Gettier and the folk

It seems like a good day for my first post to the blog.

A number of readers might be interested in this paper, which reports new experimental findings on laypeople's take on Gettier cases. Comments welcome. Title and abstract:

A conspicuous art: putting Gettier to the test

Professional philosophers say it’s obvious that a Gettier subject does not know. But experimental philosophers and psychologists have argued that laypeople and non-Westerners view Gettier subjects very differently, based on experiments where they ascribe knowledge to Gettier subjects at statistically significant rates. I argue that when effectively probed, laypeople and non-Westerners unambiguously agree that Gettier subjects do not know.

Interested parties will also want to check out this recent paper, if they haven't already. (You really should have already.)

Comments

[Marcus Arvan]

Very cool studies, John. Have you given any thought to doing similar studies in the philosophy of language (to test earlier studies claiming that non-Westerners have different intuitions in Kripke/Godel cases)?

[John Turri]

Thanks, Marcus. Yes, it had occurred to me that a related maneuver could be used whenever it's claimed that various folk disagree with what (some) philosophers claim is obvious. It won't always work, but it's worth a shot. Anyone with the time and inclination should feel free to try. (I probably won't have time to do it for a while.)

[Eddy Nahmias]

Really fascinating studies, John! I have to think more about them, but I look forward to hearing what others have to say, especially those whose earlier results are powerfully challenged by these results.

Can you offer your thoughts about what explains these earlier results? Presumably, you think the studies were set up so that people were making some sort of mistake(s) or being influenced by pragmatic features (your b and c). I thought a big part of the differences would derive from your use of "only thinks he knows" rather than the previously used "only believes" but it looks like that's not a main issue.

This methodology is great and should be used for other areas in x-phi. I've tried similar sorts of things but would love to hear people's thoughts about how to use your "trident" to probe intuitions about free will and responsibility.

[John Turri]

Thanks, Eddy!

Ori Friedman – who has been extremely generous to me throughout this project, and whose work with Christina Starmans on Gettier is fantastic – thinks that the main weakness of this set of studies is that I don't include enough controls, and that the controls I have in place should be better. I tend to think that they're good enough to get my proposal on the table as worthy of serious consideration.

Also, Ori and Christina observed a difference between Gettier cases involving 'apparent' versus 'authentic' evidence, which I tread over a bit in the version of the paper I have up on my site now. In earlier versions, I reported that I found no significant difference between tri-partitioned apparent and authentic cases (in both sorts, K-ascriptions were at floor). But that might just be because the tests I'm using aren't powerful enough to pick up on any remaining difference.

Correct, the results I've gotten provide no evidence that switching from 'only believes' to 'only thinks she knows' makes a difference.

I actually do have a hypothesis about what explains the earlier results. I've got a whole other series of experiments written up which identify a related effect. But these other experiments have nothing to do with knowledge or Gettier cases specifically. So it'll take another set of experiments, and another paper, to see if my hypothesis is on the right track. If it pans out, I'll post it when it's in good enough shape!

[Chandra Sripada]

Hi John,

This is a really excellent paper! I love the tripartite approach, and think the results are quite persuasive.

One thought. On page 20, it looks like you've got a pretty strong trend that people ascribe knowledge less in Husband than Burglar. The n's are very low, but if you increased your sample, this difference is sure to be highly significant. If this is right, then there may be a second factor (in addition to the tripartite structure) that explains your results compared to predecessors:

A noteworthy feature of most of Starmans and Friedman's cases is that they rely a lot on items being replaced by replicas, or pairs of improbable events that originate from a single causal source. You apparently notice this as well based on what you say in the first full paragraph on page 19. So it may be that avoiding the specific types of cases used by Starmans and Friedman also contributes to your eliciting the philosophically standard Gettier intuition. That is Husband, even presented without the tripartite structure, may have differed from Control (and also Burglar) and have been similar to Authentic Gettier.

Just a thought. Anyways, fabulous paper, and thanks for sharing!!

[John Turri]

Hello Chandra,

Thanks very much for the kind words!

I think you're right about that being a relevant factor. Wesley Buckwalter and Peter Blouw (a PhD student here at uWaterloo) made similar observations in hallway conversations.

