Minsky

Thank you. It’s very nice to be here.

Minsky

Well, I think that’s the idea, all right. Many people say, “I don’t see how a machine could think, or a machine could be conscious, or a machine has feelings,” and I understand why people are so skeptical of this, because usually when you talk about a machine you’re talking about a typewriter, an automobile engine, or something, and it has none of those properties. We don’t have any experience in our lives except with people—who are the most complicated machine in the world—and so it’s natural to be skeptical. But what I suggest in this book, and the kind of theory I work on, is that the mind is many hundreds of different kinds of computers. It took us four hundred million years to evolve from the time that we were fish to humans, and what happened in that time isn’t that the brain just got bigger, but it got more complicated, and the book suggests—it’s like psychology fiction, you see, because nobody really knows how the brain works. And so I paint a picture of how you could have a lot of stupid things—because computers are stupid—that could get together and do the kinds of things that we admire in children and people, and the sort of things that we’re used to in a person.

Minsky

Well, I think that’s a funny way to use words, and that’s absolutely right. I hear this all the time. But then, if you ask people, “Well, what do you mean by consciousness?” they say mysterious things. I think we can explain consciousness the way science explains other things. You work for a while, and you try to say what is it that we’re really talking about—what are the phenomena, what happens in consciousness that I have to explain? And if you talk to most people they have a very fuzzy idea of consciousness. They say, “It’s being aware of everything that’s going on.” Well, we’re not. “It’s knowing what your mind is doing.” Well, we don’t. When I talk, I haven’t the slightest idea of the processes that produce the words. So how I make the words is not conscious, and when you talk to me, and these sounds come in and I make sense of them, yes, I’m conscious of the words in a sense, but I’m not at all conscious of the tremendously complicated processes. We ought to have more respect for ourselves, and the joke—I think—is that when a person says, “I’m not a machine,” they’re showing a lack of respect for people.

Minsky

Well, because we’re the greatest machine in the world. The brain has a hundred billion cells. They’re organized in a way that took—I might sound like Carl Sagan with these millions, but there’s no way to avoid it. If you look at the evolution of animals, five million years or so ago there were no people. There were chimps—or rather, there was the common ancestor of the chimp. And that’s a great animal, and it does lots of things. It can even learn words. It can’t make complicated sentences. And before the chimpanzees we have all the other kinds of mammals that you and I know: dogs and cats. And something wonderful happened in the last few million years—and that’s a very short time, because it’s one percent of the time since we were fish. Now, when I say that to say we’re not a great machine is to not respect ourselves, I think—you see, when a person says, “I’m conscious, I’m just something different; it’s different from anything else in the world,” that’s sort of pleasant and boastful, but it doesn’t give you anything to be proud of. It just says, “There’s a little gleaming jewel here, and all my virtue comes from it, and nothing I did earned it.”

Minsky

But if you think about it twice, you see, it’s rather just the opposite of that. If we’re descended from the gods, well, that’s nice, but it’s no reflection on us. There’s no virtue to you to have been created all in one piece. So I think that the so-called humanists who feel that dehumanizing the person is bad are shielding themselves from the glory of having—we’re like Prometheus, not like God. We’ve struggled for this four hundred million years through the slime, and in every era, when this animal learned to climb a tree, and learned to take care of the children when it’s cold, and all the different things, then natural selection added another little part to the brain. And I’d guess there’s something like four hundred kinds of computers, and each of them has developed specially. The mystery of consciousness, to me, is not, “Isn’t it wonderful that we’re conscious?” but it’s the opposite: “Isn’t it wonderful that we can do things like talk and walk and understand, without having the slightest idea of how it works?” And so that’s the mystery—not that there’s some magic that brings everything together, but that the mind is really incoherent in a sense. We don’t know how we work, and yet the thing works. It’s like a big corporation with no one in charge, and I think it’s wonderful.

Minsky

Well, I don’t think there will be a spark. I think we’ll have a more miserable experience than that. But I think the picture you painted is right. I don’t know if the general public knows how many wonderful things have happened since around 1950. People have made computers or other machines do a little bit of vision. We have machines that can read letters. We have machines that can recognize the sounds of some words. Right now there’s no machine that can look around and tell a dog from a cat. But there’s a reason why there’s no such machine—because when you look at something like a hand, you’re not seeing a hand. You’re getting a lot of information, and your visual system is finding the dangers and finding the parts, but another part of your brain has this knowledge that you’ve built up since childhood.

