The History of Artificial Intelligence [Documentary]

Hello, and thanks for tuning into Futurology, an EarthOne subsidiary. This video is a compilation of some of the best documentaries on AI and is an introductory video for this channel's AI series. Which will take us on a journey: from the ancient origins of the field of artificial intelligence, to the birth of modern computational-based artificial intelligence, leading up to and encompassing the various types of machine, deep and reinforcement learning algorithms and then concluding with more grandiose sci-fi topics. The reason we wanted to do this is to give further context into how the field of AI has evolved, the prominent groups and figures responsible for it and most importantly so we can appreciate the advances in technology we all often take for granted today! Enjoy the documentary, consider subscribing and let us know your thoughts on the subject matter in the comments below!

Good Evening. I'm David Wain and as all of you are, I'm concerned with the world in which we're going to live tomorrow. A world in which a new machine, the digital computer, may be of even greater importance than the atomic bomb. Can machines really think? Even the scientists argue that one. I don't believe that we can say yet that machines do think. I have a basic question which I always ask, and that is, are these producing anything really new? Until I see a machine producing genuinely new things, I will not agree that machines think. I confidently expect that within a matter of 10 or 15 years, something will emerge from the laboratories which is not too far from the robot of science fiction fame.

[Music]

The Thinking Machine starring David Wain. In a moment, that story.

[Music]

[Applause]

[Music]

With me tonight is Professor Jerome B. Who is Director of the Research Laboratory of Electronics at MIT. Dr. Wiesner, but what really worries me today is what's going to happen to us if machines can think? And what interests me specifically is, can they? Well, that's a very hard question to answer. If you'd asked me that question just a few years ago, I'd have said it was very far-fetched. And today, I just have to admit, I don't really know. I suspect if you come back in four or five years, I'll say, "Sure, they really do think." Well, if you're confused, Doctor, how do you think I feel?

Professor Wiesner, I don't think I have to tell you that the conception of the robot, a thinking machine, has been man's dream for centuries. Also his nightmare, of course. Up until recently, the exploitation of that dream has been largely in the hands of its fiction writers and my colleagues in the motion picture industry. You remember a great robot in the silent film Metropolis? No, I don't think I ever saw that one. That I'd like you to see. That'd be very nice. Okay. Charlie, roll him.

[Music]

[Applause]

[Music]

[Music]

[Applause]

[Music]

That movie, that professor, is still way ahead of us, you know. Is he? He is. And you know, if I'd seen that movie as a youngster, I've probably been afraid of science. Doctor, when that film was made, it wasn't too long ago. The thinking machine concept was still in the realm of make-believe. What do you think about it today? Well, the reason that's such a very hard question to answer is we know so very little about thought processes or about information that makes up thought processes. You know, there are many things which machines can do today which, if they were done by human beings, you would certainly call thinking. Have you ever watched a child try to learn the alphabet? I've got a piece of film here that I'd like to have you see.

Do you know that letter? No. It looks like an M, but it isn't. It's a W. And then what? Tie down, isn't it? I'm going to make an X there, and you see if you can draw over the same way that I make down, up, down, and up. Can you do that the same way? Honey, that's right.

[Music]

That's fine. What letter is it? W. That's right. Good. Now, you see, of course, she makes mistakes at first. The real question is, how does she ever learn to get them right? I can copy yours. That's right. You copy mine. What do you mean to tell me that you don't even know yet how a child learns the letters of the alphabet? Psychologists who study the problem have a lot of ideas about what goes on, but no real explanation. A letter is simply a pattern, and we don't yet know how the brain recognizes patterns. Do you know what letter that is? Oh, no. It's a P. But do you think you'll ever understand these problems? I think so. That's one reason we're so interested in the computer and what it can be made to do. Are you suggesting that the computer can do what this child is doing? Yes, I am. The computer is a remarkable machine. And later on in the program, we're going to see it write an original TV Western. But first, let's see how the youngster is doing here. What letter is it? Well, P. Now, can you tell me what this letter is? Do you remember this one?

[Music]

It's a W.

[Music]

Now, let's see if the computer has as much trouble with the alphabet. You mean we're going to see the computer do with the child? Eh, I hope so. If that really proved that machines can learn, I'll tell you what it does. It compares the letter you write with a letter you first showed it. And as it gets more and more information about what is called a W, for example, it is more and more able, or able to make a judgment about whether the letter that's being written on the screen is, in fact, a W or not. Let's see. Okay, Laurie. Let's read on the program again. Well, at least I know what a program is. It programs the rules you want the computer to follow. Now, what are they doing? Showing the computer the alphabet for the first time. Yes, in just the same way the teacher did with a child. Incidentally, for convenience, you see, we're using P's and W's. Okay. Now we have those characters, and the machine will try some.

[Music]

Call it a day. It's pretty poor, but on the other hand, it doesn't have much knowledge. Well, here's a P.

[Music]

Try another P. Should get it this time.

[Music]

Wow. Well, now, that's luck. That's pure luck. No, as a matter of fact, one thing the computer can't do is fool the teacher. On the next try, look at the line below the letter. The longer the line is, sure the computer is of its answer. Good enough. That's a pretty good percentage, as indicated by the bottom bar. Very good. Don't you call that learning? Well, I know that if my wife and I saw our kids going through that process, we call it learning. So machines really can learn. You know, we don't really know much about the thought processes, and this is why I'm so hesitant. But we are studying the problem. That's being studied in a very many places and by people in many disciplines. You mean psychologists, psychiatrists, philosophers, and so on? Yes, and electrical engineers and mathematicians and others. Well, if the computer is this important, live, and I heard more about it. Well, the computer is a relatively new thing, and we're just really getting an appreciation for the full range of its usefulness. Many people think it's going to spark a revolution that will change the face of the earth almost as much as the first Industrial Revolution did. Well, now, this brings me back to my original question. Can machines think? I mean, by that, thinking that process we try to avoid when we have a problem to. So you mean like I'm trying to do in avoiding your question? Do you remember that puzzle about the cannibals and the missionaries? I do remember that. That's when we tried to cross the river with the missionaries and the cannibals without the missionaries getting eaten. Yes, that's the one. Well, out of Carnegie Tech, Professors Simon and Newell are doing some very interesting work trying to understand that kind of logical problem. Let me show you. This is Professor H. A. Simon. Three missionaries and three cannibals are trying to cross a broad river so that they can reach a town on the other side. They have a rowboat. Cannibals. They have a rowboat which will, however, hold only two at a time. All of them know how to row. There's one difficulty. No group of missionaries and cannibals may be on either side of the river at any time if the cannibals in the group outnumber the missionaries, because the missionaries would be eaten. Now, the problem is to plan a series of trips that will get all three missionaries and all three cannibals across the river in the boat without being eaten. You have any questions? Now, reviving. Well, you may start now. Please remember to say aloud as much as you can about what you're thinking, and you may move the missionaries and cannibals to try different combinations of them. There's only one boat. There's just one boat, and only two can get in at a time. Why don't you take them across and see how it would work, remembering that the boat always has to come back with someone to get the next load? And then I have the missionary bring the boat back and take one of the missionaries. Careful, that man is going to get eaten. Well, I guess I wouldn't do that. Is it possible for any of them to swim? No, I'm afraid they're not. Well, I'll bring one of the missionaries over myself. Poor Barbara. You know, I sympathize with her. I remember how baffled I was the first time I tried that problem. Well, Barbara is bright, and she'll get it after a bit. It's a matter of fact, in a moment, I'll show you her solution. That's why Professor Simon asked her to talk aloud as she works. They'll have a record what she says, and then bring the other one over, and that's it. Very good. That's the correct solution. And now we all go and try the same thing out with the computer. On the coast. Oh, no. Now, don't tell me you're going to try to get that computer to do the same thing. Is it found the solution already? Yeah. And in just a second, we'll have a chance to look at the answer. Is that the answer you have there? Yes. As a matter of fact, it's both answers. We've typed out Barbara's solution on computer paper, too. Can you tell which is which? Do I have to answer that? How does the machine do it, Doctor? Well, Professor Simon thinks that logical processes are really quite simple and only appear complex because there are so many of them in cascade. What does that mean? That the machine tries every possible answer? No, it does just what a person does. It tries those things which seem most likely. Well, how does the machine know it's likely? Well, a person knows what's likely on a basis of experience. The machine knows what's most likely on a basis of probabilities or reasonableness that has been programmed into it. That is, told to it in advance. Well, what else can computers do? Well, many things, though, as I've already said, we're just really beginning to understand the capabilities of the computers. I've got some film to illustrate this point which I think will amaze you. That man isn't playing checkers against a computer, is he? Sure. And it plays pretty well. Now, which colors the machine playing? Black. The player pushes those switches to tell the machine his moves. Incidentally, he has to make the machine's moves on a checkerboard as well. Well, then, how does he determine the machine's moves? By watching those lights on the console. Machine also prints out its moves. Who's that watching? He's Dr. Al Samuel, MIT graduate, now with IBM. Dr. Samuel programmed the computer to play checkers so he could study machine learning. Let's see what it's printing out. Huh? How in the world did it do that? Back to easier. How does a computer work? Well, it would take too long to give you a detailed explanation, but I think I can give you a simple explanation of the principles. If you want, you can make a little box they can put signals into, and every time you put a pulse into it, it adds. And with these devices, you can make adders, you can make multipliers, and you make devices which do a number of other mathematical operations. Now, with these various building blocks, and it takes just a very few of them, you can organize much more complicated mathematical problems. And what you do is what we call programming. You lay out a series of steps, and you have to tell a machine every blessed single step you wanted to do, because it can't do anything unless you do this and this. We call a program. We feed it into the machine, and it carries out the operations we've told it to do, one at a time. Of course, there is a saving grace here which is very important. And once the machine has learned how to do something, it can print out into its permanent memory the tab, or on a piece of paper of the kind we've seen, the instructions for doing this, and you never have to think about that particular thing again. Well, does the nervous system work like a computer? No, not really. Though there are many similarities between computers and living nervous systems, but neurophysiologists who work on the problem think there are many more differences than there are similarities. Fact of the matter is, though, that both systems use electrical signals. There is electrical potentials and pulses. Well, then, by following the complicated steps that a computer takes, the steps and stages in solving a problem, can you learn how or more about how the nervous system works? Well, what you can learn is a good deal about thought processes, or at least simulated thought processes, by this method. But it's very dangerous to carry this analogy too far. Well, if I told you all about my particular problem, could you solve that on a computer? Well, it probably takes a psychiatrist, right? I think I better ask my question over again. You have to tell a computer what to do. In other words, all of the computers that we've seen then program, that is, have been told by men what to do. Yes, but you know, that's really not a valid argument, because there's every evidence that men are programmed too. That is, they have certain built-in programming. Now, wait a minute, Dr. Wiesner. Are you suggesting that men are born with something put in their brains like men put information into a computer? Yes, they're not only born this way, but they get programming in other ways. That is, they're really two ways in which you can get programming. There is a hereditary part, that is, the part you're born with. You can also get information or programming in your brain by experience, that is, by learning. Well, surely in higher animals, learning is more important. We like to think that anyway. Can you give me an example of programming in something that's alive? Yes, I can. I've got something over here. Can you tell me what this silhouette's supposed to represent? Well, it doesn't remind me of anything in particular. If I hold it up here like this, does it look like a goose to you? Yes, it does. Now, if I turn it around, what do you make of it? Well, we're still thinking about birds. I'd say it was a hawk. Did pretty well. Let's see if a duck can do as well. The ducklings used in this experiment had been raised in isolation. I've never seen any other birds. And there. Is our friend the goose?

