved. But Baumann adapts this case by
imagining someone who sets their (working) watch by using the broken watch, oblivious to the
fact that it is broken. Can this person subsequently know what the time is on their watch?
Baumann poses this problem by considering a range of time intervals. So just after setting their
watch, Baumann thinks that it is clear that one cannot gain knowledge of the time, as it is just too
lucky that one’s belief is true. But what about several hours later, or several days or weeks later? At
some point it seems that we would be willing to ascribe knowledge of the time to our agent. And
yet, argues Baumann, all that has happened in the interim is that time has passed, and he claims
10
that the mere passage of time shouldn’t make a difference here. Accordingly, either one has
knowledge in both cases even despite the luck involved, or one lacks knowledge in both cases due
to luck. And, of course, if one opts for the first horn, then one is committed to there being cases
of knowledge that are unsafe and hence veritically epistemically lucky.
I think the error in Baumann’s reasoning is in his claim that the passage of time doesn’t
make a difference in this regard, as clearly it can (as indeed his cases show). Effectively what
Baumann is offering us is a transition from a case of veritic epistemic luck to one where there is
only circumstantial epistemic luck in play. If one happened to previously set one’s reliable watch
by unknowingly using a broken watch that by luck gave the right time, then that is circumstantial
epistemic luck. If one now properly forms one’s belief about the time by using one’s working
watch, then one’s true belief so formed is not a matter of veritic epistemic luck—so formed, it
couldn’t have easily been false. In contrast, forming a belief about the time by looking at a broken
watch, even if the belief is true, is subject to veritic epistemic luck, as this belief, so formed, could
easily have been false. What Baumann’s cases illustrate, however, is that there are bound to be
cases where the distinction between circumstantial and veritic epistemic luck is not sharp.
For example, if I only set my watch by using the broken watch a few moments ago, then is
that circumstantial epistemic luck or veritic epistemic luck? The difficulty here is that temporal
proximity to the broken watch makes us inclined to treat the use of the broken watch to set the
time as being part of the basis for the belief formed by looking at the working watch. That’s why I
think we are tempted to treat this as veritic epistemic luck rather than circumstantial luck. Increase
the temporal gap between the setting of the watch in this way and the use of the working watch to
determine the time, however, then I think we are now inclined to treat this as a straightforward
case of circumstantial epistemic luck. Contra Baumann, the passage of time clearly does make a
difference, and it does so because it has a bearing on where we draw the line in terms of what is
part of the basis for the belief and what is rather part of the prior conditions that are subsequently
being epistemically exploited by the subject. Inevitably, as with most distinctions with real-world
application, there will be threshold cases where it is hard to judge which side of the distinction a
case should fall. But that is only to be expected, and certainly doesn’t show that the distinction is
illusory.
A second kind of counterexample trades on an important difference between two ways in
which a belief can be subject to veritic epistemic luck. I’ve elsewhere characterized this distinction
in terms of intervening and environmental epistemic luck.19 Standard Gettier-style cases like the farmer
case we considered above fall into the first category. They concern scenarios where something gets
in between the subject’s justifiable believing and the truth of what is believed, such that there is a
complete disconnect between the two. In the farmer case, for example, although the farmer
11
justifiably and truly believes that there is a sheep in the field on the basis of what she sees in the
field, her justifiable believing is disconnected from the truth because she doesn’t actually see a
sheep at all.
While all Gettier-style cases involve veritic epistemic luck, it isn’t always of an intervening
kind. Consider instead the famous ‘barn façade’ case.20 In this scenario we have someone looking
at the front of a barn and forming the belief that what they are looking at is a barn. The twist in
the tale is that, unbeknownst to our agent, in this area almost everything that looks like a barn is in
fact merely a barn façade, with our agent looking at the one real barn in the vicinity. This is a case
of veritic epistemic luck, as the subject’s belief, so formed, could have easily been false. That is,
there are close possible worlds where the subject is looking not at the genuine barn but at one of
the many nearby barn façades and hence forming a false belief as a consequence.
