In this chapter, Pinker proposes a number of assumptions that the mind works under to interpret the visual world. Considering these assumptions are exploited in most optical illusions, I thought it might work well to use illusions to demonstrate them. (Fortunately for our eyes, none of these are autostereograms.)
Assumption 1: "The world often contains parallel, symmetrical, regular, right-angled figures lying on the flat ground" (217); these figures are affected by perspective. The brain, says Pinker, employs probability to guess the nature of objects in space (244), and finds it probable in most circumstances to assume that such objects tend toward the parallel/symmetrical/regular persuasion. If two lines seem to taper toward one another at an angle that suggests they are parallel but moving away, for example, the brain finds it plausable to assume they are parallel.
Pinker provides the example of Ames's room, seen below:
From this website.
The effect of the room, explained on pages 215-16, is reinforced by the multiple lines present in the photo: not only do the floor and ceiling seem to align in parallel, but so do the windows and the checkerboard squares on the floor.
Another illusion operates similarly, appearing regular and right-angled . . .
From this website.
. . . until viewed at an angle.
---
Assumption 2: "Surfaces are evenly colored and textured [�], so a gradual change in the markings on a surface is caused by light and perspective." (217) Light itself, for that matter, can be attributed to two further assumptions: shading and reflection (245-248). When an object's coloration varies sharply in value, the mind might attribute that variance to either shadow (like a fold in the object) or an external light source (like a lamp); in both cases the picture is 'equalized' to compensate for this factor, and light and shadow are read as being themselves, instead of variations in the object's pigment. When we see a crumpled up piece of paper, its shadows don't keep us from assessing it as all white.
The effect is so strong that optical illusions like the one below are very convincing. Squares A and B are the same color, but the equalizing effect makes one seem 'gray' and the other 'white':
From this website.
As for illusions that more generally exploit the assumption re: color/texture consistency, well, those are just paintings. A flat object, the canvas, can be made to seem three-dimensional if the color on it is gradated in a way that suggests depth; the mind presumes that the gradations indicate several distinct surfaces. The effect is especially striking when artists go the full mile and take away the frame.
From this website.
From this website.
-
Assumption 3: "Objects have regular, compact sillhouettes." (217) If an object appears circular with the exception of a gap in the shape of a right angle jutting into it, an angle which exactly shares its boundry with a square object, chances are you are not looking at a scene in which a square shape is resting next to or behind a circle-with-a-square-shaped-hole-in-it; chances are you are looking at a square in front of a circle.
Ames does a good job of exploiting this.
From this website.
So does this painting, which works from the opposite principle: the shape seems unified, so we assume it can't be the result of intersecting sillhouettes.
From this website.
-
The fallacies suggested by optical illusions say a lot about how vision works in our heads. On page 254, Pinker suggests some of how that information might be applied to artificial intelligence; if our visual assumptions which are exploited by the illusions can be written as logical precepts, then computers can be programmed to follow them. (Of course, the assumptions that go into interpreting a two-dimensional vector a la the folded-paper-with-a-strip-on-it graphic are far simpler than those that go into percieving the real world; if there is a concrete, finite set of assumptions our mind is making, they're a long way from being discovered.) If computers make the same mistakes we do in percieving an Ames Room or the caf� wall illusion, then presumably they are operating under the same visual system humans are. This feels like a promising road toward developing some nifty AI vision tests. (Maybe even vision-variety Turing Tests. "What do you see going on in this picture?") At the same time, I do wonder: if we want to program computers to see, is it necessarily desireable to emulate the human visual system so exactly as to emulate its mistakes as well? If we could develop a computer that could see through optical illusions, would that be better? (More daringly: would that be vision?) The answer may well be Chomsky's, that here is a case where we really are pursuing computer research in order to better understand ourselves.
. . . But now that the imagined fusion of computers and vision has been proposed, there's no way to get around asking the other question�one which I think has come up in some of our earlier reasons�namely, how much vision shapes our own thoughts, and to what degree a computer can think if it can't see. There are all sorts of notions, like "cuteness," which are derived from a visual context; how would a computer make sense of them? How much do they affect our own thinking, actually? . . . I actually just now thought of the natural answer to this question, namely blind people, who are obviously exercising thought despite being unable to see. Although I do wonder how a blind person might perform on a Turing-esque test. (Meanwhile, I guess the computer question should be more safely revised to include the other senses: to what degree can a computer think if it can't see/hear/taste/smell/touch (i.e., interact with its environment)? And this one has been asked before.)
Last note: however big a part sight plays in cognition and thus in AI, Pinker does seem to be implying via terminology that vision itself might be--or maybe ought to be--the working of a sort of computer. Recognition modules employ geometry (273); mental pictures are accompanied by instructions for interpretation (297); "the concreteness of mental images allows them to be co-opted as an analogue computer" (291) (!), even. I get the idea that the mind-as-computer theory is one of Pinker's more controversial claims; Jerry Fodor, for one, has issue with it. (The "massive modularity" section: "If, in short, there is a community of computers living in my head, there had also better be somebody who is in charge; and, by God, it had better be me.") But if the computer-esque vocabulary is in fact being credibly applied, well, that's certainly exciting. I'm not sure what I believe about the whole thing--aside from a general sense that mind can operate without vision better than the other way around--due to my overwhelming lack of knowledge in all things cognitive science-ish, but computing vision does seem like it would be a significant success in the search for thinking machines.
-
Actual last note: I'm not sure which visual mechanism this is the result of, but behold: rotating snakes!