[Music]

There wasn't any reaction at all, was there? No, there wasn't. That's what interested Professor Melzack. You're seeing what happened when he first did this experiment in the London Zoo.

[Music]

You are a duck. Would you worry if you saw a goose? You mean he will worry when Professor Melzack changes that goose into a hawk? Well, let's watch. That's a really frightened duck. Well, not all ducks react this way. Some are simply worried when they see their first hawk.

[Music]

Well, Mr. Wayne, well, when I gather from this is that this duck that never had been exposed in its life to any other kind of a bird can now suddenly differentiate between a goose and a hawk. Now, you call that some kind of programming. To me, that's what I would call instinct. Well, I'd agree with you. Instinct is the word you use. I'm using programming, or at least that part of programming which is determined by heredity. You see, animals seem to start life with a large part of their nervous system knowing what to do. That's the reason why we breathe, and why our heart beats, and why babies know how to cry, or when to cry when they need help. Well, do you mean that instinct turns out to be some kind of programming that we're born with? Yes, that's using our language. Let me show you some very interesting research that was done on the frog last year at MIT by Professors Leffert and Morano, which seems to indicate that some animals, at least, are born with very much more built-in information than we used to suspect was the case. Well, does it show that we're born with a program? Well, what it does show is that in the frog, at least, the frog's eye reports only very specific information to the brain. You mean that the optic nerve of the frog doesn't report everything it sees? No, it reports only very specific things, things which seem to be very important to the survival of the frog. Let's take a look at the next film.

Well, now, the frog doesn't pay any attention to those dead flies. Didn't he? Hungry? Sure, but his eyes don't see them. He's hungry, all right. Watch. But if he didn't eat any of the dead flies provided, he'd eat the fly on the string. Well, you know, that's what Professor Leffert's trying to find out. It seems that the frog only sees things that move. He's looking through the microscope into the frog's brain, preparing to put a tiny electrode into one of the fibers in the frog's optic nerve. You mean when the fiber sees something, it'll send a signal to the electrode that we can see? Yes, and you'll hear it too. Now, Professor Levin is putting a target, a semi-circle, a hemisphere in front of the frog's eye, and then using a magnet on the back of the target, he can move a small metal disc around the target until he finds the point at which the particular fiber is looking. There he's found it. See, it sends in an electrical signal through the electrode. But what's the professor doing now? Well, he's trying to find out just what kind of things make this particular fiber react. It looks to me as though this fiber reacts every time something moves. It does. Well, that could explain why the frog didn't eat the dead flies. They didn't move, so he couldn't see them. Well, that's Professor Leffert's suggestion, anyway. He said to himself, suppose the fibers in a frog's eye look only for specific kinds of things in the world around them. Now, as we've seen, one kind of fiber apparently only reports movement. Here, Professor Leffert is looking for a different kind. You see, this fiber reports to the brain when anything small enough to eat moves into view, and it keeps reporting as long as the object remained there. Well, why did it stop reporting when the light went out and then failed to report when it came back on? Well, think of the light going off as a shadow passing overhead. And I'll say, and apparently in a dangerous situation, the frog's eye only reports the danger, not food. Watch now as a long bar has passed through the field of vision. The bar is too straight and too big to be a bug, so you see there is no reaction. Professor Leffert's work suggests that the fibers from the frog's eye only report specific things to the brain, things related to the frog's survival. One group of fibers looks only for sharp edges. Another group, this one, seems to be a bug detector. Well, now, I was always under the impression that the eye reported all light patterns. So are we all? I ought to add, though, that Professor Leffert's theories are not yet accepted by all people in the scientific community, or at least insofar as the implications of our human vision are concerned. Well, if Professor Leffert's theories are correct, we would have some explanation for instinct, wouldn't we? Well, the existence of built-in coding or instinct was never really in doubt. All this is a very interesting demonstration of it. It's a demonstration of at least one kind of instinct, because it's obvious that the frog didn't learn to recognize flies, but was born with this ability already in its brain. Well, now, if I may say so, people like frogs? Are they? Well, no one really suggests that people are pre-coded to the same extent that frogs are, but it's very clear that people also are born with a certain amount of information built into their nervous system. Let me show you something that'll prove this point. We're going to Geneva to the laboratories of Professor Jean Piaget and his associates. This is Dr. Barbara Helder, working with a youngster of five. By the way, how's your French, Mr. Wayne? Well, it's not very good, I'm sorry to say. Well, Dr. Ian Helder here is filling one glass with milk, then she's asking a child to fill his with exactly the same amount as she has. And hers.

[Music]

You've a little too much, is it? Now the same, exact one of them. Shoes? Okay, so perfect. Yep. He's a real stick with me. Who's that? Nick wanna know exactly when I'm in Rosa for wasn't my bra morally North Sea conveyer. I would prefer to drink my milk in the tall glass. I have poured it all in the tall glass. Hey, what you and I have? We still the same amount? No, there's more soup in this glass. But how do you know? Because it is taller. Yes, sure. Let's delight to listen now. How could that child make that mistake? Well, apparently, when we're born, we rely exclusively on our eyes to form our concepts of the world around us. He apparently had certain preconceived notions of the world around him, and some of these are wrong, and as he grows up, he's going to have to learn to correct them. Well, then, our eye apparently is programmed to tell us that that which is taller holds more. Yes, and in nature, isn't that generally the case? Well, I've always been under the impression that seeing is believing. Well, actually, it's just the opposite. These experiments indicate, but we tend to see only those attributes of objects which our nervous system is designed and programmed to see. Then seeing isn't believing, but believing is seeing. Well, it works both ways, but certainly believing is seeing. Watch this next piece of film.

[Music]

That's a Wiener. I've seen this before. Now, that window isn't square. You see, one side is longer than the other. Yes, but no matter how many times you've seen it, I'll bet you still can't see through the illusion.

[Music]

What's the window doing now? It's revolving. I know. Who? It's revolving, of course, it is. But I'm certain that you don't see it that way. You see it as oscillating back and forth. As a matter of fact, you're right. You see, your eyes tell you that anything that is longer is closer to you, and your experience tends to confirm this. So you have to see the long side of the window as though it were in front, even when it's in back. It just can't see it revolve. Now, let's cover the window with a cloth. The cloth sides are equal in length. Now, maybe your brain will let you see the window rotate. Look at the bottom corner of the cloth. Now, do you see it revolve? Yes, I do.

[Music]

Now, let's put a tube through the window. What do you see this time?

[Music]

Well, that tube is bending back on itself. You know why? That's because you've assumed the tube is made of rubber. Now, if you tell yourself that it's made of steel, you'll see it cut through the window instead of bending. Am I right? Well, I have to think real hard that it's steel. Right? Right. It's still. It's still. It's... Well, I'll be. Yes, you are right.

[Music]

Now, then, doesn't that prove that you tend to see what you believe? You mean, professor, that I am somewhat programmed? In other words, that there are rules built into me that make me react sometimes similarly to a machine? Yes, you and everyone else. Well, I think I'm beginning to get your point. I am born with certain rules built into me, but I think I can think. That, as I say, I can think. Therefore, I shouldn't be too upset when a machine thinks, just because it has rules built into it by man. That's logical, isn't it? What can you show me where a computer can do anything original? I think that might help to convince me. What depends an awful lot on what you mean by original. Would you regard writing a television Western and being original? How do I have to answer that? But do you mean to tell me that a computer can really write a play? Well, we can write pretty good plays. Matter of fact, we will see the play written by the computer in just a moment. But first, from AMF, an expression of one company's aims in your interest. You were saying that a computer can write a play? Sure. I'll show you. It won't be as good as Shakespeare, but it'll be better than a lot of the Westerns you look at. I suspect by now you've seen enough of computers in action, so let's just watch the computer print out its play. That's Harrison Morse, who did most the work on this computer program.

[Music]

[Music]

[Music]

[Applause]

[Music]

[Music]

[Applause]

[Music]

Well, I'll tell you one thing. If that computer ever learns to act, I'll tear its transistors out by the roots. You know, that's as close to magic as anything I've ever seen, Doctor. Well, you know, it isn't really magic. Let's have Doug Ross, who's on the staff at MIT and who supervised the writing of the program for this playlet, explain it to us. Well, we had a lot of fun working on this program, but we're not just playing games. We're trying to illustrate some important things about artificial intelligence. Just as a human playwright must obey certain rules in order to have a meaningful and understandable play, one that seems natural for people to actually act out, we must make the computer aware of the same kinds of rules. So what we're trying to show are that intelligent behavior is rule-obeying behavior. We're trying to show what these rules look like, and we're trying to show how a computer can be made to do creative work. In the type of play that our program is designed to write, we have a robber enter, wait for the sheriff, and when he enters, they have a shoot-out, and one or the other will die, and the winner, if any, will pick up the money and we'll all walk out the door. The human playwright would know already things that we have to teach the computer by programming. For instance, if the gun is in the hand, and the hand is on the robber, and the robber is in the corner, the human knows immediately that the gun is also in the corner. But we must make the computer able to keep track of all these things. This switch shows how the computer will choose reasonable alternatives for the sheriff, depending upon whether or not the sheriff can see the robber and the robber sees the sheriff. The sheriff may wait, advance on the robber to get a better shot, or try immediately to shoot the robber. We've given the computer rules for determining reasonable behavior, and we have also given the computer rules for modifying those rules. For example, we have an inebriation factor which controls the actions of the robber, depending upon how much he has had to drink. The more the robber has to drink, the more inebriated he will become, so that he becomes less and less intelligent in his behavior. Notice that the computer is still being intelligent and not violating physical rules, but the character of the actor is changing as he becomes more and more inebriated. And that is one of the main points that we're trying to show here is that intelligent behavior is rule-obeying behavior, and there is no black magic about doing these things on machines. It's marvelous to do the mind machines, but far from miraculous. Well, now, that makes it seem almost reasonable. But even so, how can I be sure that the computer wrote the script and not the scientists? Well, you know, if we had written the script and stored it in the machine's memory, it would print out the same play every time, doesn't it? No, it doesn't. As a matter of fact, all though the plot is the same, the play it writes is different every time. It's sort of like a mystery story writer. Matter of fact, we've written about 50 plays, and here are several of them the machine has printed up. Would you like to see another one of them? Certainly would. Okay.

