https://youtubetranscript.com/?v=M04xnKNChtI
Now, here’s something interesting, you can think about this for a minute. I went and saw an autistic woman speak at one point, her name was Temple Grandin. She’s really worth looking up, Temple Grandin is a very interesting person. She’s very seriously autistic when she was a child. But her mother and her worked her out of it so that she could be… She’s very functional, she works as a professor, I don’t remember where, it’s in the Midwest somewhere. Now, she’s famous not only for being a highly functional autistic person, who talks a fair bit about what it’s like to be autistic, but also for designing slaughterhouses across the United States. And the reason she can do that, as far as she’s concerned, is because she thinks like an animal thinks. And so she doesn’t… and she’s identified maybe at least part of what the core problem is with autism. So, the talk I heard her at was in Arizona, and it was a really entrancing talk. She showed some really interesting pictures of animals. So, what she’s done is she’s redesigned slaughterhouses so that when the animals enter the slaughterhouse, they go in like a spiral, basically. They can’t see what’s around the corner, and the walls are high so they’re not distracted by anything outside. So, one of the things she showed, for example, was a bunch of cows going through a standard sequence of gates, essentially. And off to the side, there was a windmill spinning, and the cows would stop, because the windmill… they didn’t understand what the windmill was, and they’d stop. Or she showed other pictures where cows were going down a pathway, too, and there was a Coke can sitting in the middle of the pathway, and the cows would all stop because they didn’t know what to do with it. Or she had another picture of cows out in the middle of the field, all surrounding a briefcase, and they were all looking at the briefcase. And the cows didn’t like anything that shouldn’t be there, and had a hard time mapping it. Now, she said… here’s a little exercise she did. She said, think of a church. Okay, so maybe you think… imagine a child’s drawing of a church. It’s like your standard house, like a pentagon, right? Which is basically how children draw the front of a house, with a steeple on top, and maybe a cross on top of it, or something like that. Which actually isn’t at church. It’s an icon of a church. And you think about how children draw houses, too. Pentagon, rectangle, what is it? Trapezoid. Chimney, almost always with smoke, which is quite interesting. I don’t know where kids get that exactly, but they almost always draw a chimney with smoke. Even though chimneys with smoke aren’t that common anymore. But anyways, you know, you can see what a child’s picture of a house looks like in your imagination. One of the things that you might want to think about is, that is not a picture of a house at all. Right? It’s an iconic representation that’s kind of like a hieroglyph. Because no house looks like that. And then you think about how a child will draw a person. Circle, stick, stick, stick, stick, stick. And you show it to somebody and they go, that’s a person. It’s like, really? It looks nothing like a person, right? I mean, you immediately recognize it as a person. But it looks nothing like a person. Well, what Grandin said was that when she thinks of a church, she has to think of a church she’s seen. She can’t take the set of all churches and abstract out an iconic representation. And use that to represent the set of all churches. She gets fixated on a specific exemplar. And she thinks that one of the problems with autistic people, and they have a very difficult time developing language, by the way, is that they can’t abstract out generalized representation across a set of entities. They can’t abstract. And then they, well, of course, if you can’t abstract, then it’s also very difficult to manipulate the abstractions. You see very strange behavior with autistic children, for example. So they don’t like people. And that’s because people don’t stay in their perceptual boxes. Like a human being is a very difficult thing to perceive because we’re always shifting around and moving and doing different things. Like we don’t stay in our categorical box. So autistic people have real trouble with other people. But they also have trouble. So, for example, if your autistic child gets accustomed to your kitchen, let’s say, and you move a chair, then especially if they’re severely autistic, they’ll have an absolute fit about it. Because you think kitchen with chair move. They think completely different place because they can’t abstract the constancies across the different situations and represent them abstractly. So I made this little diagram. I made this little diagram here to kind of give you a sense of what you might be doing when you’re abstracting perceptually. And so you could say, think about something that’s that complicated. It’s sort of my model of how complex the world is. But the world is a lot more complex than that. But the world is made out of everything is made out of littler things. And those littler things are made out of littler things and so forth. And those things are nested inside bigger things and so forth. And where you perceive on that level of abstraction is somewhat arbitrary. It has to be bounded by your by your goals. That’s the other thing is that your perceptual structures are determined by