Some commentators have argued that knowledge is possessed in cases of environmental
epistemic luck, even despite the veritic epistemic luck involved.21 One rationale for this approach
is that the target belief enjoys positive epistemic support that is lacking in parallel scenarios of
intervening epistemic luck. In the barn façade case, for example, our agent is actually seeing the
barn in question, and hence is in cognitive contact with the target fact in a way that is absent in
cases of intervening epistemic luck. Relatedly, it also seems plausible to say that in cases of
environmental epistemic luck (and in contrast to intervening epistemic luck) the subject’s belief
amounts to a cognitive achievement, in the sense that her cognitive success is attributable to her
manifestation of cognitive agency.22 The idea that cases of environmental epistemic luck are more
credible candidates for knowledge than cases of intervening epistemic luck thus has some
plausibility. Nonetheless, there is a substantial cost to allowing that environmental epistemic luck
is compatible with knowledge (i.e., as opposed to merely saying that such cases can enjoy certain
kinds of positive epistemic support that are not applicable to cases of intervening epistemic luck).
In particular, it involves conceding that knowledge can be had even in cases where there is a high
degree of veritic epistemic risk—viz., where one’s belief, so formed, carries a high risk of being
false.23 Allowing that cases of environmental epistemic luck count as knowledge thus carries a high
degree of epistemic revisionism.
Many putative instances of unsafe knowledge involve environmental epistemic luck. For
example, Ram Neta and Guy Rohrbaugh (2004) offer a case where a glass of water could very
easily have been poisoned but wasn’t (and had it been, it wouldn’t have been detectable).24 They
claim that one can nonetheless know that what one is drinking is unadulterated water, even though
it’s just a matter of luck that one’s belief is true. Although they don’t characterise it as such (as this
terminology came later), this is a case of environmental epistemic luck, for while nothing
intervenes between the agent and the target fact (concerning the unadulterated water), the subject
12
is nonetheless in an environment where the taste of water is not a good guide to whether it is
unadulterated water (just as in the barn façade case one is in an environment where barn frontages
are not a good guide as to whether the structure is a barn). That’s why the belief, so formed, could
have very easily been false. Once we understand that such a case is an example of environmental
epistemic luck, however, then while we can agree that there are positive epistemic factors in play
that are absent in cases of intervening epistemic luck, we can also see that the high level of veritic
epistemic risk in play makes it an implausible candidate for knowledge, at least unless one is willing
to embrace a substantial form of epistemic revisionism.25,26
13
REFERENCES
Baumann, P. (2014). ‘No Luck With Knowledge? On A Dogma of Epistemology’, Philosophy and
Phenomenological Research 89, 523-51.
Becker, K. (2007). Epistemology Modalized, London: Routledge.
Black, T., & Murphy, P. (2007). ‘In Defense of Sensitivity’, Synthese 154, 53-71.
Black, T. (2008). ‘Defending a Sensitive Neo-Moorean Invariantism’, New Waves in Epistemology,
(eds.) V. F. Hendricks & D. H. Pritchard, 8-27, Basingstoke: Palgrave Macmillan.
Chisholm, R. (1977). Theory of Knowledge, (2nd ed.), Englewood Cliffs, NJ: Prentice-Hall.
Cohen, S. (1998). ‘Contextualist Solutions to Epistemological Problems: Scepticism, Gettier, and
the Lottery’, Australasian Journal of Philosophy 76, 289-306.
Comesaña, J. (2005). ‘Unsafe Knowledge’, Synthese 146, 393-402.
Coffman, E. J. (2015). Luck: Its Nature and Significance for Human Knowledge and Agency, London:
Palgrave Macmillan.
Dretske, F. (1970). ‘Epistemic Operators’, Journal of Philosophy 67, 1007-23.
¾¾ (1971). ‘Conclusive Reasons’, Australasian Journal of Philosophy 49, 1-22.
Engel, M. (1992). ‘Is Epistemic Luck Compatible with Knowledge?’, Southern Journal of Philosophy
30, 59-75.
Goldman, A. (1976). ‘Discrimination and Perceptual Knowledge’, Journal of Philosophy 73, 771-91.
Hales, S. D. (2016). ‘Why Every Theory of Luck is Wrong’, Noûs 50, 490-508.
Hawthorne, J. (2004). Knowledge and Lotteries, Oxford: Oxford University Press.
Hetherington, S. (1998). ‘Actually Knowing’, Philosophical Quarterly 48, 453-69.
¾¾ (2016). Knowledge and the Gettier Problem, Cambridge: Cambridge University Press.
Hiller, A., & Neta, R. (2007). ‘Safety and Epistemic Luck’, Synthese 158, 303-14.
Kelp, C. (2009). ‘Knowledge and Safety’, Journal of Philosophical Research 34, 21-31.
Lackey, J. (2008). ‘What Luck is Not’, Australasian Journal of Philosophy 86, 255-67.
