Tuesday, 27 September 2022

Are Illusions Even a Thing?

Traditional vision science is very excited about illusions. These are cases when perception seems to break down; there is a mismatch between what is out there and what we experience, and traditional approaches consider these breakdowns as clues to how vision has to work, given what it is working with. 

Ecological psychologists don’t like illusions. Typically, they occur when information is either made ambiguous or faked, and in general we think these are the wrong situations to study perception in. We sometimes engage with the literature on these effects, but usually to show how the trick is the result of not thinking ecologically. 

Rogers (2022) has taken this basic analysis but gone one interesting step further. He’s argued that the notion of ‘visual illusion’ is simply not a clear category; it’s not a useful way to describe any of the effects people study. He argues that there simply is no sufficient definition of what an illusion is that works, and that what we call illusions are just either tricks (as above) or inevitable consequences of how the visual system works. 

I am broadly on board with this additional step, and it’s made me think hard about what illusions are and how best to respond when people use them against direct perception. 

Rogers first argues that no-one actually has a good account of what an illusion even is. The standard definition is Gregory’s: an illusion occurs when there is a mismatch between what is really out there, and what we experience. An illusion is an error. (Rogers also spends some time looking at a modern analysis by Todorovic (2020) which has, according to Rogers, basically the same idea at the core and therefore the same limitations.) 

The problem here is that there are many ways to describe ‘what is really out there’, and a long-standing ecological objection to illusions is that the mismatch is an artefact of the researcher having the wrong description. Gibson spent the first 4 chapters of the 1979 book describing an ecological reality for a reason! So the mismatch is often just an analysis error, not a perceptual one. A corollary to this is that illusion results are often telling us about our understanding of the relevant physics, and not about perception. 

Rogers then describes three ways to classify candidate illusion effects that he suggests covers the space without ever needing the notion of error. I think an interesting step going forward will be to try and apply these three categories to all and any illusion discussions from here on out, to see if it holds up. 

The first are effects that would never fit any definition of illusion. He discusses the Ames Room, where a non-rectangular room is made to appear as if it were rectangular when viewed from a particular place. Rogers identifies that it’s simply unfair to consider this an error by the visual system, because it is being meticulously fooled – it is being presented with what Rogers calls a facsimile, something designed to look a certain way. It’s not an error for the visual system to perceive a good facsimile as the other thing; how could it possibly do anything else? If it’s not an error, then it cannot sensibly be called an illusion. (I blogged Runeson's excellent analysis of the Ames Room, which Rogers is endorsing here). 

The second are effects due to how the system works. His first simple example is thresholds. If there ‘really is’ a dim light present but the visual system doesn’t register it because it is below threshold, it’s not an error; it’s simply a result of the way the system works. In the same way, (my example) it’s not an error when a microphone fails to respond to light but does to sound, because that’s how microphones work. Colour metamers is another example for Rogers; we don’t consider these errors but simply what happens when a trichromatic system is operating normally. These are simply how the visual system measures the world, and calling these effects errors/illusions is just to not understand this fact.  

The third are effects of using impoverished situations. Rogers point is simply that it has to be true that the system will fail in some way when you remove information the system requires to work properly. Calling this an error is unfair. Rogers doesn't quite spell this out clearly, but I think this is a combination of the above two situations: the impoverished setting can't actually be seen any other way, because of how the perceptual system works. 

There is an additional concern here, though, that Rogers focuses on; not only is it unfair to consider these errors/illusions, they aren't good experiments. The logic of these studies is usually that systematically removing potential sources of information (typically framed as cues) tests whether those were, in fact, important. But this subtractive approach to perception assumes that the system is linear: that the full system and the system missing one cue only differ in that one cue. But this isn't the case; Rogers reviews some examples here, but my favourite is always Mon-Williams & Bingham (2008). They tested distance perception under full cue and restricted cue conditions, and basically showed that a) people will use the non-specifying height-in-the-visual-field when that's all there is, but that b) they show no signs of using it when better information is present. Studying perception in a restricted-cue task tells you how perception works in that task, but not how it works in the typical full-cue setting. 

Conclusions

I am basically on board with this analysis, because it is, basically, the ecological analysis of illusions (Rogers cites Runeson (1988), one of my all time favourites, and some Gibson, but I did find the framing of the paper to be a bit Rogers-centric, as if he had come up with all this. I'm not mad, just a little disappointed :) The novel part is the extra step, saying not only are illusions not methodologically useful or the result of an uninteresting trick, but that given all this the category of illusion-as-error is probably not a real thing. That's quite a bold move and I like it. 

The paper only discusses visual illusions, and so on Twitter Gavin Buckingham raised his standard objection of 'what about non-visual illusions like the size-weight illusion, etc?'. I'm inclined to think there's no reason Roger's analysis won't apply here (he uses examples but only to illustrate the deeper points and those will stand, I think). Worth working through, however! As I note above, if this paper is to have the impact it should have, any discussion of illusions should get filtered through the three classifications above, to see how well they hold up. 

References




1 comment:

  1. I know very little about this subject, so I'm aware that my arguments may be flimsy, but I'm not really convinced by the "illusions aren't errors" argument

    In particular, I'm not sure what you mean when you say that a light too dim to be registered is not an error, "simply a result of the way the system works". I agree that this case doesn't fall under the common conception of "error", but can't "errors" (by definition) always be attributed to the way a system works?

    I'm also kind of uncomfortable with the use of the word "unfair". I mean, I agree that if you intentionally deceive someone, then punish them for the error you provoked, you've been pretty unfair (unless the error can be improved with punishment I guess).
    But on the other hand, it seems to me that any punishable act can be considered as some kind of error. And it feels like considering something as an error would make it more understandable and less blameworthy, not the reverse; and certainly, in a psychological context, "errors" are rarely considered blameworthy.

    Of course, calling something an "error" doesn't tell us how it occurred, and so if we want to understand how something works, we sure can't stop at "it's an error!"


    Sorry if this is unclear... I guess my main question is, do you think the idea of "error" makes sense as a category?

    If it doesn't, then why not criticize that idea directly?

    And if it does make some sense, then maybe illusions could be defined as errors occuring in tasks involving description, for example?

    I guess then we would have to define "description" and so on, and probably it still wouldn't be a perfect category, but then again what is?

    ReplyDelete