Empirical Tests of Interest-Relative Invariantism

According to advocates of Interest-Relative Invariantism (or pragmatic encroachment), whether or not someone knows something at a time (or has evidence for it) depends in part on what is at stake for that person at that time in acting on it. There has been a good deal of controversy recently about how this thesis reflects the intuitive judgments of vox populi, with a number of papers arguing that the folk do not have interest-relative intuitions. Summarizing this work, Jonathan Schaffer and Joshua Knobe say, in a recent paper in Nous, that IRI models “seem to predict the exact opposite of the data.” In this paper, forthcoming in the journal Episteme, Chandra Sripada and I critically discuss some of the prior studies, and present the results of own experiments, which provide strong support for IRI. We discuss our results in the light of complementary results by Angel Pinillos, and conclude by discussing an intellectualist challenge to IRI due to Jennifer Nagel.

This figure summarizes our results. One key discovery we made in running our surveys is the difficulty in presenting non-philosophers with a true low-stakes case. We repeatedly found that non-philosophers tended to interpret the basic low-stakes case as a high-stakes case. As we argue in section 1.3, this is to be expected on general Gricean grounds.


Empirical Tests of Interest-Relative Invariantism — 9 Comments

  1. This is a very nice paper — definitely an important contribution to the discussion in this area.

    I just wanted to ask a quick question about the interpretation you offer for your findings. You find that stakes have no significant effect on attributions of knowledge in what you call the ‘basic case.’ You then say that the reason for this null result is that people naturally assume that an epistemic subject in a high-stakes case will take care to gather more evidence, consider that evidence more carefully, etc. So there might be one force making people less inclined to attribute knowledge in the high-stakes case, but then there is another force making people more willing to attribute knowledge in that case. The end result is that we get no effect at all.

    I think this hypothesis sounds very reasonable, but I notice that you find in your own results that participants say that the epistemic subject actually has *weaker* evidence in the high-stakes case than in the low-stakes case. It’s just that participants aren’t any less inclined to think that the epistemic subject has knowledge in the high-stakes case.

    Do you think you could say just a little bit more about how you are understanding these findings?

  2. Josh – the first section goes through some potential confounds. The one you discuss is emphasized heavily in Nagel’s work, and seems reasonable enough. As we say later in the paper, though our results provide good evidence for some of these confounds (like the problem of narrator cues – in the basic case, subjects are treating low stakes lime Hugh stakes), we do not find evidence for this particular suppressor effect. On the other hand, nor do our findings rule against it – the effect may still dampen the stakes-sensitive effects, without (as we show) not eliminating them. Personally I suspect the main culprit has been the problem of narrator cues.

  3. In short you have misinterpreted us a little bit. We don’t argue that the basic case is explained by the suppressor effect you mention. Rather the data suggests that what has happened is that subjects are interpreting low stakes cases as high stakes cases. Suppose there were no stakes effects. What you would expect to see is people being fairly liberal about the standards for knowledge and evidence in both low and high stakes cases. That’s not what we found. What we found in the basic case was that subjects were uniformly demanding in their standards for knowledge and evidence (uniform, that is, across low and high stakes). The only way to elicit the more liberal standards for knowledge and evidence was to work much harder than anyone had previously done to dampen the effects of relevance and other Gricean maxims that would lead subjects to suspect it was really a high stakes situation after all.

  4. Oops I read your comment too quickly. You are worried about the asymmetry between knowledge and evidence in the basic case. Good point. We don’t know how to explain that. If it were more robust one may want to use as a basis for IRI about evidence but not knowledge, or as an argument against E=K. However it just doesn’t seem robust enough a difference to ground dramatic conclusions, IMO.

  5. Hello!
    Let us suppose we’re trying to evaluate S’s belief that infinite sets has infinite sub-sets. Let us also suppose S satisfies the classical + anti-Gettier conditions for knowledge in this case. Now, what would an IRI defender say we need to take into account as to what is at stake for S in acting on that belief, in order to attribute knowledge to S or refuse to do so?
    Looks like S can know that infinite sets has infinite sub-sets in spite of what is at stake for S in acting on it. What is the criterion to define ‘what is at stake in acting on a belief’?
    Thank you!

