Knowing What You Don't Believe

(by Blake Myers-Schulz and Eric Schwitzgebel)

Virtually every introduction to epistemology (online examples include the Stanford Encyclopedia and the Internet Encyclopedia of Philosophy) highlights the debate about what is commonly called the “JTB” theory of knowledge – the view according to which for some subject S to know some proposition P, it is necessary and sufficient that

(1.) P is true.

(2.) S believes that P is true.

(3.) S is justified in believing that P is true.

According to the JTB theory, knowledge is Justified True Belief. Perhaps the most-discussed issue in the last 40 years of epistemology is whether the JTB theory is true. Debate generally centers on whether there is a way of interpreting or revising the third (justification) condition or adding a fourth condition to avoid apparent counterexamples of various sorts (e.g., Gettier examples). Nearly all contemporary analytic philosophers endorse the truth of conditions (1) and (2): You can’t know a proposition that isn’t true, and you can’t know a proposition that you don’t believe. Few assumptions are more central to contemporary epistemology.

However, we – Blake and Eric – don’t find it intuitive that (2) is true. We think there are intuitively appealing cases in which someone can know that something is true without believing (or, per Eric, determinately believing) that it is true. We have four examples:

(A.) The unconfident examinee (from Colin Radford 1966, one of the very few deniers of (2)): Kate is asked on an exam to enter the date of Queen Elizabeth’s death. She feels like she is guessing, but enters the date correctly. Does Kate know/believe that Elizabeth died in 1603?

(B.) The absent-minded driver (from Schwitzgebel in draft): Ben reads an email telling him that a bridge he usually takes to work will be closed for repairs. He drives away from the house planning to take an alternate route but absent-mindedly misses the turn and continues toward the bridge on the old route. Does Ben know/believe that the bridge is closed?

(C.) The implicit racist (also from Schwitzgebel in draft): Juliet is implicitly biased against black people, tending to assume of individual black people that they are not intelligent. However, she vehemently endorses the (true and justified, let’s assume) claim that all the races are of equal intelligence. Does Juliet know/believe that all the races are intellectually equal?

(D.) The freaked-out movie-watcher: Jamie sees a horror movie in which vicious aliens come out of water faucets and attack people, and she is highly disturbed by it, though she acknowledges that it is not real. Immediately after the movie, when her friend goes to get water from the faucet, Jamie spontaneously shouts “Don’t do it!” Does Jamie know/believe that only water will come from the faucet?

In each case, we think, it is much more intuitive to ascribe knowledge than belief.

So, naturally (being experimental philosophers!), we checked with the folk. We used fleshed-out versions of the scenarios above (available here). Some subjects were asked whether the protagonist knew the proposition in question. Other subjects were asked whether the protagonist believed the proposition in question.

The results came in as predicted. Across the four scenarios, 75% of respondents (90/120, 1-prop z vs. 50%, p < .001) said that the protagonist knew, while only 35% said the protagonist believed (42/120; 1-prop z vs. 50%, p = .001). Considering each scenario individually, in each case a substantial majority said the protagonist knew and in no scenario did a majority say the protagonist believed. (A separate group of subjects were asked “Did Kate think that Queen Elizabeth died in 1603?” [and analogously for other scenarios]. The “think” results were very close to the “believe” results in all scenarios except for the unconfident examinee where they were closer to the “know” results.)

We think epistemologists should no longer take it for granted that condition (2) of the JTB account of knowledge is intuitively obvious.

[Cross-posted at The Splintered Mind.]

Comments

[Mark Phelan]

Hi Eric (and Blake),

I think these results of yours are pretty interesting, but, as I discussed in my comments on Blake’s paper at the XPS meeting at the Eastern APA, I don’t think they clearly support your conclusion, that belief isn’t required for knowledge. (Well, maybe your conclusion is just that epistemologists shouldn’t take it for granted that it’s intuitively obvious that belief is required for knowledge. But I wouldn’t have thought anyone needed an argument for that! So, I’ll discuss the more interesting potential conclusion.) I think this is an area where a pragmatic explanation of the empirical results is warranted.

