Intuitively, justified true belief is not always knowledge

I admit it’s old news that JTB doesn’t intuitively equal K (50 years old this year in the Western tradition, about 1200 years old if you follow the Indo-Tibetan tradition), but if anyone still wants empirical confirmation, a little more is available here, in a paper co-authored with psychologists Valerie San Juan and Raymond A. Mar, forthcoming in Cognition.

We looked at a variety of Gettier cases (like the Russell/Scheffler stopped clock case) and at “Skeptical Pressure” cases (in which it’s mentioned that clocks are sometimes broken, and that although the clock was working the agent didn’t double-check this).  We found that laypeople do tend to have the intuitive reaction that these are not cases of knowledge, despite evaluating them as cases of justified true belief.  Unsurprisingly, but to my great relief, we didn’t find significant variation correlated with gender or ethnicity.

An earlier study in Cognition had suggested that laypeople only have the knowledge-denying response for false lemma Gettier cases, but we found the effect robust for both those cases and fake-barn type cases. The earlier study used a narrow range of cases of one particular type, in which an object is replaced with an indistinguishable replica; we used a broader variety of stories.

In any event, I wanted to share my report that crawling out of the armchair doesn’t necessarily leave one with the sense that it needs to be set on fire.


Intuitively, justified true belief is not always knowledge — 10 Comments

  1. Hi Jennifer,

    Congratulations on the Cognition acceptance. That’s really impressive!

    I was just wondering how you saw the relationship between the no-false-lemma view and the view advocated in the earlier Cognition paper, namely, that laypeople tend to ascribe knowledge in Gettier cases involving authentic evidence, where authentic evidence is (roughly) evidence that makes the belief true when (initially) based on it.

    Don’t their authentic-evidence Gettier cases involve false lemmas?

  2. Hi John,

    I find the earlier Cognition paper a little murky (although interesting). Starmans and Friedman say a few things about Gettier cases that I don’t agree with (e.g. that Gettier cases all share the feature that “the fact that justifies the belief… is causally disconnected from the fact that makes the belief true”). And I’m not sure that some of their Gettier cases are actually most naturally read as Gettier cases; I think they have a somewhat idiosyncratic view of what it is to be Gettiered (on which more below).

    S&F draw a contrast between apparent evidence, which they describe as “evidence that only *appears* to be informative about the world, but coincidentally leads to a true belief”, and authentic evidence, which isn’t like that (they don’t give any positive characterization of authentic evidence, but we gather the impression that it leads to non-coincidental true belief formation). They claim that laypeople follow a rule of K=JTB+authentic evidence in attributing knowledge. One of my worries concerns the phrase “informative about the world”, which needs to be understood in a way that won’t invoke the notion of knowledge (if the analysis is to be non-circular). In saying that apparent evidence only appears to be informative about the world, they are suggesting that it is not informative, but here they can’t mean that it does not lead to true beliefs about the world. Gettier cases do end up with the agent believing something true. So the work is being done by the rider that apparent evidence (presumably unlike authentic evidence) “coincidentally leads to a true belief”. This is worrisome for several reasons. First, perfectly good knowledge-supporting evidence can in fact lead by sheer coincidence to true beliefs (I trust you to come up with intuitive examples!) and second, the appeal to the coincidental looks to me like a mere placeholder for something explanatory. Insofar as we see coincidence (or luck) as a catch-all for “unexplained (good) outcome” and see true beliefs as ordinarily being formed by processes that ordinarily give rise to knowledge, then to say we have landed on an evidentially supported true belief by coincidence is just to say that we have managed to achieve justified true belief without knowledge. I think the analysis secretly relies on the term it is trying to analyze.

    A better way to make sense of what is going on in the paper is just to look at the examples. The ones that worked for them (to elicit knowledge denial in laypeople) were examples in which the agent initially forms a false belief (as in Havit/Nogot) and then later infers something true. They claim that in these cases the denial of knowledge was “not due to the agent being Gettiered” but just to the involvement of apparent evidence — where I’d want to say that what they’re calling the involvement of apparent evidence is itself just a really nice way of being Gettiered.

    The cases that didn’t work in eliciting knowledge denial are perhaps a bit tricky. Here’s an example: a manager in a bookstore puts a $20 bill into the cash register and then goes into the back room. While she is back there, an employee sees that the $20 is tattered, and replaces it with a fresh bill. Fill in the blank: The manager [really knows/only thinks] that there is a $20 in the cash register.

    My concern here is that the employee who makes the substitution has taken care to restore the world to fit the manager’s original state of mind (after all he is carefully replacing the bill, not stealing it). So I don’t think this case is necessarily a Gettier case at all (of course there are several ways of fleshing out the case, imagining what is going on in the mind of the employee and manager–but my first pass on it was to read it as a case in which the employee at the end knows that his manager knows there is a $20 in the till).