If you see a telephone, something remarkable happens, because when you see the telephone, it’s not just that the word ‘telephone’ occurs to you. You know what it means. It’s more than seeing. You know that you talk into this part and you listen here, and that it’s probably going to ring sometime, and then there’s a lot of behavior. What I think is going on in the brain is that yes, there’s a lot of machinery for division and there’s a lot of machinery for remembering experiences, and there are machines for interpreting the sounds of words into processes that we don’t understand yet. I think the 1990s is going to be a wonderful adventure, because we’ve had forty years of preparation, of working on little parts of artificial intelligence, we call it—machines that can see a little bit, that can do geometry, that can do some mathematics, do a little bit of language. But it’s chaos, in the sense that for forty years all sorts of wonderful people have developed little bits. The idea in my book is to say, is there a way to make one computer program, or one machine, do all those things? I don’t think so. If you had one machine that knew how to walk and how to see and all those things, I think it would get confused, and so I think what has evolved in the brain—and the brain scientists have confirmed this all the time—is that if you injure a certain part of the brain, you can’t recognize faces, and yet you can recognize other things. If you injure another part of the brain, you can’t remember how to work your hand. I think each of these parts of the brain has its own knowledge, and the reason it’s possible for us to function is that we’re like a social organization. One part of the brain, if I wanted a drink of water, I would go over there to get it. But the part of my brain that wants the drink of water doesn’t know anything about walking, but it can exploit the other one. It can sort of request to the walking machine, “Say, you know how to do that. I don’t have the slightest idea about how to move muscles.” And so in your head is something magnificent. It’s like a whole city. Maybe I just made up the number four hundred because I like the idea that maybe one of these developed every one million years, very slow. And so when I say that a person is a machine, I think that’s much better than saying there’s a little spark from which all of your virtue comes, because then you don’t deserve it; it’s just a spark.

Minsky

That’s right. If we were just a simple thing—a simple machine with a lot of memory, or a simple soul with a lot of memory—there’d still be a terrible management problem. I think what’s happened is that by having the brain divided into separate segments, each of which gets very good at something—I think every part of the brain is good at two things. First it’s good at doing its job, and second it’s good at learning which other parts of the brain it’s connected to are good for helping with different problems. And so you’ve got a smoothly functioning organization in most people. And we forget there are a lot of people for whom this organization doesn’t work; there are a lot of people in institutions, there are a lot of people who get caught into kinds of behavior that really destroy themselves. So nobody’s perfect, we might say. But it’s amazing, I think, that anything so wonderful and complicated could work.

Minsky

Yes, and the same for the animals too—that they’re different from anything else. Let me give an example of that. When we’re talking about something, and somebody comes in, you’d say hello and then I’d have a good chance of being able to get back to you and remember where I was. So one of the nice things that a person can do is tolerate some interruption. Now, in fact since about 1965 or so, computers have had a certain ability to tolerate interruption—run one job and stop it for a moment while it prints something, and come back. For computers to have this ability to tolerate an interruption, we had to do something. And I know the people who invented these priority-interrupt systems—

Minsky

Right.

Minsky

That’s right. In order to make a time-shared computer, which you didn’t have in the 1950s, the first few years, we had to invent things called short-term interrupt memories, which would help the computer, if it were interrupted, to just store a little bit of information about where it was in this job and go to the other job. Now, I think chimpanzees, and in fact all mammals, have a certain amount of short-term memory of various sorts, and psychologists every year discover new ones. Neuroscience is the most exciting field now. I think for the last twenty years it was microbiology and genetics, and now you see many young people moving into the brain sciences because that’s suddenly starting to break through. And it’s not so much because of discoveries in biology, but because the work in artificial intelligence, of understanding how to make machines see a little bit, got so intriguing, and people could see if we knew a little more about the brain and we used this knowledge from this other field of artificial intelligence, we’d get something good. Well, there were these experiments teaching chimpanzees and gorillas words. Koko here in San Francisco learned hundreds of words, and she could make little sentences. Now, why couldn’t she make the rest of human grammar, the kind of thing that every three- or four-year-old child can learn?