I worked this out in detail at the APA, but the basic idea rests on recognizing the different ways in which people in ordinary conversation use the term “belief” and the ways in which these uses differ from the (at least) quasi-theoretical account of belief epistemologists are endorsing when they claim that, intuitively, belief is required for knowledge. I take it that what epistemologists mean by “belief” in (2) is something like the account the Stanford Encyclopedia attributes to contemporary analytic philosophers of mind: “[They] generally use the term "belief" to refer to the attitude we have, roughly, whenever we take something to be the case or regard it as true.” It’s pretty clear that ordinary usage of the term “belief” can depart pretty widely from this. For example, a resistance fighter in occupied France may say, “I believe in the liberty of the French people.” But it’s pretty obvious he’s not claiming that the French people are liberated.

In the same way, and more to the point, I think the term “belief” is sometimes reserved in ordinary conversation for a very strong conviction that something is the case. But this should not be the notion of “belief” epistemologists are endorsing when they claim that belief is required for knowledge. I know and (in the theoretical sense) believe that my car is parked on Grove Street, but crime being what it is in New Haven, I have a much weaker conviction that this is the case than I do that this is a table (which I also know and believe). In other words, there are degrees of belief, and sometimes the term “belief” is reserved for strong belief. (But knowledge doesn’t come in degrees, so we don’t ever reserve the term “knowledge” only for strong knowledge. Consider: “I strongly believe p,” and, “I somewhat believe p,” are each more apt than either, “I strongly know p,” or, “I somewhat know p.”) When, in ordinary conversation, people use the term “belief” to pick out strong beliefs, they are using the term to pick out a more restricted class of mental states than most epistemologists would (or, at least, should) think is required for knowledge.

All that being said, I think the specific problem with your studies is this: They prime people to use a restricted notion of belief (strong belief). The agents in the studies lack a strong belief state that the target propositions obtain. So, participants respond that these agents do not believe the target propositions. There is no correspondent priming in the case of knowledge. So, when asked if the agents knew, subjects respond that they did.

Now, I particularly hate when people just invoke pragmatics to explain away results without pointing to a particular pragmatic mechanism, so here’s my story about how your vignettes prime people to use a restricted notion of “believes”: Your vignettes (I think) present people with a clear case of a justified true belief. When you ask, after such vignettes, did, for example, Kate know that Queen Elizabeth died in 1603, you put participants in a position to provide a maximally informative answer. Roughly, you put them in a position to respond whether they think the case of justified true belief you’ve just presented is a case of knowledge. But, when you present the same case and ask is this belief, if they interpret belief broadly, then your question doesn’t allow them the opportunity to make an informationally sufficient contribution to the conversation. Each of your cases is clearly a case where an agent has at least a weak attitude that the relevant proposition is the case, i.e. if the agent has a belief (in the sense of interest to epistemologists). Who could be interested in finding out if people think that? But, you could plausibly be interested in finding out if participants think the agent has a fairly strong belief that the relevant proposition obtains. So, they interpret “believe” as “strongly believe”, and most people think Kate doesn’t strongly believe Queen Elizabeth died in 1603. In other words, if you were asking participants if Kate believes in the general sense, you would be setting them up to violate Grice’s maxim of quantity. Being good little conversational participants, they interpret your question in a way that doesn’t require them to break a conversational maxim.

In fact, for the XPS session I tested this explanation by running a quick study (with Jonathan Phillips’s help) on Mturk. I asked people in one condition, “Did Kate believe that Queen Elizabeth died in 1603?” In the other, I asked, “Did Kate believe on some level that Queen Elizabeth died in 1603?” The second formulation would force them away from the interpretation available in the first, and to a notion of belief familiar to the epistemologist (that invoked in 2). If I was right, we should expect a similar proportion of people to say Kate believed on some level as say she knows—many more than would answer affirmatively to the believes-question. This was what I got. 35% claimed Kate believed—a proportion similar to what you and Blake found. But 80% of participants responded affirmatively to, “Did Kate believe on some level that Queen Elizabeth died in 1603?” A proportion similar to the 88% you and Blake found for the knowledge question (for that particular vignette, as reported at the APA). The difference between the two questions I asked was statistically significant: x2(1, N = 40) = 8.3, p = .004

So, I think there’s some support for my pragmatic explanation of your results—enough, anyway, to block us concluding from those results that belief isn’t (even intuitively) required for knowledge (in the way epistemologists mean).