[Music]

[Music]

[Applause]

[Music]

But I can see there's one thing that the computer doesn't know. In television, the bad guy's supposed to lose, you know. You know, there are many things the computer doesn't know. And even worse, sometimes the computer just doesn't work. But what happens in almost anything, for example?

[Music]

Well, what happened there? Well, you know, making a program for a computer involves a fair amount of trial and error. What you've just seen with one of the errors. Well, that's wonderful. Well, now, look, if a machine can do all these things today, what's going to happen tomorrow? Well, it depends on many things. Depends on how much we are able to find out about learning processes. As I told you before, there are several ways we're going about trying to understand information processing systems, including the nervous system. And one very important one is working with humans. This experiment is taking place in Professor Walter Rosenblith's laboratory at MIT. Through these earphones, Dr. Geyser will put a series of rapid clicks into the subject's ear. Now, he's going to record the signal coming from the electrodes in the headset, response to that signal, as it appears on the outside of man's skull. But hasn't this kind of recording been done for good many years? Well, no, recording from the skull has been done for a long time, but what you used to get was a picture of all the electrical activity in the brain. In these experiments, the computer lets us concentrate on the specific electrical activity that's a direct result of the clicks that the man hears. In other words, Dr. Weisenner, the computer allows you to learn a little more about signals in the human brain than you knew before. Well, the peaks and valleys you see there are a result of the clicks that that man is hearing. Incidentally, as it almost always does, pure research has a number of applied, usually unpredictable, values as well. In this case here, we're seeing the John Tracy Clinic in Hollywood. These techniques are being used to determine deafness in children. Another way of studying these logical processes we've been talking about that take place in the brain and in giant computers is going on at the Lincoln Laboratory. This is a laboratory which we operate for the Department of Defense. Here is one of the world's largest and most versatile computers. We call it the TX2. It took a great many men several years to build, and we're still working on it, improving it all the time. This enormous collection of wires, tubes, transistors, and circuits connects the largest computer memory now operating. That's the memory over there. All that a memory? Yes, it is. You know, it contains about two and a half million memory units, primarily ceramic cores in this case. Huge as it is, it holds only a tiny fraction of the elements that are housed in the human brain. People estimate that our brain holds about 10 billion neurons. Ten billion? Yes. And unbelievable as that is, this comparison is probably very misleading because neurophysiologists believe that each neuron fulfills many more functions than a single vacuum tube or transistor. So you have to use huge installations like this to find out how signals move between the brain's neurons. Well, you know, this is really pretty small to simulate anything as complicated as the brain. This gadget only has about 1,300 neuron-like elements. If you put signals into the two edges, you can make a wave travel through the device and study the behavior of signals in such networks. That's what Belmont Farley is going to do here. Our scientists everywhere are using computer machines like this to study learning. Yes, they are. You know, once an idea like the computer exists, people everywhere begin using it. That's one of the reasons why there can't really be any secrets in science for very long. A matter of fact, I recently saw work similar to this in Russia. Well, the fabulous machines like this are being developed everywhere. What's gonna happen to us all tomorrow? Who's going to be in charge? Machines or men? Man, I hope. You know, you can always pull the plug. A French artist named Jean Tingley tries to prove this point, as a matter of fact, by building machines that do absolutely nothing.

[Music]

You know, modern research can chew up money faster than thing. Waste machines doing.

[Music]

[Music]

[Music]

You know what he's doing here? I'm pumping that bicycle. He's dropping a picture.

[Music]

No, that really is wild. Well, how seriously, professor, do you think that one day machines will really be able to think? Well, I think so, but people still disagree about it. Let's hear what a few scientists have to say about it. I don't believe that any of the machines that we know today can think. I have a basic question, which is, do these machines produce anything really new? When you consider the great new ideas produced by men like Newton and Darwin and Galileo, you'll find that initially they had to throw away the old rules that they'd been brought up with. Now, machines do what they've been told to do. They obey the rules that have been fed into them by man, and we know of no machines at present which have means of overcoming this limitation. I have little doubt that we'll be able to produce machines and computer programs that will behave in a fashion that we speak of as intelligent. That these will be of great aid to man in terms of relieving him of intellectual work that is not fit for human production. Where my doubt comes in is whether we shall be able to produce machines and machine programs that are capable of creative thinking. I doubt very much, with the usual type of human vanity, that any artificial information processing system will ever be able to do this kind of inventive things. That I rather doubt whether it's going to be possible to do this in our lifetime. I'm convinced that machines can and will think. I don't mean the machines will behave like men. I don't think for a very long time we're gonna have a difficult problem distinguishing a man from a robot, and I don't think my daughter will ever marry a computer. But I think the computers will be doing the things that men do when we say they're thinking. Now, machines can't write good poetry or produce deathless music yet, but I don't see any stumbling block in a line of progress which will enable them to in the long run. I'm convinced that machines can, we'll think in our lifetime. Well, now, that's pretty unnerving stuff, don't you think? It's going to have tremendous repercussions in the days to come. I'm sure that it will. It's going to have many effects, direct and indirect. What would you say those effects would be? Well, in the direct effects, we're going to put machines to work for us in many ways. By indirect, I mean that we're going to learn many things while we work with the computers that will help us in other fields that we're interested in, for example, in the field of mental health, social problems, and economic problems, just to mention a few. But how do we put the computer specifically to work? Well, you know, my colleague at MIT, Professor Norbert Wiener, says we're living through the Second Industrial Revolution today. The first Industrial Revolution being the replacement of manual labor by machinery. That's right. And the second, then, you say, will be the assistance of the human mind by the computer. And I'm certain that as time goes on, we're going to find ways to do many things using the computer which the unaided mind just could not do by itself. The future of the computer is just hard to imagine. Let's listen now to Professor Shannon, whom we heard briefly at the beginning of the program. And also, I won't like you to remember that when he talks about robots, he doesn't mean well, a pretty girl like our good friend Dr. Rote. Violence? Yes, it was fun created at the beginning of the program, but what he really means is machines that can do things that man wants them to do. In discussing a problem of simulating the human brain on a computing machine, we must carefully distinguish between the accomplishments of the past and what we hope to do in the future. Certainly, the accomplishments of the past have been most impressive. We have machines that will translate to some extent from one language to another, machines that will prove mathematical theorems, machines that will play chess or checkers, sometimes even better than the men who designed them. These, however, are in the line of special-purpose computers aimed at particular, specific problems. What we would like in the future is a more general computing system capable of learning by experience and forming inductive and deductive thought. This would probably consist of three main parts. In the first place, there would be sense organs, akin to the human eye or ear, whereby the machine can take cognizance of events in its environment. In the second place, there would be a large, general-purpose, flexible computer program to learn from experience, to form concepts, and capable of doing logic. And in the third place, there will be output devices, devices in the nature of the human hand, capable of allowing a machine to make use of the thoughts that it has had, of the cognitive processes, in order to actually affect the environment. Work is going on in all of these fronts simultaneously, and rapid progress is being made. I confidently expect that within 10 or 15 years, we will find emerging from the laboratories something not too far from the robotic science-fiction pain. In any case, whatever the result, this is certainly one of the most challenging and exciting areas of modern scientific work. Exciting and challenging, but doesn't it worry? Well, sure, it worries me. But you know, the problems posed by the computer are really no different than the problems we have with other products of technology. It's going to take a great deal of wisdom on our part to manage them, but if we do, we're going to make a much better world. Thank you, Dr. Weisenner.

[Music]

I'm Claude Shannon, a mathematician here at the Bell Telephone Laboratories. Let's just be sure these two stood an electrically controlled mom. He has the ability to solve a certain class of problems by trial and error means and then remember the solution. In other words, he can learn from experience. Like his classical namesake, these two times the problem is finding his way to amaze his objective. As they go here in the corner, he is now exploring the maze using a rather involved strategy of trial and error. As he finds the correct path, he registers the information in his memory. Later, I can put him down in any part of the maze that he's already explored, and he'll be able to go directly to the goal without making a single fault turn. Of course, solving a problem and remembering the solution involves a certain level of mental activity, and if something perhaps akin to a brain, a small computing machine serves these suits for a brain. Authority to himself is much too small to contain even a small computing machine, and we have placed the brain cell, if you like, a pea soup behind the mirror here. This is a bank of relay telephone relay, and the job they do for T-shirts is similar to the job they do in your dial telephone system. Each time you use your telephone, the dial system has to remember the number that you dial, then guide your call through the maze of connection, the thousands of separate lines that a dial clicking off in a fraction of a second. It must find a trunk line for you to carry your call. It also has to remember in itself what sequence of steps are necessary to make the connection. Poison here at the Bell Telephone Laboratories, we're concerned with improving your telephone system, making it work better to give you a more efficient service. One of our continuing projects is on dial switches. Incidentally, the things we learn for the telephone system have other applications too. We can use the telephone relays to build computing machines, machines that can solve mathematical problems in a few minutes and would otherwise take many days to solve. We at Bell Labs also use this knowledge to build gun director equipment for the Armed Forces. The many factors which affect the aiming of gun and guided missiles are computed and applied by these machines in fractions of a second. Please do. This is a simple demonstration of some of the things we can do with telephone relays. He's now had a chance to find his way through the maze. From there, he's reaching the goal. Now, let's see how well he remembered what he'd learned. This is capable of other types of intelligent behavior. He can add new information and adapt to changes. If the entire maze or part of it, for example, it's changed, he will explore the change area, replacing the old information it's no longer a value by the new information he's just learned. Here, let me show you what I mean. Suppose I have changed the maze around a little bit. Let's say I take this partition and move it over here, and I move this one over here. There are over a million million different ways you can set up the mazes on this board. Now, I'm gonna take the mouse and put him back in the same starting point, and let's see what he does. If replacing the old obsolete information with what he is now learning about this new situation. Those copper whiskers of his. How I'm running up against the wall. Let's try a different direction. Now he's through the change section, and his memory is correct to the rest of the way to the goal. For a bar magnet mounted on three wheels, this is rather agile. Here again, no, it isn't the mouse that does the actual work. The motor and mechanism that really move him around are monitored underneath the maze floor. Let me show you. The mouse is actually moved by this electromagnet, which can move on a carriage in two different directions, driven by a pair of motors here and here. The position of the mouse is sensed by the machine in terms of a number of reed switches, which are located under the different squares of the maze. If the mouse is brought into one of the squares, it closes the appropriate reed switch. This gives a signal to the electromagnet to move over to a position underneath that square, and then the relay circuit takes over the control of the electromagnet and thereby the mom.