the goals that you have at hand. I mean, some of that’s that’s not completely true because your perceptual systems also have limitations, right? There’s things you can’t see or hear even if you need to. So there are limitations built in. But within that set of limitations, you’re still trying to tune your perceptions to your motivated goals. And that’s also very useful to think about when you’re trying to understand artificial intelligence, because for human beings without goals, there’s no perception because there’s no filtering mechanism that you can use to determine the level of resolution at which you perceive. Anyway, so there’s the there’s a thing made of smaller things which are made out of smaller things. And it’s so it’s kind of my iconic representation of the complexity of the world. And then you could think, well, what is this? How can you see this object? And I think if you just look at it, you can detect it’s like a Necker cube. You know, those cubes that that are line drawings that you can see the front of and then it’ll flip to the back. Have you seen those? So this is kind of Necker cube like or at least it is for me in that when I look at it, my perceptions play around with it. Sometimes I focus on the kind of cross like shape in the middle. And sometimes I can see these other lines. And then sometimes I’ll focus on that square. And sometimes I can see the little dots there, maybe one dot. And my perceptions are going like this, trying to fit a pattern to it. And you can kind of detect that when you’re watching it. And so I would say, well, you have the options of perceiving this in its full complexity or you can simplify it. Essentially, there’s lots of ways you can simplify it, but some of them are laid out there. So you take the complex thing, you make a low resolution representation of it. So that’s it’s rough. That’s the rough area that all those dots occupy. That’s the rough area broken down to its four most fundamental quadrants. That might be how you would look at it. If this was a map of an orchard and you were trying to walk from south to north, that would be a useful representation. This combines this and this. That’s the highest level of resolution that you can perceive this object at. That’s lower resolution than the object itself. So the first issue is how should you look at things? Well, that’s a problem that intelligence has to solve. So that’s one of the problems that intelligence goes after. And then I think what happens is we have the thing in itself and then we simplify it with a perception. And that’s like an iconic representation. And then we nail the iconic representation with a word. And that’s how we compress the world’s complexity into something that we can manage. We take the complex thing, make it into an icon and represent the icon with a word. And then when I throw you the word, so to speak, you decompose it into the icon and then decompose it even further into the thing. If you know the icon and you know the thing. And so then we can use shorthand, right? Because you have representational structures and so do I. And I’m just tossing you markers about your representational structures and you can unfold them. That’s what you do when you’re reading a novel. Because the novel comes alive in your imagination in your own idiosyncratic way. And it wouldn’t if you didn’t understand the references of the novel, right? The novelist has to assume that your basic perceptual structures and your intuitions and your instincts are basically the same as his or hers. Because otherwise they have to assume that. Because otherwise they would be lost in an infinite regress of explanation. So it’s problematic often. For example, if you start reading Victorian novels, you may find that it takes a while to get into them. Because the presuppositions, the expectations are slightly different and so is the language. You have to update the representations. But anyway, so that’s roughly, as far as I’m concerned, that’s roughly a representation of what intelligence is doing in the world. Or a big part of it. It’s how in the world do you look at things so that you can use them for the purposes that you need to use them for. And then the next problem that intelligence has to solve, which is related, is once you’ve got the perceptual landscape sorted out, how do you abstractly represent the action patterns that you’re going to implement in the world? So it’s how do you perceive where you are now? How do you perceive where you’re going? And how do you construct up and then implement strategies that enable you to move from where you are to where you’re going? So it’s a continual process of mapping and movement. So it’s navigation. That’s what we’re doing in the world all the time. We’re not just navigating through it because we’re mobile creatures. We’re navigating through it, attempting to make the world manifest itself in accordance with our wishes. And that’s the fundamental problem that intelligence has to solve. And animals have their perceptions to rely on. But we have our perceptions and our ability to abstract from those perceptions multiple times and then to abstract finally into language. So we live in a very abstracted world. And it also means that we can learn a lesson in one place and generalize it across many other places, which is also something that animals have a hard time doing because they don’t know how to do that initial perceptual generalization.