(Sorry this comment was long. As I mentioned, I’d already thought and written about this, and am glad to have another venue to discuss it.)

[Eric Schwitzgebel]

Thanks for the thoughtful, detailed comment, Mark! I regret not having had a chance to chat with you about this at the APA.

I am not sure what it is to "strongly believe". One possibility is confidence. Although Kate lacks confidence, Juliet can be read as highly confident. (In the original Schwitzgebel 2010, this is clear, and we could easily tweak the scenario to bring out the confidence more strongly. Would you predict that doing so would flip the results?) Ben and Jamie, also, would be highly confident if they paused for a moment to reflect. Maybe they momentarily lack confidence? I don't know I am sure to interpret that. When I am not thinking about the roundness of the Earth, which is most of the time, do I lack confidence in its roundness?

Alternatively, "strongly believe" might mean something like "disposed to act in accord with that belief across a broad variety of situations". If so, then I at least (I can't speak for Blake) would agree with your analysis of the folk response, only adding that it's not clear that the philosophical view of belief is, or should be, any different.

I don't see the pragmatic problem with our questions. If we were asking a question with an *obviously* true or false answer, that might force an alternative interpretation, but I take it is not obviously true in all these cases that the protagonist believes; they seem like cases on which opinion might reasonably divide. (The knowledge cases seem more like obviously yes cases to me, but that problem is the reverse of the problem you assert.)

Also, "believing on some level" seems to me a strange sort of attitude. If I were a respondent, I would be torn about whether to ascribe belief in such cases. "Believing on some level" seems like a weak, qualified way of putting it, so I would probably agree to it. "Believing on some level" might be a form of in-between believing. Thus, my own philosophical theory of the relationship between belief and knowledge in these cases (I ascribe determinate knowledge, in-between belief) can easily accommodate it.

[David Chalmers]

Cool. Three other hypotheses that come to mind: (i) A lot of people use 'believe' for "merely believe", i.e. believing without knowing. (ii) A lot of people use 'believe' (at least in these cases) for "occurrently believe", while using 'know' more broadly to include dispositional states [this one works better for ABD than for C]. (iii) A lot of people in these scenarios tie 'know' to a report-disposition (along with a justification+ condition) while tying 'believe' to a more demanding action-disposition. One could also combine e.g. elements of (ii) and (iii).

I suppose that (i) and (ii) could easily be accommodated by the JTB proponent (but perhaps you could easily rule them out by testing some control scenarios), while (iii) would require some more fancy footwork.

[Eric Schwitzgebel]

Interesting possibilities, Dave!

On (i): In piloting we tried giving each subject several scenarios with both "know" and "believe" prompts. We found that most subjects answered y/y or n/n to know and believe on the majority of scenarios (though most also answered y/n on at least one scenario). We also saw scenario order effects. We decided it would be cleaner to do an entirely between-subjects design. The fact that y/n was a minority response to most scenarios in piloting suggests against the "merely believe" hypothesis. (It is also one reason we don't come out strongly in favor of saying that these scenarios are intuitively knowledge without belief.) But your question makes me wonder if we should more formally do a within-subjects test to rule out the possibility.

On (ii): Right, I think that move won't work for C. It also seems a little dubious for D where it is a narrow temporal window at issue, making the occurrent/dispositional distinction less compelling.

On (iii): This is not far from my own view of the matter, although I don't think a report-disposition is quite right for K. We have some data from another scenario suggesting that we can get K attributions where there is at least a temporary disposition to counter-report (the protagonist emotionally declares the opposite of what it seems she really knows). We thought that scenario slightly problematic (for independent reasons) so we haven't included it in our analysis, but the K/B split is essentially the same as in the other scenarios.

My own view is that K has a lower dispositional bar, roughly *capacity* (instead of [modal] *tendency*, for belief) to appropriately deploy the information in question, although knowledge unlike belief requires justification and truth.