[Music]

[Applause]

[Music]

When thesis is exploring a maze, he rotates his trial direction for any square in a clockwise manner, north and east and south and west, until he's able to escape from the square. He also takes account of the direction by which he entered the square and what his previous knowledge from the last time he was there. A theorem from that branch of mathematics known as topology guarantees that his method of exploration will eventually solve any possible maze. Of course, how do we mean that these who can solve any problem that can be solved like the rest of it? He occasionally finds himself in a situation something like this.

[Music]

[Music]

[Applause]

When the hero of any up-to-date science-fiction show has a problem, he usually turns to his trusty computer for an answer. Somehow or other, science fiction computers always manage to think of something.

[Music]

Sometimes they think so hard they get uptight, they blow their cool. This computerized robot spent the better part of a one-hour TV show thinking himself into falling in love with another robot, of course. But how about real-life computers? Do the so-called Thinking Machines really think?

[Music]

[Applause]

Well, do computers really think? To answer that, we'll have to find out what the word "think" means. Let's see if our computerized friend can help. My data bank dictionary has many definitions of Barbour. Thanks. What's the first one, please? To think: call to mind, to remember. People can't always recall things the instant they want to, but storing information and recalling it, if something electronic, computers do remarkably well. In fact, it is this ability that makes their operation automatic once programmed. It can refer to its own memory for instructions and data. Of course, memory devices aren't new. They came into existence as soon as man learned he could use substitutes or symbols to represent the things he wanted to remember. The next time he wanted to know how many cows he owned, he could refer.

To the sack of pebbles instead of grinding up the herd again.

[Music]

Put the pebbles on a frame, and numbers could be stored simply by changing the pebbles' positions. Mechanical memories use all kinds of symbols, such as lines carved in marble so you won't forget the glory of the past, or strings on fingers so you won't forget to pay your light bill.

[Music]

The familiar light switch is much closer in time and spirit to the kind of memory computers actually used: lamp on, lamp off. Either way, this simple equipment is storing one bit, or binary digit, of information. With more lamps, greater quantities of information can be stored, but mechanical switches and lamps are too slow. Modern computers use fast magnetic memory devices.

[Music]

Such as tapes or disks stacked like jukebox records, or tiny cores woven together like Indian beads. Large computers use these and other types of memory devices to achieve tremendous storage capacities. The computer being queried by this librarian can give her instant information on the whereabouts of any of the thousands of books in her charge. If remembering were the only definition of thinking, we'd have to put computers in the mental giant category. It is not to think, so subject to the process of logical thought.

Playing chess certainly involves a high degree of logic, and this computer at MIT has been programmed to play a very respectable game. Its human opponent studies his move on an actual chess board before feeding it into the computer. The computer then performs logic operations to determine its next move and displays it on this screen. It has even won first place in an amateur tournament.

But how can a collection of wired hardware be capable of logic? Basically, logic is a predictable series of facts or events, such as closing this switch and this one to ring the bell. In fact, computer people call this a logic AND circuit. The same components can be rewired into a logic OR circuit to ring the bell by closing either this switch or this one. AND, OR, and many other kinds of elementary logic circuits are the basic building blocks that form the complex logic networks we call computers.

The computer controlling this electronic telephone switching office makes millions of logical decisions every day, all with the infallible logic it takes to quickly connect one telephone to any of the more than 100 million other telephones in this country. That would be pretty good thinking if logic were the only definition.

There is also visualization: to think, to form a mental picture. All of us are capable of translating symbols into mental images, but computers are nearly as imaginative.

[Music]

They are very good, however, at taking abstract data, processing it, and producing pictures on cathode ray tubes. These are the complex motions of an orbiting satellite as seen from outer space. Even more interesting is the computer's ability to simulate designs or systems. This scientist at Bell Telephone Laboratories is designing an electronic circuit by drawing with a light pen on the face of a television screen connected to a computer. He can make changes in the design and simulate its operation without the necessity of actually building it.

[Music]

When it comes to memory, logic, for forming images, we computers are pretty good thinkers. Just a minute, aren't there any other definitions? Yes, Joe. Topor, save, or recognize. Mary, when it comes to recognition, computers are still pretty inept. They can be made to see by means of optical or magnetic sensors, such as those used by this bank check reader. But so far, computers are limited to recognizing simple, well-defined patterns like post office zip code numbers, provided they're tight and properly positioned. But teaching a computer to generalize, to recognize that all these symbols mean the same thing, it's more difficult.

When it comes to language translation, computers aren't very bright either. Some progress is being made, but the problems are enormous. Essentially, they boil down to one fact: there is no such thing as an absolute one-to-one correspondence between the words of one language and those of another. For example, when "out of sight, out of mind" was put through one translating computer, it came out as the foreign language equivalent of "invisible imbecile." Imagine what mechanical translation would do to slang. It is a frightening thought. May I go on to the next definition?

To think, to have feeling or consideration for. Computer programmers have been known to fall in love with their computers, but no computer was ever observed returning the sentiment. In fact, computers are absolutely devoid of feelings, and a good thing too. They neither play favorites with programmers nor get angry at their mistakes. Best of all, they never get bored. Like other machines, they can do the same monotonous chores all day long without complaining.

[Laughter]

[Music]

But they can be programmed to simulate human emotion. Getting arty? I'll bet your next definition involves creativity, right? Thank you, Joe. Create or devise. Creativity, a word just as hard to define as thinking, still seems to be a uniquely human capability. A computer has been used to produce animated pictures as well. One animated film produced by a computer is on the subject. To produce animated films on a computer, the technique has already produced some interesting results.

[Music]

The pictures. The music you are listening to was also produced by a computer.

[Music]

[Applause]

The computer, an ingenious collection of electronic hardware, was created by man. It is also man who creates the programs that make the computer the useful tool that it is. Without a program, a computer is no more productive than a player piano without a music roll, or a jukebox without a record. Still, whether the big machines are creative or not is irrelevant when you consider their usefulness and efficiency. Billions of correct mathematical operations between errors, the equivalent of a thousand people computing for a lifetime without making a mistake.

But do they think? Well, let's say they carry out some processes that are similar to human thought. Or better yet, let's just say it all depends on what you mean by thinking.

You.

[Music]

How do we differ from the computer? What else do we? The issues this class of kindergartners are struggling with have exercised some of the world's smartest minds during the 40-year history of the computer. For computers are not like other machines. Their ability to manipulate concepts gives them a mind-like quality, and from the very beginning, some people saw them not as a calculating device, but a mental aide, a thinking machine.

The thinking machine. Hello again. With me tonight is Professor Jerome B. Weizenberger, Director of the Research Laboratory of Electronics at MIT. Dr. Weizenberger, what really worries me today is what's going to happen to us if machines can think. And what interests me specifically is, can they? Well, it's a very hard question to answer. Could you ask me that question a few years ago? I had said it was very far-fetched, and today I just have to admit I don't really know. I suspect if you come back in four or five years, I'll say, sure, they really do think. Well, if you're confused, Doctor, how do you think I feel? We're just really beginning to understand the capabilities of computers. I've got some film to illustrate this point, which I think will amaze you. Batman isn't playing checkers against the computer, is he? Sure, it plays pretty well. Now, most computer scientists saw it as a mere number cruncher. A small group thought that the digital computer had a much grander destiny. Being a general-purpose machine, it could be programmed to do things which in humans require intelligence: play games like checkers and chess, and solve brain teasers. The field became known as artificial intelligence, and no one institution would be linked with its fortunes like the Massachusetts Institute of Technology. MIT, in 1958, two young mathematicians, Marvin Minsky and John McCarthy, set up a department to explore this exciting new intellectual frontier. They attracted a series of brilliant students.

[Music]

One of the first, Jim Slagle, got the idea of programming the computer to solve problems in freshman calculus. A complex function which Jim Schlegel decided to write a program which would try to solve the kinds of problems that MIT students do in the first year calculus mathematics. Merton was doing what we call symbolic integrals, and he wrote a program which consisted of more or less a hundred kinds of rules or suggestions. The machine would try various ones, and then it would use a special set of rules I called the rules of "tear." If, if the thing got too complicated, it would say, "That's no good, it's too complicated." But if it seemed to be getting simpler, it would follow it further. With the amazing thing, and this was 1960, just a couple of years after we started, then got an A on the MIT exam, and it was frightening. It was doing as well as the average student, or maybe slightly better. These first small exercises in artificial intelligence had turned out so well that hopes for the future were very high indeed. They were delirious. But in 10 years, they had gone from machines that moved bits around to machines that actually proved logic and played games, and everyone was amazed that these things which what they thought were just number crunchers could be minds, had intelligence, thinking intelligent thoughts.

Intelligence is a mysterious activity. While the brain, the hardware which enables us to think, is physical, philosophers have tended to locate our mental activities in a different, abstract realm that we call the mind. So that I can imagine that we're on the moon. Close your eyes now. Keep your eyes closed now. Imagine you're not on this planet Earth, but at a school on the moon. Right now, on if you live on the moon and you live in moon houses, and their mind is a place where we can hold and manipulate ideas as if they were real things. Have you island the moon now, living in moon houses, getting ready to go to the moon Montessori School? My house, would you live in there? Don't you wear the ideas can be about real things we know about the world from our senses, but they can also be completely imaginary, dealing with worlds we have never seen. A computer, too, can conjure up worlds, both real and imaginary.

The pioneers of artificial intelligence reasoned that if a brain can also be a mind, then so can a computer. The fact that the hardware of the brain, with its neurons, was completely different from the computer, with its vacuum tubes, was irrelevant. It was the thoughts that they manipulated which were important. The pioneers in the field of artificial intelligence had little interest in the question of how the brain was actually constructed. That is, they viewed, and continued to view, the mind as something different from the brain. The mind is a symbolic processing entity. The brain is, so to speak, the hardware on which the mind runs. Mind is analogized to software. Brain is analogized to hardware. Just as software runs on the hardware of a computer, the mind is what, so to speak, runs on the on the wetware of a brain.

Blindly copying nature's way of doing things wasn't always a good idea. Attempts at artificial flight based on the way birds fly had been a disaster. They had this interesting analogy that just like when you made planes, you didn't have to make them with flapping wings. The biology didn't make any difference. We just bypassed biology and evolution and made planes with motors, and we made rockets. And if we'd spent our time trying to make things that flap their wings like people did in the early days, we wouldn't have ever had anything that flies. So it doesn't matter at all how the brain makes intelligence. We don't have to model the brain any more than you have to make something that flaps its wings in order to make something that flies.