Minsky

She used sign language, because her throat wasn’t so good for talking. But she learned a lot of words and a lot of nouns and verbs and some adjectives, a lot of the stuff that young children learn. But at a certain point it couldn’t go further. Here’s a theory that comes from our laboratory about this. If you look at a sentence, you’ll find that adult sentences have interruptions. I could say, “The man who wore the white shoes went to the mountains.” Now, that’s really two sentences, as everybody has known since modern linguistics. First there’s one sentence hidden in there, “There was a man who wore white shoes.” The other sentence is, “The man went to the mountains.” What you’re doing is, I’m telling you something about the man who went to the mountains, but suddenly it occurs to me maybe you don’t know which man it was, so I interrupt myself and I interrupt you, and I say this other thing about who wore the white shoes. Now that’s something Koko was never able to do: make what you call a relative clause, because you mustn’t lose track of the fact that he was going up the mountain—I almost lost track of it myself right now, because of jet lag. So what you can do is you have a whole mental state. You want to say something. You interrupt it for a moment to solve a sub-problem, to develop another idea, and you can get back. Now, I think chimpanzees, generally, and gorillas can do this to two levels, but one of our students, Mitchell Marcus, at MIT some years ago tried to make language machines, and he discovered with two levels you couldn’t do so much. With three levels he could do almost all of the kind of grammar that young children have. So there’s something that might have happened a million years ago, or two or three, we don’t know. But you see, when people say the mind is so mysterious, yes, but you have to think, and take science to get—isn’t that such a simple idea, that maybe the difference between the animals up to then and the later ones was just having another layer of being able to interrupt you after you’ve been interrupted and get back both levels? Animals don’t seem to be able to do that. That’s why they can’t use tools.

Minsky

Of the interrupt circuits. Same thing that we did in computers in 1965. And that’s a baby theory, and my book has three hundred pages, and each page has a little theory like that. I don’t really care if two hundred and fifty of them are wrong. I mean, having fifty correct ideas in a field like that would be too good to expect. But the idea of this book is to tell people: don’t think that you’re being dehumanized when you try to understand yourself as a machine. In the past you’ve thought of yourself as a sort of jelly, a sort of unstructured thing which is just plain good, but no reason for it. I think it’s better, maybe for children, to think of yourself as a package of hundreds of skills. My colleague Seymour Papert works with children, and sometimes you find a child who’s doing badly—he invented the language Logo, which is very popular in schools these days.

Minsky

Because they can pick it up so quickly, it’s very clear.

Minsky

Yes, once the kid gets started, they really get into it. One of the things that we had in our experience here was that you would find a child, for example, who was having trouble with mathematics, and you’d talk to the child and say, “What’s the problem?” He says, “I’m no good at math.” See, that’s the bad side of the wonderful philosophy that we have a wonderful spark: that there’s a central being in your mind without much structure, and it’s either good or bad, you see. The child has no choice. He has to say, “I’m not good at mathematics.” Now, how could that be? But we find once a child has learned more about computers, the child can say, “Well, I’m good at lots of things. I have a bug in the part of my mind, the processes that do arithmetic. I’m not ashamed that I can’t carry the one, or forget it; that’s just something in a little bit of memory. Maybe I’ll do something about it.” When I was a kid, if you had to carry a one, I’d put it in my pocket, or I’d put it on a tooth. You just tell the child, “You’re a complicated thing. You’ve got a lot of parts. Most of them are great. Look, kid, you can talk. Even our scientists don’t understand that. So you’re having trouble with math; there’s not much to it. Why don’t you try to figure out which of your skills need developing?” So what I’m saying is, thinking of yourself as a society gives you hope. It’s almost the opposite of what people think, that it devitalizes you.

Minsky

The whole is more than the sum of its parts, all right. The whole is exactly the parts and the way that they communicate with each other. I think in general holistic thinking is not a good thing; on the whole, I should say. Usually you find it’s the way people stop facing problems and stop working on details, and convince themselves that all you need is a blurry idea of the thing. Now, what is true is if you have a complicated system you must have a childishly simple overview of it. For example, suppose I tried to understand IBM. This is a huge company, hundreds of thousands of people, they make thousands of products. I can look at it as a whole and say, “Well, it’s like a big, hungry animal, and it’s got a brain which is the people who do this or that.” I’d probably be wrong; it may be that the people who do the hardest job are in marketing, and the scientists are not.

Minsky

That’s right. So I think that to understand anything, you have to understand it at many levels. You need a very simple way of looking at the thing as a whole; you need a more complicated way of looking at middle-level parts; and you need a great many very small theories at the bottom. And so I don’t mean to say I don’t like holism, but I think the idea that there is a way to understand something deeply all at once is a dangerous and deadly illusion. On the other side, if you understand a lot of details—and not just one, let’s have five different ways of parodying. Let’s think of IBM as an animal; let’s think of them as a machine that makes money; let’s think of IBM—I’m running out—let’s think of it as like a field where you plant seeds and they grow—whatever you want. It’s important to have not one holistic view of something, because there’s no correct one, but have a whole lot of layers. And that’s what happens when children grow up. Whatever they learn, they make more and more layers until they stop developing.