[Thomas Nadelhoffer]

Eric and Blake,

I, too, am sorry I didn't get to hear your presentation in New York! That being said, I wanted to suggest that self-deception may be another case where it makes sense to say P knows that x is the case but P doesn't (or won't/can't) believe that x is the case. For instance, a parent repeatedly finds clear and convincing signs of her teenager's drug use. But rather than accept and admit what she knows to be the case, she engages in self-deception--whatever that involves (see Mele's Self-Deception Unmasked for a convincing empirically-informed account)-- so that she no longer believes it to be the case. So, while deep down she knows her teenage is using drugs, she has somehow hidden or shielded herself from this dark and painful truth. Instead, she clings to the less painful belief that is everything is fine.

Now, I realize self-deception is a puzzling phenomenon--so, it might not be the sort of case you want to rely on in motivating your rejection of 2. But I, for one, would be really curious to see some studies on lay intuitions concerning self-deception!

[Eric Schwitzgebel]

Thanks, Thomas! We're inclined to agree. In fact, we already have a self-deception scenario that we are considering running in a follow-up.

[Fritz Warfield]

Knowledge that P, on the traditional JTB account, does not require that one "believe that P is true" nor does it require that one be "justified in believing that P is true".

The necessary conditions in these neighborhoods are, instead:

that one believe P
and
that one be justified in believing P.

One can believe P without believing that P is true. Neither knowing that P now believing that P requires having a concept of truth.

I'll be surprised if when I check the sources you cite they make the same error about the JTB account. If they do they need to be corrected.

That point is not a point about whether knowledge requires belief and whether your examples show or do not show that knowledge requires belief.
As for the examples, as I understand them so far I'm not moved. With some it's a problem of form -- one doesn't make a case for knowledge without belief by giving a case in which knowledge is more comfortably attributed than belief. That's consistent with both being present and consistent with both being quite plausibly attributed.

The old Radford example may well be a case in which one knows but doesn't believe that one knows. But that's not a case of knowledge that P without belief that P.

[Jonathan Weinberg]

I basically agree with Mark's take on things, except I don't think that this needs to be taken as semantic vs. pragmatic, so much as multiply-semantic, i.e., there are a range of different meanings for "believe" in ordinary parlance, and different ones will get cued up in different contexts. And really, this needn't even be different _meanings_ per se; it could just be different stored exemplars, or different "concepts on the fly", that get cued up, even if all have the same ultimate semantic value, i.e., beliefs.

Also, fwiw, I find "believes on some level" to be pretty idiomatic, and a touch of googlistics backs this up somewhat: 31,300 hits for "believes on some level", 329,000 for "believe on some level". About 3.4 million hits for "on some level, I believe", but scanning the top 20 a lot of them are spurious, like "...on some level. I believe...". I do find belief-on-some-level more intelligible than in-between-belief; the former is rendered sensible by taking the mind to be substantially disunified, and one system can represent p as being the case even when other systems don't, and so one can believe p at the level of the one system, while not believing it at the other. We can cash out different levels in terms of different representational systems, but I don't know how to cash out in-between belief.

Also, I don't see why it should be a problem for epistemologists if, in fact, some weaker sort of belief than determinate belief is sufficient for knowledge. The epistemological community on the whole is generally committed to _some_ sort of belief requirement on knowledge, but why take them to have a similarly strong commitment to any particuluar form of belief?

Which brings me to one disagreement with what Mark said, here: "maybe your conclusion is just that epistemologists shouldn’t take it for granted that it’s intuitively obvious that belief is required for knowledge. But I wouldn’t have thought anyone needed an argument for that" Actually, I think that many philosophers do indeed take it to be intuitively obvious that belief is required for knowledge. So clearly those philosophers, at least, deserve an argument!

[Eric Schwitzgebel]

Thanks for the comment, Fritz. You are right about the potential difference (on some accounts) between "believes that P" and "believes that P is true". "Believes that P" is the more accurate way to put it -- though perhaps slightly less comprehensible to non-specialist readers of the blog. Thanks for the reminder.

I would expect that only a minority of philosophers would be moved by our examples. We acknowledge being in the minority opinion about this. Our aim is not to provide examples that we think everyone will find intuitively compelling but rather to provide examples on which there may be a legitimate difference of opinion.

We also acknowledge that -- as in any psychological test -- there are alternative possible explanations of our results. We find a vague gesture toward the possibility of alternative explanations unpersuasive. We think that the simplest, most appealing explanation for the results across the four scenarios is that there are cases that ordinary folk are inclined to regard as cases of knowledge that they are also not inclined to regard as cases of belief.