In the early 1960s, Hubert Dreyfus was a young philosopher working at MIT, one of the few representatives of the humanities in an institution which lives and breathes science. Students of his who were immersed in the exciting new frontier of artificial intelligence began trying to convince him that for philosophy, the writing was on the wall. Students would come into my courses on Heidegger and Wittgenstein and say that, "You philosophers have had your 2,000 years and you haven't come up with much, but we are beginning to. In fact, we've already practically finished understanding perception and particularly intelligence, reasoning, and so forth, and in effect, the ball has passed to us." And I was amazed. I didn't, I had heard nothing about that.

Can machines really think? Even the scientists argue that one. I'm convinced that machines can and will think. I don't mean the machines will behave like men. I don't think for very long time we're going to have a difficult problem distinguishing a man from a robot, and I don't think my daughter will ever marry a computer. But I think the computers will be doing the things that men do when we say they're thinking. I'm convinced that machines can and will think in our lifetime. We expect that within a matter of 15 years from the robot science fiction fan. Things, however, were to turn out rather differently.

The MIT scientists wanted their computer mind to interact with the world, so they built a gripper for a hand and a TV camera for an eye. The task they set it to stack blocks was, on the face of it, child's play, but it turned out to be more difficult than anyone could have imagined. It's easy enough to get a picture into a computer. The trouble is that a block or a box is different. You move it this way, it's a different shape, and so you almost never see the same thing twice. Sometimes there's shadows on it, sometimes it's darker or lighter, different boxes have different surfaces, sometimes there's things written on it. So that even though to you or me or a child, the idea of seeing a block seems simple, it's actually very, very complicated.

But beyond the problem of recognizing blocks, the program had some rather strange ideas about what happened to blocks when you let them go. For example, we have the robot build a tower of blocks, and guess what it did? It started with the top block. But if they're in space and let it go, because the machine didn't know that if you let go something, it will fall. Know that gravity, and it didn't know about the kinds of things that every two-year-old child knows. So it took several years to do that.

Across the Atlantic, things weren't going any faster. At the University of Edinburgh, another ambitious vision project was underway. This documentary from 1971 showed just how intractable some of the problems were. Freddie is a computer, and his world consists only of his round board and the objects on it. We immediately see this as a cup. Now, the computer tries after the first phase of processing. We have something which looks like this. The computer has sorted out the four major regions: top of the cup, body of the cup, hole in the handle, and this irregular region, which is the shadow. After painstakingly working out where the object ends and the background begins, the computer is ready to name what it is looking at, and it took 10 minutes. A two-year-old child would recognize it immediately.

The problems of moving and seeing at the same time were even more taxing. At Stanford, in the early 70s, Hans Moravec bravely tried to get his cart, a wagon connected to a massive computing engine, to cross a space avoiding objects in its path. Each flash represented 15 minutes of thinking time, which tied up the department's computers.

[Applause]

Each meter of travel of the robot was accompanied by 15 minutes of computation, and the crossing of a large 30-meter room took five hours. A four-year-old is already equipped with wonderful circuitry to look at the world and detect the presence of objects and to know their imminent collisions and to avoid them. So a four-year-old can toddle across the room with apparent effortlessness. The Stanford cart used a hierarchy of programs approach. For anything but effortless, the computational problems of vision were so immense and required such different kinds of computing engines than those available that many researchers decided to avoid the area and concentrate instead on a pure form of intelligence, a truly disembodied mind, one that didn't need eyes or hands.

In 1950, the British computer pioneer Alan Turing had speculated that one day machines would think. He proposed a test, the Turing test, which a machine would have to pass before it could be considered truly intelligent. Imagine you were sitting in a room communicating via a screen with an entity somewhere else. It may be a person or a computer program. I question and answer. You have to determine whether you were talking to a human or a machine. This entity doesn't need eyes or limbs. In principle, it doesn't need a body at all. All it has to do to beat the challenge is to use language intelligently.

[Music]

One of the first computer programs which sought to use language convincingly in this way was Joseph Weizenbaum's program Eliza. Here is a conversation between a computer (the capital letters) and a person: "How do you do? Please tell me your problem." "Women are all alike." "In what way?" "They are always bugging us about something or other." "Can you think of a specific example?" Although it seems to understand the words, it is in fact based on a series of tricks. Eliza understands nothing about the meaning behind the words. Often, Eliza just turns a reply into another question. "Well, my girlfriend maybe come here." "Your girlfriend made you come here." But it is geared to look for important words like "mother," "father," or "dream," then it will respond more aggressively. "She and my mother value therapy." "Tell me more about your family." And once you realize this, it can be easily fooled. "My opinion is that necessity is the mother of invention." "Tell me more about your family." Such a program could not possibly pass the Turing test and could not be said to use language intelligently.

Eliza was a light-hearted foray into a field which was full of difficulty. For remarkably, it would turn out that the human ability to understand spoken or written sentences was vastly more complex than solving calculus problems. Even if you take a simple everyday sentence like, "Mary saw the bicycle in the store window. She wanted it." It might refer to the bicycle, the store window, the store. In which it refers to depends not so much on English as on your knowledge of what people want and don't want. If I had said, "She saw the bicycle through the store window. She looked at it longingly and pressed her nose up against it." That "it" probably refers to the store window and not the bicycle. Now you have to bring in knowledge of human anatomy and the fact that people emotionally like to be near things that they want, even if they can't have them, and so forth. They hadn't reckoned with ambiguity when they set out to use computers to translate languages.

Electronic translates Russian into English, instead of mathematical wizardry. A sentence in Russian is to be fed. One of the first non-numerical applications of computers. It was hyped as the solution to the Cold War obsession of keeping tabs on what the Russians were doing. Claims were made that the computer would replace most human translators. Present, of course, you're just in the experimental stage.

[Music]

An adequate speed to cope with the whole outlook of the Soviet Union. Leave just a few hours computer time a week. When do you have to be able to achieve the speed? Our experiments, oh well, perhaps within Mr. McDaniel. Does this mean the end of human translators? Which I guess for translators of scientific and technical material, novels. But despite the hype, it ran into deep trouble. Yeah, analysis in Tosa Raeleen. The cantatas straightforward Russian passage poses no trouble for a human translator. So spear comp Romania, it's operable if scale algebra. In the present paper, we propose using matrix Boolean algebra, and we describe some results which we have already obtained in this way. But when it was given to a computer, this was all it could come up with: "To forum president, waiting for it to be offered to be proposed for investigation, research, analysis, expiration, paper." I say, "Dude, let's take advantage of matrix Boolean algebra and to be described bro series result garden in that in two to four own in this direction trend order permit."

Humans all over the world, even though they have never met, even though they have different languages and traditions, share a vast amount of common knowledge. And whenever Debora who could evoke knowledge of what human beings are and what they strive for, their goals and beliefs, their sensitivities and fears, before they could translate or understand languages, computers would have to know these things as people do.

[Music]

We're so good at using and understanding language that before computers came along, we never realized just how difficult the task was. Language is just full of ambiguity. What words mean depend on their context. Imagine the problems an alien from outer space would have understanding these newspaper headlines: "What kind of pattern are the police looking for?" Is this about breakfast food or Britain's Labour Party?

[Applause]

[Music]

What we began to see is that the things that people think are hard are actually rather easy, and the things that people think are easy are very hard. We could do the calculus with just a few hundred pieces of program, but to learn language, to recognize faces, to walk and to put your clothes on and do the kinds of things we expect every child to do, we still can't do with the robots of the eyes of 1990s. Once so promising, the fortunes of artificial intelligence now looked bleak indeed.

In 1972, Hoover Dreyfus wrote a history cataloging the failures of AI called "What Computers Can't Do." The same year, in Britain, funding for projects like Freddie was cut off following a scathing report of AI's progress by Sir James Light Hill. A male part of the field. Have the discoveries made produced the major impact that was promised? Minsky and his colleagues didn't give up. They set about working on machine learning, represent knowledge in computers so the computers could use that knowledge to resolve ambiguity in language learning. One project did succeed. Terry Winograd's program SHRDLU could use English intelligently, but there was a catch. The only subject you could discuss was a microworld of simulated blocks. "The block which is taller than the one you were holding and put it into the box." In this case, it needs to do a whole set of things, one of which is figure out what is meant by words like "one" and "yet." We use those in normal everyday language in a way which has to be interpreted by looking at the context in which they appear. In this case, it types back out: "By 'it,' I assume you mean a block which is taller than the one I am holding, which is only one of several possible things I could have meant, and needed to use a set of rules of thumb about how people use words like that in order to decide in this case which one I intended." Despite Winograd's achievements, people outside AI circles were not impressed. Some predicted that artificial intelligence was doomed. But AI would not die out. In the dark years ahead, this man, Edward Feigenbaum, would somewhat restore AI's fortune. He realized that while microworlds might not be very large, they might be large enough to be useful. He reasoned that the intelligence displayed by experts, scientists, doctors, specialists, might be as easy to capture, if not easier, than a world of simulated blocks.

Feigenbaum started with the expertise chemists use when reading a mass spectrograph trace like this and interpreting it as a three-dimensional structure like this. Feigenbaum and his colleagues captured the rules in a system called DENDRAL. Now, a computer program was expert in a small area of chemistry. Other so-called expert systems followed, like this one, which simulated how geologists find mineral deposits. All kinds of areas of medicine and science where the knowledge used was deep but very narrow succumbed to Feigenbaum's approach. Paradoxically, these admired specialties were easier for the computer to simulate. There was much less ambiguity, and they were not very large. When you actually codify this knowledge and put it to work, it turns out that you can achieve expert behavior in useful but narrow areas with a few hundred pieces of knowledge, maybe a few thousand pieces of knowledge. Experts are often shocked and startled to find out that in the end, it amounted to a few hundred rules. "Is that all? I learned that what I'm doing every day, I'm really exercising just a few hundred rules. All of me that relates to this task can be canned in a few hundred rules." But a human expert knows many things outside his specialty. The same is not true of an expert system. Outside their field of knowledge, they are hopeless. So you get a blood disease analysis program. It's brilliant at deciding which blood disease the patient has on the basis of a lot of objective tests, like the sugar in the blood, presumably, and white red cells and the white cells and a lot of more complicated things. It will tell you what kind of, say, meningitis that is, and it will tell you with more reliability than your family doctor. Terrific. But if you ask it, "What is a germ?" It doesn't have the slightest idea. Or, "What is a patient?" Or, "Do people prefer to live or die?" Then it breaks down. That's the brittleness.

There was an expert system which had to approve or not approve automobile loans, and it granted a loan to someone who put down that they had 20 years of experience on the same job, even though they also put down that they were only 19 years old. A another situation like that of brittleness that I saw firsthand. System that was done in France for skin disease diagnosis. As a kind of joke, we told it about my 1980 Chevy, and it asked questions like, "Are there spots on the body?" And it said, "Yes." "What color spots?" "Reddish brown." "How old is the patient?" "Ten years old." And it eventually said, "The child has measles."