    The “Gettier cases” that didn’t work for S&F all have this kind of substitution-of-identicals structure. To their credit they do seem to be aware that it might be a problem that what they call the “double luck” events are causally connected, so they run what they take to be a control case in which there are two burglars, one of whom steals a blue Bic pen and the other of whom “absentmindedly leaves his own identical blue Bic pen” in just the same spot. It’s still not entirely clear that the story is really one in which there is knowledge loss — for example, it’s not stipulated whether the second burglar did or did not see the first one taking a pen, or take that as a reason to put a replacement there (however absentmindedly). We have to think about what would be most plausible as a construal of the story, for participants not expecting a Gettier case. It’s a strange coincidence that the second pen would be identical, and put in exactly the same spot; it’s arguably less strange that one thief would be seen as covering up for the other by trying to make the scene look as it did originally. If participants are construing the story this way, it’s not a Gettier case either. (Philosophers may take the effort to read a case as involving independent coincidences, as charitable readers of philosophical arguments aimed at showing the possibility of JTB without K: S&F are of course right that philosophers have proposed substitution cases like this as Gettier cases.)

    As a further concern, we may also in cases of this kind be putting a heavy burden on the mental state attribution powers of the layperson: stories in which a burglar steals a worthless object (why?) and puts an identical replica in its place (why??) are somewhat taxing already. It would be really interesting to see cases with more psychologically plausible narratives developed to test the points that S&F want to make.

    Meanwhile, I should also note that our paper mislabelled the two Gettier cases in Appendix A: we got the “authentic evidence” and the “apparent evidence” tags reversed. Many thanks to Christina Starmans for noticing that. We will fix it!

  3. Hi Jennifer,

    It seems to me that the basic difference between authentic evidence and apparent evidence is best construed as a difference between evidence that makes the belief true when (initially) based on it, and evidence that doesn’t do that. (This perhaps comes out best in the description of the piggy-bank cases, p. 278, right-hand column, end of the penultimate paragraph.) As best I can tell, that doesn’t smuggle knowledge in through the back door. Does it seem that way to you?

    I take S&F to have established the following: in an identifiable (even if somewhat fuzzy) range of cases where the protagonist perceptually detects the truth and justifiably continues to believe that truth over a short period of time, the default position for a lot of people – perhaps a small majority – seems to be that the protagonist continues to know, whether or not she’s Gettiered. Waxing figurative for just a moment, we might put the point this way: at least in the short term, the epistemic inertia of initially successful perception [and hence knowledge?] can, to some extent, inhibit Gettierization from having the expected effect. Do you think that’s a fair assessment of their findings?

    Incidentally, I’ve observed non-skeptical results with cases that don’t involve any other agents, other than the Gettier subject. Even in (authentic-evidence) cases where purely impersonal forces render the original basis false and in the process install a backup truthmaker (not described as “identical” or ending up “in the exact same spot”), it’s not easy to get participants to deny knowledge at rates exceeding chance. So I suspect that S&F’s results aren’t being driven by cognitive load caused by extraordinary ToM tasks. Moreover, S&F have false-belief control conditions for the same thief and two-thief scripts. Those don’t seem to be any less taxing, but participants overhwhelmingly deny knowledge there.

    Anyway, congratulations again on the Cognition acceptance! It’s great to see these important issues gaining the attention of the wider cog-sci community.

  4. I just noticed something else about the Emma case (which, I gather from the last paragraph of your previous post, is an authentic-evidence case – sorry if I got that wrong and this is a tangent).

    The script says that Emma “could not tell the difference between a real diamond and a cubic zirconium fake just by looking or touching.” That’s basically saying that she can’t know by perception that it’s a real diamond.

    Instead, what she has to go on is the standing testimony provided by the sign labeling the items “Diamond Earrings and Pendants.” However, “almost all of the pendants had cubic zirconium stones rather than diamonds,” due to the dishonest employee’s thievery.

    As I understand the case, then, (1) Emma’s belief that what she’s holding is a diamond is essentially based on what the sign says, (2) the sign says something false (i.e. the items in this tray are diamond), (3) what the sign says doesn’t make Emma’s belief true when she bases her belief on it, and (4) participants are all but explicitly told that Emma doesn’t know by perception that it’s a diamond.

    In light of all that, if S&F’s theory makes a prediction about the case, I suspect that it predicts a tendency to deny knowledge.