I'm working on single-stage vignettes with a single causal source of luck, to see if they can elicit similar response patterns. I *think* I've cracked the code {fingers crossed}, but what makes it really tricky is pairing it with the right control(s). With Gettier cases, at least, I've found that there is a painful trade-off between a good control, on the one hand, and a simple and natural enough storyline that you can be confident won't confuse people, on the other.

[Christina Starmans]

Hi John,
First of all, I think this is a really interesting and innovative approach to testing Gettier intuitions, and I appreciate your effort to include the kind of control conditions that haven’t often been present in previous investigations.

However, I have some concerns. My main concern is that your controls aren’t as closely matched to your Gettier scenarios as they could be. Because of this, the results don’t tell us whether laypeople are sensitive to the Gettier factor in particular. There are many other psychologically important differences between your control stories and your Gettier stories which might be responsible for the difference in knowledge attributions that you report – for example, in the coin scenario there is a burglar in one condition but not the other; there are two coins in the Gettier story but only one coin in the control; and there is an extreme difference in the ‘noticeability’ of the coin at the end of the story.

To expand on just one of these, take the difference in ‘noticeability’. In the control condition the coin is easily visible at the end of the story, and in the Gettier condition the coin will never be seen again. This is a psychologically (but not philosophically) relevant factor that might influence laypeople’s knowledge attributions. In fact, Ori and I have some pilot data which we are currently following up on which suggests that noticeability indeed has this kind of effect, even in non-Gettier scenarios. So while this is an interesting fact about how laypeople attribute knowledge, it doesn’t tell us anything about how laypeople reason about Gettiering, specifically.

I don’t know whether these factors are actually driving the difference in knowledge attributions you report, but to really conclude anything about laypeople’s Gettier intuitions it’s important to create ‘minimal pairs’ – that is, 2 conditions that differ only in the variable we are interested in.

One final point: the ‘forged’ version of the coin story and the mule-zebra story are, of course, apparent evidence cases, and so your findings agree with those in experiment 4 of our paper; laypeople do not attribute knowledge to agents when their beliefs are based on apparent evidence – though as we show, this is not related to the Gettiering in the scenario, but simply to the nature of the evidence for the belief.

[John Turri]

Hi Christina,

Thanks very much for your thoughts! These are very useful points, and I'm looking forward to learning how those pilot data turn out.

When you say that the approach doesn't tell us anything "about how laypeople reason about Gettiering, specifically," could you please say what you mean by 'Gettiering specifically'?

I ask because it's not clear to me that anyone yet knows what is essential to Gettiering. Without that, it's hard to say what would be involved in creating a minimal matched pair.

Incidentally, if you look at the literature on Gettier cases, they're almost all apparent evidence cases. There is virtual unanimity on those. The ones involving authentic evidence, such as fake barn cases, only showed up later, and have been much more controversial, with many people saying that they're not even Gettier cases, and that the person knows it's a barn. (For example, I say that.) But that's not only an authentic evidence case, it's case involving present and ongoing perceptual contact with the object.

[Ori Friedman]

Hi John,

Perhaps no one knows exactly what is essential to Gettiering. But surely we know that many factors are not essential.

For example, it would be difficult to argue that adding a thief to the control condition is sufficient for turning it into a Gettier case. If that were true, then the victim would be Gettiered about all objects in the room, not just the coin. Likewise, it would be difficult to argue that Gettier cases require that one item be replaced with a different-looking item, and that stories with identical replacements don't count. If that were true then the numerous Gettier cases featuring twins would not count as Gettier cases.

So although there might be some fuzziness regarding the “ultimate” minimal pair comparison (i.e. the most minimal comparison possible), surely we can come up with non-Gettier controls that closely resemble their corresponding Gettier cases, and that don’t have lots of obviously unnecessary differences. In your own coin story, for example, a thief could have entered the room, and then moved the coin to some hidden spot in the room. I’m not suggesting this is the perfect minimal comparison possible. But it would certainly be more closely matched with the Gettier story, and would allow us to rule out the possibility that knowledge attributions were lowered because of the presence of the thief or because of differences in noticeability, etc.