The brittleness of expert systems has been likened to the fascinating condition of idiot savantism. David is an idiot savant, a human being brilliantly gifted in one area, but backward in every other sense. "What was the 17th of December, 1974?" "The 10th of June, 1917?" "The first of March, 2014?" Jenny calendars that far ahead, but it's again absolutely correct. "Can you tell me what eight add-on seven is?" "Actually, eight age 17, age seven is 14. Attach the candy is not. You don't think it's 15?" When outside this area, he is unable to function as a fully competent member of the world. David can't even add up properly. A deep but narrow mind will always break when it meets new situations. General human intelligence somehow creates a broad model of the world, enabling us to cope with all kinds of situations. To capture this in a computer program, we have to study not experts with their deep and narrow knowledge, but children who excel in knowledge which is broad and shallow.

[Music]

"Notti couldn't stop her giggles." "It's not me, Mom," she said. "It's Ned." Language researchers were hard at work trying to get computers to follow simple stories as children do. They discovered the problem wasn't what the story said, it was the huge number of things that were left unsaid because they were too obvious to be worth saying. Made another pudding. Typical of the kinds of stories which researchers tackled in the early 1970s was this: "Water it was Jack's birthday. Jane and Janet were going to Jack's. Let's give him a kite," said Jane. "No, he already has one," said Janet. "He'll make you take it back." Could the computer understand this story and answer questions on its meaning? Although it seems easy, it presupposes a vast amount of knowledge. We assume they are going to a birthday party, but it doesn't actually say that. You notice that in that story, it's not clear that they're going to a birthday party. It was just that it was Jack and they were going to Jack's. It isn't clear why they were buying a kite. It doesn't say anything about birthday presents, and it doesn't say that you bring birthday presents, and it doesn't say the kite was a birthday present. But they thought they had an answer to that. The answer was supposed to be: build it into a birthday party frame. The thing after microworld, which looked promising, was scripts and frames. Then it, the idea was, there is a stereotypical birthday party, and a child has a birthday, then the child has a party, and then people come to the party, and they bring gifts, and so forth. So even though the story doesn't mention kites, the computer would fail in presents and kites.

Since we judge words by their context, why not give the computer a context by building frames or scripts for the situations it might meet? A birthday party frame would contain all the things which usually happened at birthday parties. Once the computer had identified the frame, it could fill in the missing knowledge. But things were not to be so straightforward. The problem surfaced in a funny way. The idea that he already has one, he'll make you take it back. The challenge was, and I think it's unanswered to this day, here's a new bit of information: the information that if you, in our culture, sadly, children learn if you get a new one just like the old one, you have to take the new one back, not the old one. And the question for them was, where are we going to store that information? It doesn't belong in the birthday-party frame. It doesn't belong in the department store frame. It's general background knowledge.

The horrible thing, general background knowledge, reared its head. And then when I heard it, as I usually did, I saw worse things. It occurred to me that the principal, if you have one, you don't want another one just like it, is not much of a strict rule. It doesn't apply to dollar bills. It probably doesn't apply to marbles or cookies. So the principal has got to be something like, well, everything else being equal, if you got one, you don't want another one just like it. But of course, in the "everything else being equal" is the whole common sense knowledge problem again. Because what is "everything else," and how equal does it have to be? I mean, if it's a cookie, is that big? Maybe one is enough. But then if you're a cookie monster, probably one isn't enough. And all of the knowledge we have about cookies and appetites and Cookie Monsters and so forth comes in to understand just a simple birthday party story. And so that looked bad.

The dream which had started out so well with checkers, chess, and calculus, and which had progressed through microworlds and expert worlds to representing knowledge in scripts and frames, had finally foundered on the common sense knowledge problem. Knowledge that is so intuitive, we are hardly aware of it. Common sense knowledge is the knowledge that everybody shares. Everyone knows that if you hold something and release your grip, it falls. They don't know about gravity, but they know that this is common sense. There's no person that you can communicate with who doesn't know the same things you do about space and time and social relations and geometry and language and whatnot. How large is this database that we all share? I suspect it's about 10 million items or units, whatever units are. The secret of intelligence was common sense, the enormous number of things which we all know to be true, so much so that we are able to communicate effectively without even mentioning them. Could such vast quantities of knowledge possibly be acquired by machines? We learn these millions of things as children growing up. Good machines learn like people. Scientists have long been interested in machine learning, but computers made very bad students because they started out from such a low level. Learning has the property that we learn at the fringe of what we already know. We learn that this new thing is similar to something we know already, and here's the difference. So the more you know, the more and more quickly you can learn. But the trouble is that we start our learning programs off with next to nothing, so they don't very, they don't have a very big fringe, so they can't really learn very much very quickly.

But it wasn't hopeless. There was a way forward, albeit a rather audacious one. Let's believe the only hope was to painstakingly feed the computer the millions of things it needed to know so that it would be able to understand language and learn by itself. In 1984, he set up a project in Texas to do just this. Because it was likened to a vast encyclopedia project, it became known as CYC. The task was, however, not to make an encyclopedia, but to input the kind of knowledge which is not in encyclopedias because it is too obvious to include. For example, CYC would have to know some things an encyclopedia would dream of mentioning about Abraham Lincoln.

[Music]

The CYC project was the ultimate test of AI. If it was possible to build an artificial mind, this apparently was the only way to do it. Before a computer could understand language, before it could learn, it had to be given millions of pieces of common sense. But critics said Lenat's ambitious 10-year project would never work because knowledge alone was not enough. Real common sense, they argued, depended on having something no computer possessed: a human body. Common sense knowledge, to a large extent, doesn't consist of facts. Not a hundred thousand, that's sure, but not a million either. No number of facts. A lot of what we know is not factual knowledge at all. It's skills. Anybody who had children must be struck by the number of years they can spend playing around with blocks, playing around with sand, even just playing around with water, just splashing it around, soaking it up, pouring it, splashing it. Seems endlessly fascinating to children. And one might wonder, what are they doing? Why aren't they getting bored? How does this have any value? Well, I would say their their acquiring the 50,000 water sloshing cases that they need for pouring and drinking and spilling and carrying water. And they've got their 50,000 cases of how solids bump and scrape, stack on, fall off. And common sense knowledge would, in this story that I believe, consists in this huge number of special cases which aren't remembered as a bunch of cases, but which have tuned the neurons so that when something similar to one of these cases comes in, an appropriate action or expectation comes out. And that's, that's what underlies common sense knowledge.

Most people acquire the common-sense knowledge they need to make sense of the world growing up as children, experiencing life in all its richness through the senses. But is this the only way to get it? There are people who never experience much of the world at all, and yet acquire and use language with common sense. In Oliver Sacks' collection of strange neurological cases, "The Man Who Mistook His Wife for a Hat," he wrote of a patient, Madeleine, born blind and unable to move her limbs. Practically everything she knew about the world was told to her or read to her, and yet Sacks records she could use language with common sense. "All my reading has been done for me. I can't read Braille, not a single word. I can't do anything with my hands. They're completely useless." If Madeleine succeeded where computers have so far failed, it wasn't because she experienced what the words meant. The limited experience she did have was processed with an organ far more complex than any computer: her brain.

The pioneers of AI had argued that they didn't need to know the way the brain worked, any more than the first aviators needed to build planes with flapping wings. But some critics had never accepted this analogy, arguing that when it came to thinking, the only way to do it was nature's way, perhaps to build an artificial mind, you first had to build an artificial brain. The human brain is not like a computer. It is made from billions of neurons connected together in thousands of ways. Experiences which come in through the senses trigger electrical signals to pass between the neurons. Patterns and regularities perceived by the senses are therefore remembered as similar patterns of neural discharges. The brain is awesomely complex, but from the early 1950s, some scientists pursued the idea of building an artificial brain to try to imitate how the brain's network of neurons learned. In the brain, experiences are recorded as the strength of the connections between neurons. A new experience coming in through the senses changes the strengths of the connections. A pattern is then subsequently remembered. Other experiences set up other patterns.

In the 1950s and 60s, scientists built a few working perceptrons, as these artificial brains were called. He's using it to explore the mysterious problem of how the brain learns. This perceptron is being trained to recognize the difference between males and females. It is something that all of us can do easily, but few of us can explain how to get a computer to do this would involve working out many complex rules about faces and writing a computer program. But this perceptron was simply given lots and lots of examples, including some with unusual hairstyles. But when it comes to a people, the computer looks at facial features and hair outline and takes longer to learn what it's told by Dr. Taylor. Andrew Cook Chan's wig also causes little pop searching. After training on lots of examples, it's given new faces it has never seen and is able to successfully distinguish male from female. It has learned.

While promising, this approach to machine intelligence virtually died out. But in the late 70s, as AI's problems seemed insurmountable, it underwent a revival. The modern perceptrons are called neural networks, and the people who build them are a growing movement called connectionists. Neural networks model the brain, not the mind. They are small learning machines, and at universities all over the world, connectionists are seeing how much they can get their neural networks to learn. This vehicle is being driven by a neural network. The network hasn't been programmed; it has learned by itself how to keep the vehicle on the road.

[Music]

Earlier, we trained the network to imitate a person driving, having the network watch the person as the person drove along about a 500-meter stretch of road. Now the network has taken over for itself. You can see that down here at the bottom of the image, this square represents the image being fed into the neural network as input. The network has basically learned to key on the position of the of the white line and the edge of the road to determine where the road is and hence in what direction it should steer. Okay, here comes a shadow. This should be interesting. See how it deals. Okay.

[Music]

While neural networks are superficially appealing, the tasks which nets can so far conquer are a far cry from common sense. This net has to be retrained if it rains. They are every bit as small as the microworlds of AI, and they have a problem which conventional software doesn't. Researchers don't has yet know how nets learn. This matters because what the net may be learning may not be what the researchers think it's learning. Early neural networks contained some big surprises. They took a lot of pictures of tanks, more or less hidden behind trees, and trees without any tanks behind them, and they trained a connectionist net to distinguish very clearly between the set of pictures where there were tanks and the set of pictures where there weren't tanks, and it worked wonderfully. And then they went out and if they trained it on those two sets, but they wanted to know, of course, whether it could do new ones that it hadn't seen before. Well, first they gave it further ones that they hadn't shown it from the set of tanks, and it got them, and further ones from the set of trees without tanks, and it got them. But then, just to make sure, they went out and took some more pictures and gave them to the connectionist net, and it failed completely. And then the, the solution turns out to be that all the pictures with tanks in them were taken on a sunny day, and the pictures without tanks were taken on a cloudy day. And what the net had learned to discriminate was cloudy how forests looking cloudy and sunny days. Now, no human being seeing big tanks behind some trees and not behind others would find that the similarity between the scenes was whether it was cloudy or sunny, because the way we're tuned, I mean, the similarities go with whether the tanks are there or not. But nets, any old thing counts as similar. My cat is in love with the prospect. For tiny neural networks to capture something as elaborate as common sense has a very long-term one, as connectionists are the first to admit. The nearest they have come to human speech is to get a network to learn to associate patterns of letters with their sounds when read aloud.

