Now, the problem perception may not seem very hard. There are trees, and people, and rocks, and planets, and monkeys in the world and you see them and you know what they are. It's not immediately obvious that this is such a difficult problem. There's a story and I'm not sure it's true, but there's a story of a very famous AI researcher, artificial intelligence researcher in the 60s, who set himself the project of building a robot. He worked on his most complicated part, but of course, the robot to make its way through the world had to be able to see and identify objects. So, the story goes that he assigned a graduate student this problem, as a summer project. I think do for next few months, figuring how we recognize things. Psychologists love this story, because it nicely expresses how simple the problem of perception looks, but how hard it actually is. So, if the story is true, the graduate student never really did finished the project. In fact, we've been working along the project of object recognition in computers and artificial intelligence for many years and only recently are we seeing some progress. So, I used to be able to say, when I gave this lecture about five, 10 years ago that, there's no machine that comes close to what humans can do regarding recognizing things. Machines are getting better and better and I have a feeling that one day I'll give this lecture and I can't say it anymore. One day it'll be machines from Facebook or Google that really can recognize objects at the ability of a person. But they aren't there yet. So, if you wanted these people who thinks that object perception is easy, you might have a conception of something like this, where what happens is, you see the world through your eyes and then you're sitting in your head and your skull looking at some sort of computer screen that indicates the world as it is. This is, for instance, the depiction in a Terminator movie, where you see the world through his eyes and what you see is the world annotated it with some symbols and some figures, but basically, the world. But of course, that can't solve the problem. That just pushes the question back, because now the world has been encapsulated into a screen, a computer screen, but how do you recognize things on the computer screen? You've solved nothing. So, the better way to think about perception is, well, you have the eye, which is a fleshy GUI sort of thing. Then, you have the retina and light hits the retina and the retina is a bunch of nerve cells and these nerve cells fire and they fire at some stimulus and not others. Then, so you get is, basically the equivalent of this, an array of numbers, numbers corresponding with the intensity of firing, where one number is fire, fire, fire and another fire, fire, fire, fire, fire. Now, you have an array of numbers, and then the problem of perception is, out of this array of numbers, this shifting array of numbers, you have to extract objects and people, where people are, what they're doing, what things are shape, what colors things are, what depth things are and so on. Now, you can sort of see why this is difficult. One way to imagine this is that, it's a certain mathematical problem, your array is a two-dimensional array that's all that's going on in your eye. You need to infer a three-dimensional world, which is what we got there, what you think about from a two-dimensional information and it's mathematically impossible to do this with certainty. It's like backwards multiplying. When given a compound number, you can't easily determined the unique numbers that gave rise to it, because there are multiple numbers consistent with that compound number. Similarly, for the world, imagine most of your retina is silent, nothing's firing, but there's neurons firing that corresponds to a shape and the shape is this irregular polygon you see here. So, what's in the world? Well, you don't know. For instance, you don't know how far away it is. The regular polygon to fill up this amount of the space in your brain, can be right next to you. It could be floating in space a foot away from you, or it could be a million miles away. So, think about the moon. The moon, you looking at a full moon in the sky, takes up some circular space on your eye. How do you know that it's not a foot away, or 10 feet away, or a million miles away? Nothing in information and eye per se tells you. You also don't know what shape this is. So, maybe it's the shape that you see it as irregular polygon, or maybe it's a square that's been tilted back. So, how do you tell? For everything you see, there's an indefinite number of possibilities as to the actual real-world things give rise to it. I'll give you real examples of this coming up and I want to do it in terms of color, actually brightness, in terms of objects and object recognition and in terms of depth. For each of these, I'll suggest that we have unconscious assumptions in the head about how the world works, that helps us take these indefinite number of interpretations and narrow it down to get to the right one.