  6. Hi Luis,

    It depends on the defender of IRI. But putting some of the differences aside, there are lots of actions I can take right now for which “infinite sets have infinite subsets” could potentially serve as reasons. For example, I could decide to defend the view in print. Or I could decide to stop inquiring whether it’s so. Or I could decide to insist to someone that it’s so. Some of these actions are insufficiently connected to the proposition that infinite sets have infinite subsets, so that even if infinite sets have infinite subsets, I don’t end up justified in performing those actions. So, that infinite sets have infinite subsets might be a reason I have to defend it in print. But that I have other things to do, that it’s already been adequately defended and I have nothing new to say, that I’m not very good at articulating my reasons — these might all be defeating reasons, and so I might not be justified in defending it in print. But, according to the IRIist, as long as it’s proper to use “Infinite sets have infinite subsets” as a reason for the relevant actions, that’s what it’s going to take to satisfy the relevant additional conditions.

  7. Hi Chandra and Jason,
    Excellent paper! So glad to be able to see the studies. A few initial reactions, for what they’re worth:
    SECTION 1.1:
    You raise a worry about most of the previous experiments (referencing Buckwalter and Feltz & Zarpentine) involving assessing knowledge claims in embedded contexts. While you note that this applies only to some of the set of previous studies (they only “frequently” used this method), you don’t make very explicit that it isn’t done in others, especially ours (May, Sinnott-Armstrong, Hull, & Zimmerman 2010). We used the same operationalization of the knowledge attribution variable that you do later in the paper: we asked for degree of agreement with “Hannah knows that _____.”
    Of course, some worry in the neighborhood might apply to our studies. While we don’t directly ask subjects about what Hannah says, she does make a self-ascription of knowledge. As Pinillos has pointed out, something like the rule of accommodation might pose a problem here: participants may have tended to attribute knowledge just to be charitable, taking Hannah at her word, etc. While you don’t want to defend contextualism (which appealing to the rule of accommodation might seem to do) the main worry here seems to apply to any vignette that has the protagonist self-ascribing knowledge. And it looks like this applies to our vignettes but not yours. I’m not sure how much weight to put on this, but it’s definitely something to consider (especially, if I recall, in light of some further work Buckwalter has been doing).
    SECTION 1.3:
    You suggest that previous data (purporting to reveal no stakes-sensitive intuitions) can be explained partly by the idea that subjects were not presented low- and high-stakes scenarios together. This, you suggest, could prevent participants from interpreting the Low Stakes case as one with in fact low stakes. From what I recall, that is absolutely true of every study except ours (May et al. 2010). In Section 4 of our paper, we ran a within-subjects experiment in which each participant received both a low- and a high-stakes vignette. We still found no substantial effect of stakes. (As was slightly different with our paper from Buckwalter and others, there was again an effect, but it did not push most participants from attributing knowledge to denying.)
    This doesn’t account for much, but it indicates that this objection to (and explanation of) our data won’t work. From what I can gather, though, you would then try to just use another of your objections/explanations. The only one left that applies is the problem of focusing on mere true belief (section 1.2). If our participants did focus too much on whether Hannah truly believes that p, rather than whether she knows, then we’ll expect the data we got (agreement in both conditions). Again I’m not sure how much weight to put on this. But, as I mentioned before, we also might have the problem of accommodation/charity that you, DeRose, and Pinillos raise (from section 1.1).
    SECTION 2:
    I’m puzzled by why you made a certain change to your final pair of vignettes (the Ignorant cases). Instead of having Sarah merely say that the noodles “may have” the nuts (presumably just expressing the mere possibility), you have her say she’s “heard” that the dishes are “often served” with the nuts. I’m not sure why you even included the mere possibility, but doesn’t Sarah’s new testimony introduce additional evidence against the target proposition? I had the same sort of worry about the new bank cases that Schaffer and Knobe tested in their paper. But at least with your vignettes the new evidence occurs in both of your cases (in theirs, it only applies to one, which could explain their difference in responses between their cases). Still, this might affect what we should conclude from the data, since we now can’t directly compare the last pair of vignettes with the other two pairs, thinking that only ignorance is varied.
    One worry, for example, is that if the evidence against p is higher in both the Ignorant cases, then we should probably expect more agreement if the evidence against was removed. That is, we should expect additional evidence against p to lower agreement with the knowledge attribution. In that case, the better-formulated Ignorance cases may have yielded data more like ours: some effect for stakes, but participants still tending to attribute knowledge. In that case—comparing it with the Implicit/Explicit vignettes—maybe we shouldn’t conclude that objective normative reasons play the role in IRI (section 6). There may be other implications as well.
    Anyway, just some thoughts. Now I want some noodles with pine nuts…