There is a way to end the network was called NETtalk. You're going to hear first what the network sounds like at the very beginning of the training, and it won't sound like words, but it'll sound like attempts. It will get better and better with time. Network takes the letters, say the phrase "grandmother's house," and makes a random attempt at pronouncing it. "Grandmother's house." The phonetic difference between the guess and the right pronunciation is sent back through the network. I grow, imitate, duck. "Grandmother's house." By adjusting the connection strengths after each attempt, the net slowly improves, and finally, after letting it train overnight, the next morning it sounds like this.

[Music]

Understands nothing about the language. It is simply associating letters with sounds. But at least connectionists have made a start, getting neural networks to imitate some of the many things human brains do in recognizing patterns and navigating the world. Today's neural networks are very small, but attempts to make networks bigger than a few hundred neurons backfire. The training time explodes.

[Music]

The brain, with its 10 million neurons, has somehow managed to solve the problem, and the way it has solved it is instructive. It is not one big general-purpose machine, but a collection of many special-purpose machines. In the normal brain, the machines for language, vision, movement, and so on, are beautifully coordinated. Somehow, the brain has managed this task of integration, and from this integration emerges our general-purpose intelligence and common sense. We might have to understand how nature builds and coordinates our many micro machines before we can build an artificial version. It is a task which might take centuries to unravel.

While so far AI is mostly a history of fascinating failures, there is a rich legacy of practical, specialized applications which bear its name. It has produced programs which played chess as well as all but the very best players. It has produced expert systems which can do some of the things people dreamed about in restricted domains, like really rudimentary robots which carry trays around hospitals are starting to appear.

[Music]

There are computers which can read books to the blind. Eggman can be provided by a legally qualified religion and can even translate limited amounts of Japanese into English. But while these practical applications have appropriated the name artificial intelligence, they have little to do with the original quest for a general-purpose intelligence based on common sense. None of them could possibly pass the test Alan Turing had proposed in 1950, and which he expected would be passed by the end of the century. None of them could use language so well they could convince us they were human.

But the quest has not yet been abandoned. A project Doug Lenat began in 1984, the CYC project, is still going strong.

[Music]

And Lenat thinks he has a good chance of success. You can think of this as mankind's first foray into large-scale knowledge engineering. I originally put the chance of this project succeeding when we started at about 10%. I think even that was a little bit optimistic. I now put the chances at about two-thirds, over 60%. And the reason why I'm a lot more optimistic now has to do with the the ways in which we have overcome various tar pits, the potential thorns that might have snagged us along the way. There are all sorts of topics that we've had to deal with: time, space, causality, belief, emotions, rationality. And each one of them has had a whole community of researchers dedicating their life to formalizing, to studying, to codifying. Lenat and his team, who call themselves CYC, are hard at work trying to capture the world piece by piece, trying to build a mind which knows enough so that it can understand language and learn how humans transported. Okay, that's logical. So here's her house, and this is walking to her car. CYC was having some success handling ambiguous phrases. As another example, we could have something like, "Mary read Melville." Well, you can't obviously read a person, but since Melville wrote some novels like Moby Dick, and since author of has high metaphor sensibility, the system knows that often you say an author's name like Shakespeare or Melville when you mean something written by them. And so down here, it says, "Probably what you meant is that Mary Shepard was reading the Moby Dick novel."

It's a profession tied. And nurses are building a mind is painstaking work. What comes to mind when you think of a nurse? --CYC's mind has to be filled with all of these details of what a nurse does: taking temperatures, giving medicines. Now, what about stressful? Stress in a job. Stress and job mean nursing is very stressful. CYC to be able to understand stories about nurses, it must know about pressures on the job, and even the fact that most nurses are female. Even though CYC runs on these computers, it has no body at all.

It is just software, a pure mind. What does an entity with no body make of all the knowledge that it is fared of a world it has not directly experienced? At night, while the rest of the team are asleep in bed, Psyche looks for inconsistencies in its database and comes up with new generalizations. And it's clear that Psyche sees the world in a pretty novel way. So, in the morning, we come in and see what inconsistencies it's found. And we also sometimes see what interesting new discoveries it's made.

This is very interesting. I wonder if this is true. This one says that two countries that have pretty much the same mixture of religions are members of pretty much the same international organizations. So I'm going to put a question mark next to that one. Here's another funny skewing based on what happens to be in the knowledge base. This basically says that most people are prominent, and that's because most of the people that we put in, aside from ourselves working on the system, are famous people from history.

This is actually another funny, another funny sort of misunderstanding where it's been told about Fred while he was shaving in the morning. And it's reached a kind of subtle problem where, on the one hand, it's decided that Fred, while he's holding the razor in his hand, has some electrical parts. But on the other hand, it believes that people don't have, in general, electrical parts. So it's trying to ask something here, like, is Fred still a person while he's shaving?

Listening to Psyche, one gets the feeling not of a developing child, but an alien intelligence who knows a lot that has a bizarre view of the world. But what would Psyche have to achieve to convince skeptics like Dreyfuss it was really intelligent? I would say, if the Psyche project could understand stories of the sort that four-year-old children could understand, I'd be very impressed. I would think it had succeeded in showing that a great deal of intelligence could be captured in a totally disembodied way.

Okay, if it fails, then I think symbolic AI is finished. It's tottering already, because there's only, only that project really that seems to offer any hope. AI is finished for as many hundreds of years as it takes before we understand this mysterious thing that our brain does and how it does it. But if it succeeds, the potential may be breathtaking. The Psyche bet, even though it was a high-risk bet, is a high payoff bet. And if and when it succeeds, the payoff will be in the form of passing the Turing test, of general natural language understanding, of powering machine learning programs to go off and learn some things that are unknown to humanity at the present time. If you look at this AI system that we're describing as a kind of mental amplifier, as an intelligence amplifier, then I think you'll see that using it, you'll be able to do things in a few decades that today people can't dream of doing.

[Music]

Hello, and I'm Geoffrey Mishlove. Our topic today is artificial intelligence, and we'll be looking at the past and at the future of this very exciting and yet somewhat arcane scientific discipline. With me is Dr. John McCarthy, one of the founders of the discipline of artificial intelligence. Dr. McCarthy is one of the co-founders of the first artificial intelligence laboratory at MIT and the founder of the artificial intelligence laboratory at Stanford University. He is the inventor of Lisp, the major computer language used for artificial intelligence, and the oldest surviving computer language dealing with symbolic manipulation. He's also the individual who first conceived of interactive computer time-sharing. He is the developer of non-monotonic reasoning, an important new form of logically conceiving of the difficult problems facing artificial intelligence today. And he is the 1988 recipient of the Kyoto Prize for his lifetime contributions to the field of computer science and artificial intelligence, something of a Japanese equivalent to the Nobel Prize. Welcome, Dr. McCarthy.

Thank you. It's a pleasure to be with you.

Back in 1966, you wrote an article for Scientific American on the field of information, and you projected out at that time what we might see for the next 20 years. And although there were some errors, I suppose, rather accurately described many developments that we now take for granted and which were at that time rather alien to the population at large. So you've witnessed a great deal of the history and the growth of a discipline which has dramatically touched probably everybody's life in the Western world today. And, and you predicted that it would do so. I wonder if we can begin by just having you reflect a little bit on what these past 20-30 years have meant to you personally.

Well, I've gotten older. Yeah. I started my work on artificial intelligence in about '56, although I became really interested before that, in '49, when I was a beginning graduate student in mathematics. I would say that the field has made somewhat less progress than I hoped, although I didn't have any definite opinion as to how fast it would progress. I think that it had, and still has, difficult conceptual problems to solve before we can get computer programs that are as intelligent as humans.

One of the issues that you're working on, and to which most of your life has been devoted, are really tackling these problems, providing the underpinnings so that we can ultimately have formal models of intelligence that would be equivalent to human intelligence.

Well, that's right. And one part of the problem is to develop language in which we can express for our computer programs the facts and reasoning about the common sense world that humans have, and that is necessary in order to behave intelligently. And I have worked on this using the tools of mathematical logic. I think one of the striking things that I find in looking at the history of artificial intelligence is that in the early years, there was some striking progress made on some rather difficult problems, like solving mathematical theorems. And people thought, well, because we could do these difficult things, we ought to therefore have no trouble doing some of the simpler things that human beings can do. And, and yet, just the opposite seems to have been the case, that some of the simpler things that any child can do, like recognize speech, have been the most difficult problems for computer intelligence.

Well, the idea that one could really do difficult mathematical problems, that is, creative mathematical things, was not really realized. That is, it could do some simple kinds of theorem proving and things like that. Now, it's certainly true that dealing with the common sense world has proved to be quite difficult. What it amounts to is that while humans can do this kind of thing very readily because it's built into us, humans have much more difficulty understanding how it is done in order to be able to make computer programs do it.

You're developing formal models that deal with how human beings do the simplest of things.

Well, there are two ways of looking at things. You can either look at it from the point of view of biology, or from the point of view of computer science. From the point of view of biology, you could try to imitate the nervous system, insofar as you understood the nervous system. Or you could try to imitate human psychology, insofar as you understand human psychology. The computer science way of looking at it says that we look at the world and we try to see what problems it presents in order to achieve goals and think about the world, rather than about the biology per se. And I would say that the computer science approach is the one that so far has had the most success. Although these cannot be regarded as alternatives, they're like they're in a race, but they interact with one another, they help one another rather than hinder each other.

Well, in the field of psychology, it used to be during the '50s and early '60s that we thought of the mind as something like a black box. You had a stimulus that went in, and a response that went out. And I think it really wasn't until people in, in your discipline, artificial intelligence and computer science, began looking at how this information is processed, that the psychologists themselves ever felt that they could have a handle on what cognition was all about.

Yes, I think that's right. I think that Newell and Simon, who take a much approach rather close to psychology, were the main contributors to getting psychology to move away from behaviorism. Behaviorism was a reaction to 19th century philosophy, which really was very bad, but it went too far in its efforts to be scientific by saying, well, the only things that were properly subjects for science were the things that were externally observable. But when computers came along, then it became clear that you couldn't do it that way. I remember there, there was an old computer called the IBM 704, and the only stimulus-response rule that it had was that if you press the start button, a little yellow light went on. All the rest, to understand this computer, you had to know what went on inside. And I guess computers have certainly had a profound effect on psychology.

