Friday, June 03, 2011

Hohwy on Phenomenal Variability

In a new article at Mind & Language (free penultimate draft here), Jakob Hohwy argues that I'm wrong in my claim (e.g., here and here) that the enormous variation in people's reports about their stream of experience reflects a pattern of massive error in introspective judgment. Rather, Hohwy suggests, the enormous variability of introspective reports reflects a pattern of enormous variability in the target experiences being reported on -- variability that introspection captures accurately. (Hohwy does allow, however, that there may be massive error in some domains and under some introspective conditions.)

Consider, for example, these differences: Some people say that their visual imagery is as vivid and detailed as ordinary vision, while others say that they have extremely sketchy visual imagery or none at all. And yet self-reported high- and low-imagers seem to perform similarly on cognitive tests that apparently draw on visual imagery, like mental rotation tasks (see here and Ch. 3 of here). Some people say that they experience 30 or more degrees of stable visual clarity with indistinctness only outside that range, while others say that they visually experience only a tiny point of clarity (1-2 degrees) bouncing rapidly around an indistinct background (see here). Could such people all be right in their radically different claims? Hohwy admits that the burden of proof is on him.

Hohwy supports his view in a variety of ways, but I found his cases for the radical variability of imagery and emotion the most interesting. Imagery, he points out, serves a wide variety of cognitive purposes. Some purposes may require detailed imagery (e.g., remembering a scene in order to paint it) while others require only sketchy imagery (e.g., matching two figures in a mental rotation). (These are my examples, not Hohwy's.) The request simply to "form an image" of, say, the front of one's house, as in Francis Galton's questionnaire, and David Marks's, and in my own, is a request without a natural, contextually-given purpose. People may form radically different images when given an introspective questionnaire, images which they accurately report, despite being generally cognitively similar on imagery tasks with more constrained or natural goals. This is an excellent point (which Russ Hurlburt has also emphasized to me), and I agree that I haven't adequately addressed it in my writings on the introspection of imagery. When asked generically to "form an image of your house" some people might paint themselves a detailed two-dimensional scene, some might sequentially drift around thinking about a few visual details of emotional significance, and others might do a minimalist 3D sketch -- even if they all have the same underlying imagery capacities and experiences when in more cognitively constrained contexts.

Let me note two things, though, that I think tilt somewhat back in favor of attributing error rather than radical variability. First, when instructed to form an image in an introspective questionnaire in a context where their image-making capacities are naturally seen as the issue under test (as in Galton's and Marks's questionnaires, I think), most people will probably think that they've been asked to form as vivid and detailed a visual image as possible in a short time frame. The majority of respondents will thus, I'm inclined to think, tend understand the request broadly similarly. If so, it remains unexplained why people should radically differ in their responses unless those differences reflect either radically different imagery capacities (which seems not to be the case judging from behavioral tests) or some sort of inadequacy in respondents' introspective reports. Second, people also make radically different general claims and capacity claims about their imagery (e.g., Berkeley's claim that there is never indeterminacy in the contents of his imagery, contra Locke). Prima facie (though I don't know that this has been well tested) such differences seem unlikely to be mirrored by equally radically different cognitive capacities. Such differences are, I believe, somewhat more straightforwardly explained as due to introspective error (possibly arising from folk theory or psychological theory, overanalogy to outward visual media, or investment in a certain self-conception) than as due to variability in the demands of different cognitive tasks.

[Well, it looks like I won't be able to do justice to Hohwy's even more interesting claims about emotional variability without going overlong. So let me save that for a follow-up post soon.]

6 comments:

Justin said...

Devils advocacy here:

You say subjects will try to form a maximally detailed image of their house. But is it obvious that they'd know which images are maximally detailed? If you just popped up on the street and asked me what types of mental images I have, and which kind were the most detailed, I'm not sure whether I could provide a useful answer. That's true even though when I'm imagining something, I'll say "this image is really detailed" or "I have a very hazy image of this."

Your second point concerning general claims about imagery is tougher to answer. If you're talking about people who haven't studied the subject, I think you can attribute it to a poor diet of examples: maybe people respond based on the first images they bring to mind and vary in those. I'm not sure what you'd say about professionals.

Eric Schwitzgebel said...

Thanks for the comment, Justin! You raise a good point in that first paragraph, though I'm inclined to think that it cuts both ways -- a defense of variability that depends on error or ignorance. And probably I put the point too strongly in writing "most people will probably think that they've been asked to form as vivid and detailed a visual image as possible in a short time frame". Different subjects might put substantially different degrees of effort into the task. It's an open empirical question whether the VVIQ prompt does engender radically different images in people with basically the same underlying imagery capacities.

On your second point: I'm inclined to think that professionals use a poor and skewed diet of examples, too, rather than making a balanced and systematic study of their imagery. Of course, this is just as compatible with Hohwy's view as with my own.

CHeno said...

Congrats on staying at UCR!!

Scott Bakker said...

Thanks for directing me to this paper, Eric. I can see now why this got you thinking about just what introspection is, because that seems to be the real issue that arises out of your debate with Hohwy. Ultimately, I'm not sure whether the unreliability versus variability tug of war bodes well for 'introspective science,' either way.

The one thing I would have liked to see Hohwy explicitly engage is the issue you raise regarding the 'cognition/perception' split, and the possibility that phenomenal variability and introspective unreliability are actually one and the same. What's the difference between 'getting the phenomenology wrong,' and 'being incapable of reliably *predicting* phenomenology'? In terms of operationalizing some notion of 'introspection' does it even matter?

Eric Schwitzgebel said...

Thanks for the comment, Scott! I'm enough of a phenomenal realist to think that there is a big difference between getting the phenomenology wrong and failing to predict it, or between variability plus accurate reporting vs. less variability and inaccurate reporting. I do somewhat mistrust my own realism about this, but I can't let go of it and feel like I'm being intellectually honest with myself. It would certainly be convenient if there were no such difference! On some readings of Dennett 1991, that's his core insight, really.

Scott Bakker said...

Dennett never did work his position through in any convincing manner I think (at least not that I've encountered). This prediction error minimization stuff, though, really does bite *Hohwy's* phenomenal realism in the ass (and he almost engages it in footnote 4). The PEM brain is a causal inference engine, one literally design to structurally overcome the inverse problem. This provides a very elegant way of reframing the problem I posed to you regarding 'causal neglect' and 'intentional cognition' vis a vis qualia and Frankish a few weeks back.

How can a machine designed to overcome its reliance on sensory EFFECTS possibly get a 'reliable metacognitive grip' on its own processes? It's simply not feasible, at least in any way evolutionary thrift wouldn't immediately rule out.

I've been thinking about this stuff for a while now (Dennett actually sent me Andy Clark's forthcoming BBS article on Action-Oriented Predictive Processing about a year ago) but I didn't realize that PEM approaches were making the inroads they seem to be. But I'll be posting on it soon...