Levy, N. (2011). Hard Luck: How Luck Undermines Free Will and Moral Responsibility, Oxford: Oxford
University Press.
Navarro, J. (2019). ‘Luck and Risk’, Metaphilosophy 50, 63-75.
Nelkin, D. (2000). ‘The Lottery Paradox, Knowledge, and Rationality’, Philosophical Review 109, 373408.
Nozick, R. (1981). Philosophical Explanations, Oxford: Oxford University Press.
Neta, R., & Rohrbaugh, G. (2004). ‘Luminosity and the Safety of Knowledge’, Pacific Philosophical
Quarterly 85, 396-406.
Pritchard, D. H. (2005). Epistemic Luck, Oxford: Oxford University Press.
¾¾ (2007). ‘Anti-Luck Epistemology’, Synthese 158, 277-97.
¾¾ (2009). ‘Apt Performance and Epistemic Value’, Philosophical Studies 143, 407-16.
—— (2012). ‘Anti-Luck Virtue Epistemology’, Journal of Philosophy 109, 247-79.
¾¾ (2014). ‘The Modal Account of Luck’, Metaphilosophy 45, 594-619.
¾¾ (2015a). ‘Anti-Luck Epistemology and the Gettier Problem’, Philosophical Studies 172 (2015),
93-111.
¾¾ (2015b). ‘Risk’, Metaphilosophy 46, 436-61.
¾¾ (2016). ‘Epistemic Risk’, Journal of Philosophy 113, 550-71.
¾¾ (2020). ‘Anti-Risk Virtue Epistemology’, Virtue-Theoretic Epistemology: New Methods and
Approaches, (eds.) J. Greco & C. Kelp, 203-24, Cambridge: Cambridge University Press.
Pritchard, D. H., Millar, A., & Haddock, A. (2010). The Nature and Value of Knowledge: Three
Investigations, Oxford: Oxford University Press.
Roush, S. (2005). Tracking Truth: Knowledge, Evidence and Science, Oxford: Oxford University Press.
Sainsbury, R. M. (1997). ‘Easy Possibilities’, Philosophy and Phenomenological Research 57, 907-19.
Sosa, E. (1999). ‘How to Defeat Opposition to Moore’, Philosophical Perspectives 13, 141-54.
14
(2007). A Virtue Epistemology: Apt Belief and Reflective Knowledge, Oxford: Oxford University
Press.
Unger, P. (1968). ‘An Analysis of Factual Knowledge’, Journal of Philosophy 65, 157-170.
Williamson, T. (2000). Knowledge and its Limits, Oxford: Oxford University Press.
¾¾
15
NOTES
1 I think Engel (1992) was the first to use the terminology of veritic epistemic luck. The general idea that there are
specific ways in which knowledge can be lucky and ways in which it excludes luck goes back to at least Unger (1968),
though note that Unger’s terminology is in terms of accidentality rather than luck. See Pritchard (2005, passim) for a
sustained development of different kinds of epistemic luck, including varieties that are not knowledge-undermining.
2
Most epistemologists grant that, in general at least, agents lack knowledge in Gettier-style cases, but for a prominent
detractor, see Hetherington (1998; 2016). See also Hetherington’s contribution to this volume. As we will see below,
some epistemologists are sceptical about particular kinds of Gettier-style cases.
3 See Chisholm (1977, 105).
4 For some prominent examples, see Cohen (1988), Nelkin (2000), Williamson (2000, passim), and Hawthorne (2004,
passim).
5 For the key texts in this regard, see Dretske (1970; 1971) and Nozick (1981). For some recent texts which
sympathetically explore the sensitivity principle, see Roush (2005), Becker (2007), Black & Murphy (2007) and Black
(2008).
6 Versions of safety-type principles have been offered by a number of authors, including Sainsbury (1997), Sosa
(1999), and Williamson (2000), and Pritchard (2005; 2007).
7 The case that is usually offered to illustrate this point is Sosa’s (1999) garbage-chute example.
8 See Pritchard (2005; 2007; 2012).
9 We thus get a variation on anti-luck epistemology: anti-risk epistemology. See Pritchard (2016; 2020).
10 Henceforth, when I write of the ‘relevant’ possible worlds with regard to luck assessments, I have in mind those
possible worlds where the target initial conditions are kept fixed.