  8. Hi Josh (May),

    Thanks for these really thoughtful comments and criticisms! Here are some responses:

    Section 1.1: Your observations here are spot on. We agree with you that in addition to the embedding issue, accommodation may be playing a role and these issues warrant further study.

    Section 1.3: We argued that when evaluating paired hypothetical cases, there are important differences in the psychological-cum-epistemic situations of philosophers versus naïve experimental subjects. The two differences we focused on were that philosophers 1) are aware of the hypothesis the cases are intended to test; and 2) get to see and understand both cases before they render their intuitive judgments. We write, “Having access to both cases greatly facilitates one’s ability to mentally move back and forth between the cases and ‘home in on’ the intuitive judgments that are supposed to differ across the two cases.”

    You are right that in May et al 2010, you presented each subject with *both* High and Low Stakes versions of the case. But this still does not address (1), and my sense is that it does not do enough to address (2). In your study, experimental subjects were given access to both cases, but they read and answered questions regarding the two cases *sequentially* (which is presumably why you found order effects). Philosophers do something different. They read both cases first, then they deliberately move back and forth between the two cases multiple times to make the differences more salient, and finally they render their intuitive judgments. So I still maintain that some key psychological-cum-epistemic differences between philosophers and experimental subjects remain.

    Section 2

    Your point is well taken that we did include an element of testimony in both versions of case 3 (the ‘Ignorant’ case) that was not present in cases 1 and 2 (the Basic and Implicit/Explicit cases). We did this because we figured that Mongolian pine nuts are a fairly exotic nut compared to regular pine nuts. Thus Hannah’s abilities as a Mongolian pine nut detector would be perceived to be quite a bit worse than her abilities as a regular pine nut detector. This difference pushes against knowledge attributions in case 3 versus 1 and 2 (indeed, my recollection is that we found as much in early pilot studies, but I don’t have the actual results handy right now). To push back a bit in the other direction, we added an element to case 3 (i.e., the testimony) that favors knowledge attributions. I think you are absolutely right to point out, however, that the symmetry between case 3, on the one hand, and cases 1 and 2 on the other, is not as tight as it could be.

    I should point out though that the discussion of objective normative reasons in section 6 relies primarily on our finding a difference b/w Ignorant Low Stakes versus Ignorant High Stakes. The fact that stakes continue to exert an effect on knowledge and other epistemic properties, even though Hannah is (faultlessly) completely unaware of these stakes suggests that, at least in some cases, it is the ‘practical reasons that there are’ (objective normative reasons for action), rather than the agent’s subjective beliefs about these reasons, that make the difference. The preceding argument does not depend much, in my view, on the fact that the difference between Ignorant Low Stakes and Ignorant High Stakes happens to straddle the midpoint, or other factors that the testimony issue might be relevant to.

    So in sum, whether adding testimony to case 3 made this case somehow different from cases 1 and 2 is an important question. But I do not see this issue as potentially taking much away from our interpretation of case 3 taken by itself, and in particular, our finding that, at least in some cases, it is objective rather than subjective normative reasons for action that affect epistemic properties and relations.

    Thanks again for the careful read Josh – as always your comments are really helpful.

  9. Pingback: Chandra Sripada and Jason Stanley on Intuitions about Knowledge | Ellen Wolchek

Leave a Reply

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