  5. Thanks for the comments, John.

    To take them in reverse order: on the Emma case, I don’t think an inability to tell the difference between a real diamond and a cubic zirconium just by looking or touching means that someone can’t see that a stone is a diamond, any more than the inability to tell at a distance whether something is a barn or a barn facade means that someone can’t know just by looking at a distance that something is a barn. You suggest that she forms her belief that the stone she is looking at is a diamond by inferring it from her false belief that all the stones in the tray are diamonds. I don’t think the case is most naturally read this way. If you focus on the proposition that the participants are being asked to evaluate–which concerns the (real) diamond she is holding–she makes her judgment about that proposition on the basis of authentic evidence, evidence that is informative about that slice of reality (the fact that the sign is misleading about other slices of reality is neither here nor there–the evidence is, as you say, making the relevant belief true when first based on it).

    You may be right that there’s some inertia with perceptual knowledge. What I’d resist in that comment is the suggestion that K is persisting through the agent being “Gettiered”. I don’t think we have a clear depiction of what it is to be “Gettiered” here. (I myself think that agents are Gettiered whenever they have JTB falling short of K; others may call agents Gettiered when there is a certain sort of separation in the causal chain producing a belief, but I think that’s not very satisfactory, in part because of cases involving deviant causal chains that seem to make a point very similar to the point Gettier was trying to make.)

    Lastly, I should also have emphasized that the apparent/authentic evidence contrast explanation of Gettier intuitions might run afoul of the kinds of cases Ted Warfield and Branden Fitelson have invented, cases in which someone does reason to K from a false belief.

    I did notice the high rates of knowledge denial for the FB controls. I think it’s easy to deny knowledge there for Gricean reasons (and this may be part of the reason why it’s harder to deny knowledge for lay participants on the difficult Gettier cases). When the story makes it clear that S believes that there is a pen on the coffee table, and that there is a pen on the coffee table, choosing the “only thinks” side of “S [really knows/only thinks] that there is a pen on the coffee table” does carry the unwanted implicature that there isn’t a pen there.

    Thanks again for your comments, and I look forward to seeing your experimental results. By the way, on perceptual knowledge preservation through dislocations and relocations, see the experimental mental state attribution literature, e.g. Study 2 of Kaminski, Call et al. 2008: “Chimpanzees know what others know, but not what they believe”, Cognition 109 (2).

  6. Hi, Jennifer. Thanks for your further thoughts on this!

    — I’ll check out that Kaminski et al. study. Sounds cool.

    — You can find a link to my results here.

    — Fair enough re perceiving diamonds. But now I’m curious: if we’re to understand the case in terms of perceiving that it’s a diamond, then why even include the false sign? And why tell participants that Emma “could not tell” diamonds from cubic zirconia? For what it’s worth, in my own case this all leads to an interpretation that the (false) signage is essential to Emma’s judgment. I guess it’d be easy enough to rule this out in a follow-up study.

    — “I myself think that agents are Gettiered whenever they have JTB falling short of K.” You don’t think lottery cases are Gettier cases, do you?

  7. Hi John,
    Good questions about the stimuli. The “could not tell” line is in there to try to keep the case variants as similar as possible to each other (I need it there for the “Skeptical Pressure” case, so I put it in all the variants other than the vanilla one). The signage isn’t wrong about what she selects, so this feels different to me than classic false lemma cases like Havit/Nogot, where there’s a clear false belief figuring in the inference to the true. The Emma case doesn’t seem especially inferential to me, which was why I wouldn’t see it as fitting what S&F say about apparent evidence, but I appreciate that there are different ways of reading it, and I agree that we could do follow ups. Thinking about some follow ups might also help disambiguate various things that could be meant by “apparent evidence”.

    In general I think it’s a really good idea to have a diverse pool of different types of case so we don’t risk confounding superficial features of a particular narrative with its epistemological import.

    For what it’s worth: I think that in the standard fair lottery case you are justified in believing that your ticket has a very low chance of winning, but not that your ticket will lose.

    Meanwhile, I’ll look at your results, thanks for the link.

  8. After Christina Starmans was in touch about our reversal of the “authentic” and “apparent” labels in the Appendix, we did a full review of all our uses of those terms in the paper, and discovered that in fact we’d also reversed them on page 18. So (unlike S&F) we did find that both authentic and apparent evidence Gettier cases triggered the majority of participants to deny knowledge, but (more in line with S&F) we found that denial was stronger in our authentic evidence cases than our apparent evidence cases. This is probably a good place to alert everyone to the fact that we are reporting odds ratios and not relative risk.

    We didn’t draw any conclusions from the relative strength of knowledge denial in our authentic vs. apparent evidence cases because we didn’t make any very systematic attempt to compare those two kinds of cases (and we’d probably need a clearer characterization of what S&F really meant by their contrast in order to do a good job of it). No doubt further work could be done here. If anyone would like to see our complete set of cases for the study, I’ve posted them here.

    Meanwhile, I’ve posted a corrected version of the paper (marked with today’s date). Thanks again to Christina for noticing!

Leave a Reply

Your email address will not be published. Required fields are marked *