Also, about most Gettier cases being apparent evidence cases, and authentic evidence cases being Johnny-come-lately's: Do you think that deep-down there are different kinds of Gettier cases? I mean, there is an obvious sense in which we can group Gettier cases into different types. But at the same time, my (amateurish) sense is that it was widely believed that a common factor unites all such cases (I’m not including fake barn type cases, though), and that the goal of much theorizing, including your own, was to account for this common factor. If so, the existence of authentic evidence cases would suggest that apparent evidence is not a factor that is essential to Gettiering.

[Christina Starmans]

Hi John,
I agree, actually, that people in fake barn cases know. But those cases are different than our authentic evidence Gettier cases, and your own. The evidence that we’re referring to is the evidence that the initial belief is based on, rather than the evidence at the end of the story when the knowledge question is being asked. The authenticity of the latter would create a fake barn type case, but the authenticity of the former is compatible with a Gettier case. We show that laypeople attribute knowledge when this initial piece of evidence is authentic (i.e. Katie is actually looking at a pen on the table when she forms the belief that there is a pen on the table), but not when it’s apparent (i.e. Zach is looking at a mule, but thinks it’s a zebra). It’s true that many Gettier cases that have been discussed in the literature are apparent evidence cases (i.e. “Stopped clock”, “Sheep in the field”, etc). But many scenarios long counted as Gettier cases, and for which philosophers have declined to attribute knowledge, are authentic evidence cases (e.g., “Ford”, Saunders & Champawat, 1964; “Burgled”, Sturgeon, 1993).

In terms of what I meant by ‘Gettiering, specifically’ – I simply meant that the difference in knowledge attributions you find between your control and authentic Gettier conditions might be due to something other than Gettiering. For example the addition of a thief, or the fact that the coin would never been seen again, might be what caused your participants to fail to attribute knowledge, rather than the double-luck formula of a Gettier case. So to create a minimal case, regardless of the precise definition of a Gettier case, all that is required is to keep everything constant between the two conditions except for the Gettiering – for example, as Ori & I did by using the same story in both conditions, but asking about a different object.

And by the way, I didn’t mean to suggest that the approach can’t tell us anything about how people reason about Gettier cases – actually I think it’s quite an interesting approach, and it would be interesting to see how the results would be affected by a more closely matched control.

[John Turri]

No fair – tag team! ;)

Seriously, you raise good points, Christina and Ori, and I'm taking them under advisement.

Judging from the discussion thus far, a more convincing experiment would have these qualities:

Q1. the evidence is authentic;
Q2. the replacement is the same sort of thing as the original, and just as easily noticeable
Q3. doesn't introduce an extra thief or other additional factor
Q4. there is a single source for both the bad luck and the good luck (this will help achieve 3)

I think that gives your view a very fair shot. My plan is to create a variation of your Katie case, with versions of it for five conditions (N=150).

C1. Base case: no luck – appears to be just a normal case of indirect knowledge of a mundane matter of fact (one stage)
C2. Base case + bad luck only – a weird dose of bad luck that results in a false justified belief (in one stage)
C3. Base case + good luck only – a normal case of indirect knowledge, along with a weird dose of good luck (in one stage)
C4. Base case + bad&good luck – an authentic Gettier case with a single source for both strokes of luck (in one stage)
C5. Base case + bad&good luck – an authentic Gettier case with a single source for both strokes of luck (in three stages)

I predict:

P1. Very high knowledge ascription in C1.
P2. Floor knowledge ascription in C2.
P3. High knowledge ascription in C3, and probably no different from in C1.
P4a. Ambivalence in C4, no different from chance, perhaps creeping up into the high 50s,
P4b. Knowledge ascription in C4 will differ significantly from C1.
P5a. Knowledge ascription at rates significantly less than chance in C5.
P5b. Knowledge ascription in C5 will not differ significantly from C2.

This commits me to a relative ordering, from high to low, for K-ascriptions:

P6. C1 --> C3 --> C4 --> C5 --> C2

If the folk view authentically justified true belief as knowledge, then we should definitely see knowledge ascription in C4 and C5 at rates higher than chance, and probably not too different from C1.