11 I first developed the modal account of luck in Pritchard (2005), but have since refined the proposal—see Pritchard
(2014). Where there is debate about the modal account of luck, it tends to concern how best to unpack this idea and,
relatedly, where there are further elements that must be added to a particular modal profile to deliver a complete
theory of luck. See, for example, Levy (2011) and Coffman (2015). For some recent dissent regarding the modal
account of luck, see Lackey (2008) and Hales (2016).
12 As explained in Pritchard (2016; 2020), matters are slightly more complicated than that (albeit not in ways that
concern us here).
13
Most famously, Nozick’s (1981, 179ff.) ‘grandmother’ case.
14 Note that basis-relativity is never understood such that the truth of the belief is part of the basis. More generally,
while we keep the basis fixed in modal assessments of epistemic luck, we do not also keep the truth of the target
proposition fixed. The appeal to the modal account of luck explains why, since it would be bizarre to allow the initial
conditions kept fixed in assessments of luck to include the target event itself (e.g., such that we assess whether a
lottery win is lucky by considering only worlds where one continues to win the lottery). (Note that later on we will be
considering the somewhat different kind of scenario where a belief is formed on a basis such that it is guaranteed to
be true across all close possible worlds. As we will see, beliefs so formed might not amount to knowledge, but if so
this isn’t because the anti-luck condition isn’t met).
15 Indeed, that’s arguably why people play lotteries but don’t otherwise bet on far-fetched events with astronomical
odds of success, since usually in the latter case there is no close possible world where the bet pays off.
16 For further discussion of the notion of risk and how it relates to luck, including some of the differences between
these two notions, see Pritchard (2015b). See also Navarro (2019).
17 For more on anti-risk epistemology, including some of the (relatively minor) ways that it differs from (and arguably
in the process improves upon) anti-luck epistemology, see Pritchard (2016; 2020).
18 Elsewhere I have referred to the intuition in play here as the ability intuition, which I claim is one of the two ‘master
intuitions’, alongside the anti-luck intuition, that governs our thinking about knowledge. Since I further claim that
these two intuitions impose distinct, but overlapping, constraints on a theory of knowledge, I maintain that we
therefore need a particular theory of knowledge that I refer to as anti-luck virtue epistemology (or, more recently, anti-risk
virtue epistemology). See Pritchard, Millar & Haddock (2010, ch. 3) and Pritchard (2012; 2016; 2020).
19 See, for example, Pritchard, Millar & Haddock (2010, ch. 2) and Pritchard (2012).
20 See Goldman (1976), who credits the example to Carl Ginet.
21 The most prominent example is Sosa (e.g., 2007, passim).
22 I take this to be an important insight from Sosa’s (e.g., 2007, passim) work on apt performances (which roughly
equate to cognitive achievements in the manner just outlined). Sosa shows that apt performances can be modally
fragile in just this environmental sense, and then applies that insight to the specific case of cognitive performances
and, thereby, knowledge. For critical discussion of this idea, see Pritchard (2009).
23 It thus follows that knowledge and cognitive achievement come apart, in that only the latter is compatible with
veritic epistemic luck. For more on this point, see, for example, Pritchard (2009; 2012) and Pritchard, Millar &
Haddock (2010, ch. 2).
24
See also the very similar case offered in Hiller & Neta (2007, 310-11). A structurally similar, though much more
complex, putative counterexample to the necessity of safety for knowledge is offered by Comesaña (2005, 397). As
Kelp (2009) persuasively argues, however, it isn’t obvious that Comesaña’s example even involves an unsafe belief in
16
the first place. In any case, insofar as this is an instance of unsafe true belief, it will be susceptible to the same response
that I offer below.
25 See also Kelp (2009), who offers a novel ‘Frankfurt-style’ case as a supposed example of unsafe knowledge where,
essentially, a demon would have intervened to ensure that the subject believes a certain proposition (regardless of its
truth), but in the end doesn’t have to as the agent believes it anyway (and the proposition is true). The belief is thus
subject to veritic epistemic luck, as there are nearby possible worlds where the belief, so formed, would have been
false, which is why it doesn’t amount to knowledge (even though, qua instance of environmental epistemic luck, it has
some positive epistemic properties that intervening cases of epistemic luck lack). Note that if the demon’s role had
been to ensure that the belief so formed was true, then while it still wouldn’t have been knowledge, the reason now is
obviously not that it involves veritic epistemic luck but rather that the subject’s safe cognitive success doesn’t stand in
the right explanatory relationship to her manifestation of cognitive agency. For more on epistemic Frankfurt-style
cases, see Pritchard (2015a).
26 Thanks to Stephen Hetherington and Blake Roeber.