Minsky

Yes, but I have a particular gripe with that, because of course I love Mozart, and I do have a talent; I can make up mediocre Mozart sonatas. I’m one of those people who can improvise classical music. I’m no Mozart, but what I can do is make stuff that people say, “That sounds just like Mozart,” and I say, well, little do they know. But here’s the interesting thing—I don’t think of this as creative at all. When you asked me that question, you made a sentence. A hundred billion neurons did something that nobody in the world understands. It’s just because every child learns to talk that we don’t think of that as creative. So I don’t think our most creative people are in any important way different, and when we admire what we call geniuses, they just think a little bit differently. I’ve had this experience. I know how to talk Mozart. I don’t know how it works; in fact I have trouble writing the music down, I just do it. But to me it’s a very clear experience. It’s the same as talking. I’m explaining something, but I don’t have the slightest idea what it is, and I don’t think of it as creative. I think before people fall for the idea that there is a creative spark they should look at themselves talking to a friend or a neighbor, and then consult a linguist and say, “Do you understand anything about how I did that?” And if they’re honest they’ll say, “As far as I can tell that’s just as complicated as what Beethoven did in the symphony, just to say a paragraph.”

Minsky

Oh yes. But you see, people say there’s such a thing as having a goal, and that’s sort of a primitive thing that a machine can’t have, but I don’t see why they say that. My friends Newell and Simon some years ago—they’re also early workers in this field—discovered a wonderful way to make a machine have a goal. You would give it a description of something, what you want—maybe let’s go back to that glass of water. You want some water in your hand. You see some over there. You say, what’s the difference between—the goal is what you want, but you don’t have to want it, all you need is the machine I’m describing. It compares this state of the water in the hand with the one there and says, what’s the difference? The difference is it’s too far away. Let’s do something to reduce that difference. So maybe I have to walk over there.

Minsky

That’s right. What I’m saying is those aren’t mysterious. Norbert Wiener talked about this in 1950. Goals are machines that reduce the difference between something that’s fixed and something you can change. Now, in this half billion years of evolution, we got brain centers that have about twenty goals. They’re triggered by various things. If there’s too little water in your blood or wherever the sensors are, then it turns on this machine that is built: it’s programmed by thousands of genes that have accumulated by trial and error to get water. When you’re hungry, it turns on something, you get food. When you’re a baby, the way you get food is to cry, and your mother gives you some, but gradually you learn other ways. And so we start out with about a dozen goals. I think the thing that made humans different from the other goal machines that are the other animals is that we discovered that if you want something, not only can you do something to get it, but it pays to learn more about the thing. So as a child grows it gets new subgoals that weren’t there in the genes, which are: let’s get knowledge because it might come in handy, and let’s get power, learn skills, so that we will be able to solve problems we don’t even have yet. So to me, the important thing about the child is the fact that, for the first time in evolution—in other words, intentionality is not exactly the thing to focus on; worms and flies have it, in the sense that you can see them pursuing a goal that’s moving around. The great thing about humans is a more neutral way of accumulating knowledge and power that you don’t need the very next minute, so that if you ask, “What makes the difference between man and animal or man and machine?” you find it isn’t what people say much. Sometimes it’s the ability to think without a goal, that as far as we can tell animals don’t have. I have a wonderful cat, and when it isn’t looking for food or a mouse, it sits there contentedly and it’s not doing anything, because it’s too dominated by goals, and when it doesn’t have one it’s just on hold. But you and I can’t stop thinking.

Minsky

Oh well, I’m very elated by what’s happened in that forty years, because I started thinking about these things in college in the late 1940s, and like other fields, every year something wonderful has happened. What I’m afraid is that if I go back to saying maybe there’s four hundred important mental functions realized here, maybe it’ll take four hundred years, in which case what I should do is work on the life-extension problem. But I think once we get beyond skepticism, because there’s skepticism about many things—once we get people to say, “don’t assume that something is a mystery, but dig into it,” that these problems will go away. We’ll understand emotions—I have a nice theory in the book of why pain hurts, which is people don’t even ask that, they just assume that it should. And what I think hurting is, is it’s reducing you to infancy. As you grow you get these many levels, and I bet the pain nerves are connected so that you can’t do this thought. If a lobster has got you by the toe, then what the pain nerves do is reduce the level of your thought so that you get down to the immediate goal and you have to do something about it. And the reason it hurts is that the rest of your brain can’t stand being postponed. So I think you take all the mysteries, and there are lots of little theories that you can make for those.

Minsky

Oh, it’s been a pleasure.