I agree with you that believing that you know is not a requirement for knowing. Our view is not meant to be structured on that premise.

[Eric Schwitzgebel]

Thanks for those thoughts, Jonathan! My view is not too far from yours, I think, except on one point. (I can't speak for Blake here.)

First, I agree that "believes" -- probably like most terms for complex affairs -- does not have a single, sharp semantic content. I think it can be sharpened up for philosophical use in a variety of ways, and practical considerations should govern one's philosophical decision about how to sharpen it up if at all. (I'm not far here from Carnap on explication.)

On "believes on some level" (which is where I disagree with you most): Generally speaking, I'm against thinking of the mind in terms of functionally-distinguishable representational compartments in which language-like representations (or picture-like or map-like, if you prefer) are stored and retrieved. I agree that the compartments view has some resonance in folk psychology (e.g. "part of me" believes this, part of me believes that), but that's not an aspect of folk psychology I take very seriously. As a metaphor, I find it problematic especially as a way of dealing with what I call "in-between" cases. As serious cognitive psychology, I think it is sometimes successful, especially as a first-pass type model, but in the long run my guess is that connectionism and dynamic systems theory and associationist networks come closer to real structure of cognition. (I have written about these issues repeatedly, as you may know, most recently in McGeer & Schwitzgebel 2006.)

On whether our findings are a problem for epistemologists: I agree that there must be some some of belief-like or belief-ish condition on propositional knowledge. If one accepts Blake's and my view of of belief, it's still very plausible that some weakening of (2) is a condition of knowledge. And maybe that is all most epistemologists need for most of their projects. But I think that needs to be evaluated.

On your last point: That is exactly the force that I think these arguments have. I don't share the taken-for-granted intuition that knowledge that P implies belief that P (which is highly relevant to my work on in-between cases of believing). Mainly, I want *permission* to retain the contrary intuition as a non-crazy intuition, at least pending some more serious argument.

(It occurs to me that this is a third use of experimental philosophy on intuitions, distinct from the positive and negative programs you and others describe. I've been thinking I might work up a separate post on that.)

[Mark Phelan]

Thanks for the thoughtful reply, Eric! I think I wasn’t sufficiently clear about the pragmatic story I have in mind. Suppose that the traditionalist is right and intuitively knowledge entails belief. Then, any intuitively obvious case of knowledge is an intuitively obvious case of belief. We both agree that these are intuitively obvious cases of knowledge—and, indeed, participants responses seem to reflect this as well. What you want to test is whether people think knowledge entails belief. But, notice, if they do, (as I’m assuming on behalf of the traditionalist they do) then people are in a position to conclude that these are obvious cases of belief.
Now the particular way in which I think your belief question sets people up to be less than maximally informative doesn’t have to do directly with the obviousness of the state being either knowledge or belief. Rather, it has to do with the specificity of the answer you set them up to give in each condition. When you ask them, “Is this knowledge?”, people are in a position to identify the state in a very specific way. When you ask, on the other hand, “Is this belief?”, were they to interpret belief broadly (in the sense of interest to epistemologists) the question puts them in a position where they cannot be sufficiently specific about the state at hand. Since they have readily available distinct uses of the term “belief”, they opt for one of these other interpretations to be more specific in their identification of the state at hand.

I think what’s going on in the knowledge/belief cases is roughly analogous to the following: Jack’s team was playing football on the old field behind the high school. The field was fairly level: no obvious holes or hills. Jack had forgotten his cleats, and while trying to make a tackle he slipped and fell. While laying on the ground, he took a closer look and noticed that all around him there were slight perturbations in the surface of the field.

Is the old field flat? Having read this scenario, I’m inclined to say no. Is the old field flat for a football field? I’m inclined to say yes. Is there any sense in which the old field is flat? Yes. Why would I answer that the old field isn’t flat to the first question, when I think that it is, in some sense, flat? Because I’m perfectly accustomed to using different senses for flat, and interpreting your question using a particular (restricted) one of these puts me in a position to be more informative than if I use another. This, I think, is basically what’s going on in your cases (it’s just that we have a lexical entry for “knows” as opposed to “flat for a football field”.)

(That’s the basic idea. I’m about to jump on a non-wifi train. I’ll think about other comments and maybe say something about those later.)