Yeah. As a psychologist myself, I'm very much a student of William James, who back at the turn of the century began writing about consciousness and the stream of consciousness. And I'm aware that for 50, 60 years, his work in that area was, was pretty much ignored until people in the field of computer sciences began to say that we can, we can have a handle on what consciousness means.

Well, there are many kinds of consciousness. In some respects, computers are easily more self-conscious than human beings. It's not hard to make a computer program look at its own program. But all that people have managed to do with it is to check summat to see that it hasn't been damaged so far. You know, what's involved, yeah, in the kinds of consciousness that people would like to program is regarding the self as an object in the world and considering to be able to think about what progress it's making towards achieving its goals and so forth. And this offers some conceptual difficulties, and we certainly wouldn't say that the problem of giving computers self-consciousness is very close to being fully solved.

Well, I resist some very deep philosophical issues. I'm aware of the disputes that took place during the 1940s with Alan Turing, who was one of the founders of the field of computer science, and in which he, he developed the famous Turing test, which suggested that if, if a computer could imitate a human being to such an extent that you couldn't, if you were sitting at a teletype, know whether you were communicating with a computer or with a real human, then, then you might as well say that the computer were, in fact, conscious.

Turing responded, as I understand the argument, by pointing out the, the old philosophical conundrum of solipsistic. Anshe, let alone a computer. I don't remember Turing discussing solipsism, but he did sort of use that as a kind of test, as for philosophers. In other words, if you wouldn't admit that something that you couldn't tell from a human was thinking, then maybe there wasn't much more to say. Now, in fact, up to this very day, some of the philosophers are willing to accept behavioral criteria, and others are not. They can say, well, it could pretend to be a human, but it wouldn't really be thinking because it would only be doing what it was programmed to do.

One, one development, and I must confess, it troubles me a little bit, that has come out of the, the information processing models of the mind that are now current, is that we view consciousness as consisting of many components. You have memory, you have emotions, you have different kinds of attention. And sometimes people say, well, consciousness is nothing more than, than the sum of its parts, so to speak, just like a machine might be. I wonder how you respond to that.

Well, a machine isn't the sum of its parts. If you, somebody took a car apart and gave you a heap of the parts, that wouldn't be a car. It has, they have to be connected in a specified way and interacting in the specified way. And so, if you want to say that the mind is a structure composed of parts interacting in a specialized, in a separate, a specialized way, I would agree with that. But it isn't just a heap of them, it's more of a system.

It's that's right.

Mm-hmm. But, well, then we get into the, the issue, and I know many people get offended when they think that you could even describe the human being as being equivalent to a system. Like, people say, you know, we have something more. We have intuition, we have spirituality, we have something that transcends the, the mechanistic aspects of our being.

Well, that view has been in retreat for several hundred years as more and more was discovered about human physiology and psychology. And I suppose, well, maybe one could use the boxing metaphor, it can run, but it can't hide.

Can you elaborate on that? I'm not quite sure what you're getting at there.

Well, there are these aspects of human consciousness that have not been realized in machines, in computer programs. And there are some difficult problems for their realization. But we optimists about AI expect to get to them.

Mm-hmm. You, you know, there's an interesting story about, you're a chess player. And back in 1968, you made a wager with a fellow who was then the Scottish chess master that in ten years, a computer would be able to beat him. And ten years later, you got together with a state-of-the-art program back in '78, and the computer nearly beat him. That's right. He won two games to the machine's one.

Mm-hmm. Now, at that time, David Levy was a graduate student in computer science. And my intention was not merely to bet with him, but to hire him to work on the chess program. However, he decided he'd really like to publish a magazine on chess rather than continue as a graduate student in computer science. I didn't consider the bet by any means a sure thing, but it came close in '78. And now computer programs, this year, I believe, or maybe it was last year, a computer program won its first game against a grandmaster. And since David Levy never made it to be grandmaster, probably the current programs could beat him. Although, in my opinion, they used too much brute force and if you're calculating power.

That's right. I would like to see contests that are more like the one design sailboat contests, where it's the cleverness that's involved, rather than who can build a monster special-purpose machine.

Now, this is very important, because your work seems to be saying, you know, that in natural human life, we use a lot of mental shortcuts. We don't solve problems by using brute intellectual force. We somehow have rules of thumb that guide us. And you're attempting to develop formalize logic that would enable machines to be able to sort of work in that fashion.

Yes, that's right. And indeed, the collection of problems on which computer brute force can be applied is rather limited. Most of the problems of common-sense reasoning are problems where there really isn't that much opportunity to apply brute force, or at least nobody's really figured out how to do so. And the, I would say the central problem of artificial intelligence involves how to express the knowledge about the world that is necessary for intelligent behavior. And I've pursued mathematical logic as the tool. This has had its ups and downs in popularity. Now is definitely an up period. It's quite popular. And part of the reason for that is that in the late 1970s, several people independently, myself among them, discovered ways of formalizing what we call non-monotonic reasoning, which greatly extended the power of mathematical logic in the common sense area.

Now, I know many of our viewers are going to have with a term like non-monotonic reasoning, and yet it may be crucial to our understanding of some of the developments that await us in the future. So could you expand on that?

Okay. So you have to say what it is. None, ordinary logic has the property that if you can draw a certain conclusion from some premises, then if you add more premises, you can still draw that conclusion. So the set of conclusions that you can draw only increase when you increase the set of premises, they don't decrease.

Now, human reasoning, and what we will have to make computers do, doesn't always have that property.

That's what you would call monotonic.

That's right.

Could you give an example?

You know, suppose we have, I tell you that I have a bird that I want you to build me a bird cage for. Then, and that's all I tell you. Then you would draw the conclusion that my bird can fly, and that you'd better put a top on the bird cage. On the other hand, if you learn the additional fact that my bird is a penguin, then you would feel that you do not need to put a top on it. So the conclusion that the bird cage required a top depends non-monotonically on the facts that I tell you.

In other words, this is an example of non-monotonic logic, and it has sort of built-in assumptions that I work with. It is when you use the word bird, I assume it can fly.

That's right. That is the kind of, and that's the sort of convention of English or of other natural languages. If I hire you to build me a bird cage, and you build it without a top, and I refuse to pay, and you tell the judge, he never said his bird can fly, the judge will side with me. On the other hand, if you did build it with a top, and I say, well, my bird is a penguin, he wasted material, the judge will side with you, because it's a convention of English that if a bird can fly, it doesn't have to be mentioned, even if it's important. Whereas a bird, if a bird cannot fly, then it must be mentioned if it's, if it's relevant.

Mm-hmm. It reminds me of when I was a child, my father would sometimes say things to me. I would ask him a question, and he would say, "If you have to ask, if you don't already know, it won't help to tell you." It's that we just, we operate in a world of all sorts of implicit, built-in assumptions and built-in understandings of context everywhere we go. And that is what you mean, I guess, by non-monotonic.

Well, that's non-monotonicity is only part of the context problem. Now that we can do formalize some non-monotonic reasoning, we see that, well, there's would deal more to context than that. And so to speak, the next mountain that has to be surmounted.

Mm-hmm. So you are attempting to use non-monotonic reasoning as a mathematical tool for, for building into computers an awareness of that when I use a term like bird, I don't have to specify all of its qualities, which ones they have, which ones they don't have. And in a straight, I would say linear, or you would say monotonic, specific form.

That's right. Suppose you want to be able to reason about birds flying. And then you might say, well, I'll put in the exceptions, ostriches and penguins and so forth. And then someone comes along and says, well, what about a bird with its feet encased in cement? And then you can see that, well, you couldn't possibly put in all the exceptions, because if you put that money in, I'll invent another exotic exception that you, you wouldn't put in. So what you have to do instead is go to a system where you will assume that the bird can fly unless you have some evidence to the contrary.

Mm-hmm. Now, this seems, I have to tell you, it seems very simple to me. Yet you're describing this as is somehow something new in the world of computers.

Well, that's right. And what I believe is that if it takes 200 years to achieve artificial intelligence, and then finally there is a textbook that explains how it's done, the hardest part of that textbook will to write will be the part that explains why people didn't think of it 200 years ago. Because we are really talking about how to make machines do things that are really on the surface of our minds. It's just that our ability to observe our own mental processes is, is not very good, and has not been very good. And we can look at that historically. When we look at Leibniz, who was an extremely smart scientist, he was the co-inventor of calculus with Isaac Newton. He wanted to make a logical calculus that would permit calculation instead of argument. And he invented binary numbers in this case, but he didn't even invent propositional calculus. That was invented by Boole 150 years later. And then Boole didn't invent predicate calculus. So what one sees is that each step in understanding of thought processes has taken time.

Mm-hmm. In fact, what you're saying reminds me very much of the story of Socrates, who went around questioning people in all of the different professions in Greece 2,500 years ago. While these people were quite competent at what they did, he said, well, they're all ignorant, none of them can tell me how they do what they do when he questioned them closely. And you seem to be engaged in a process very much like, you know, the second step beyond what he was doing.

Yes. Well, Socrates was, as I understand it, mainly interested in demonstrating people's ignorance. But now we are really trying to say, well, how can we make computers actually carry out these processes? So it's a whole different program.

That's right. You know, in a sense, what is your sense of the, the likely future? You, you described twenty years ago how with computer time-sharing, we would all have access to information utilities, and, and that has come to pass. And I, I don't want to pin you down to a specific date, because I realize it, you couldn't say honestly, but what are the sorts of things that are achievable?

As I say, I think their conceptual breakthroughs that have to be made. Mm-hmm. And the extreme is that, well, some smart young fellow has just done it, he just hasn't told us yet. And the other extreme is that it may take a couple hundred years, maybe five hundred even, depending on how many conceptual problems there are, that it might take five hundred years before we have computer programs that are as intelligent as human beings. Now, I'd really be inclined to bet on something like 50, although very exceedingly unlikely that I'll be around. But I, I don't, I simply don't know how long it will take.

No, but it sounds like you're making a firm bet against the critics of artificial intelligence who say that in theory, it's philosophically impossible to, that's right, replicate human intelligence.

I see their arguments as faulty. And I don't see that human intelligence is something that humans can never understand. So ultimately, the project that you and your colleagues in the field of artificial intelligence are engaged in, one, one might view it as is the most noble project of all, the one that Socrates actually urged people into, which is to know thyself, and sometimes against great odds, to, to attempt to do the, what may be in effect, one of the most difficult tasks facing humankind.

Well, it's certainly a difficult task. John McCarthy, we're just about out of time right now. I want to thank you very much for coming here and sharing yourself with us this evening.

It's been a pleasure being with you.

Well, thank you for inviting me. Thank you. And thank you for being with us.

[Music]