[Mark Phelan]

Just to be clear, I don’t think the sort of interpretation I was suggesting for the believes question—Does Kate strongly believe?—is an implication of what’s meant. I agree with Jonathan that it’s semantic, in the sense that it contributes to the earliest available proposition in the relevant instances of linguistic processing. I just accept radical pragmatic intrusion into semantics. So I think something like the conversational maxim pushes participants towards the particular ad hoc concept of “believes” for the particular questions. (So, basically, I think, we don’t disagree at all about this Jonathan, and I just shouldn’t have introduced the messy semantics/pragmatics issue.)

As for the claim that that no one needs an argument that epistemologists shouldn’t take it for granted that it’s intuitively obvious that belief is required for knowledge, I guess I was just registering my opinion that no one gets to take the intuitive obviousness of entailment claims (knowledge entails belief) for granted in philosophy. Many may, but I don’t think they deserve an argument that they shouldn’t. If anyone raises the issue, they’re required at least to give some sort of justification for the purported intuitive obviousness (and if you teach epistemology, you’re probably accustomed to doing so). But, as a practical matter, an argument that forces many epistemologists to defend the entailment may be worthwhile.

[Eric Schwitzgebel]

Thanks for clarifying, Mark! I now feel more of the pragmatic pull that you are worried about. I am willing to grant this: "Believe" is not the maximally specific ordinary-language description of the attitude, so it may be a less natural response, pragmatically, than "know". I don't think the "flat" case is quite parallel, but maybe something more like asking of the 1st place winner whether she placed at least 3rd.

If people shift to another sense of "believe" on the basis of such relatively light pragmatic pressures, that other sense of "believe" must be a fairly natural one. I'm happy with that: There is a fairly natural sense of "believe" -- my own *preferred* sense, philosophically -- on which knowledge is not sufficient for belief. So the claim that K -> B is ambiguous, and it is true on only one sense of "believe". Then there will be a need to defend that sense of "believe" as the best one for the purposes of philosophy. That's exactly the result I want.

Keith Lehrer and Robert Shope do not need convincing that K -> B requires arguing for. But I find the arguments in favor to be weak and usually to collapse into intuitions about cases, intuitions that diverge from my own intuitions. And most philosophers seem to think that they can just toss out K -> B as an intuitively obvious, undeniable premise. And least, that's what they tend to do when I say things, in my work on in-between cases of belief, that imply the falsity (or perhaps indeterminate truth value) of it!

[Eric Schwitzgebel]

P.S. to Mark re. the first paragraph: What I am granting is that there is *a* disambiguation or sharpening of "believe" on which....

[Olgalednichenko]

[1] Bush is True

[2] Karl Rove belives [1] is true

[3] karl rove defends and justifies - all truths

:)

he is on a book tour too

regards
olga lednichenko

[David Zoller]

Interesting ideas here! I think I agree with your original train of thought, and I'd be very interested in the results of further XP along these lines.
To my mind, however, the most useful experiment here would not be asking people their intuitive responses to your cases--which invites all the problems by previous posters on how ordinary people construe the ambiguous term 'believe'--but rather to see what the average person is herself willing to say that he or she 'believes' and 'knows.'

Consider instead making use of the survey situation to prove your point.

In giving somebody a survey, you already have a scenario where a normal person is pulled out of their comfort zone and their day-to-day, and asked to pony up what she 'believes'. My point is that you've already broken some of the eggs--why not make an omelet? If you can show some meaningful results that a significant percentage of people refuse to 'believe' x when asked, despite the fact that something like x seems to inform other responses this could be fascinating.

While I'm not particularly familiar with XP, I work in other fields--on mind and the social supports for knowledge--and I'm very interested in hearing more precise results on this. As an outsider, I'm less interested in surveys of what the hoi polloi think and significantly more interested in surveys that demonstrate how they think.

[Eric Schwitzgebel]

Thanks for the suggestion, David! I agree that something like that would be an interesting follow-up.

[Eric Schwitzgebel]

I am closing this post to new comments, since the comments are directed to an email account I use only sporadically.

For further comments and discussion, please go to the parallel post at The Splintered Mind:
http://schwitzsplinters.blogspot.com/2010/03/knowing-what-you-dont-believe.html