Help us improve this podcast: take our short survey.
Marvin Minsky is often regarded as the father of modern AI, but when Miles visited him in 2010, Minsky wasn’t a proud father. In fact, Minsky was disappointed with the lack of progress in the field and had reservations about its future. Unfortunately, Minsky is no longer with us and can’t answer our questions about machine learning and new robots, but his answers from this interview in many ways still hold. Go back in time on this episode of Miles To Go.
Miles O’Brien: Hello and welcome to another edition of Miles To Go, I’m Miles O’Brien.
Last week, I did a piece for the PBS NewsHour. The subject: the Future of Work. In it, we discussed the concept of cobots–that is to say, collaborative robots–and how these kinds of robots, along with artificial intelligence, are being slowly deployed in various industries. The idea is that robots and humans together can do a lot more than they could separately. That’s the theory, anyhow.
I always think the question comes down to this: where’s my robot butler? I thought I was supposed to have one by now! Along with my flying car, of course. I truly am a product of the Jetsons generation and clearly spent too much time watching TV!
Matter of fact, as I was putting this piece together, I was reminded of my very first PBS NewsHour piece. It was on the subject of robots. It aired back in 2010, kids–remember those days?–and it was called “Will a Small Step for Robots Lead to a Giant Leap for Robotkind?”.
That story gave me the opportunity to sit down with the legendary and unfortunately now late Marvin Minsky, the father in the field of artificial intelligence. Unfortunately, Minsky passed away in 2016, so I couldn’t go ask him again what he thought about the progress in robotics and AI over the past 8, almost 9 years, but I imagine his answer would be similar to the question I posed to him, which is: Where is my robot butler?
Let’s have a flashback in time and listen to Dr. Minsky.
Miles O’Brien: Okay, so why, why — I am tired of waiting for the answer to this question, much less the robot, where is the robot?
Marvin Minsky: When I was a child, there was the beginnings of a field called science fiction, and one of the first things I read had been written, I think, around 1920 by Hugo Gernsback, who was a bit of an inventor and he had a book about an intelligent robot. And that seemed fine, and it tells you robot was going to happen sometime in the future.
And then of course there was Jules Verne from the previous century who had all sorts of wonderful machines and ways to get to the Moon and all that sort of thing.
And H. G. Wells, who wrote many science fiction stories about wonders that would surely happen in the next 100 years.
And then, when I was getting through grade school, I ran out of those, but luckily, around 1940 or so, another star appeared in the literary sky called Isaac Asimov and he was writing books about intelligent robots, which were actually smarter than the people who were using them. That seemed very thrilling, and clearly that was certain to happen in the next 50 years or so. So one could just wait for them.
Miles O’Brien: We’re still waiting.
Marvin Minsky: We’re still waiting. Isaac became a friend of ours. And I started to work on robots at some point when I finished high school and went to college.
In the early 1960s, we started what was called an artificial intelligence project right here at MIT, and in the first ten years wonderful things happened. One of the first programs could do a lot of calculus and it got an A on the more symbolic part of the MIT first year calculus course.
Then a couple of years later, a student named Dan Bobrow wrote a program that could solve word problems in high school algebra–you know, a car took three hours to get from here to there, and it used up so many gallons. About half the time if you typed a simplified version of that problem in English, this program could get the answer. So it was like a 10th or 11th grade high school student.
And consistently through the 1960s, many programs were written that did some of the things that a fairly advanced high school or college student could do; analogy, tests and so forth, and so artificial intelligence seemed just around the corner.
But the interesting thing is that it was easy to do the things that people considered hard, like calculus, but making a program that could figure out simple arithmetic for commonsense situations was almost impossible. It was only for carefully formulated kinds of questions you would put on in the exam that the programs could behave smartly, but for commonsense there was nothing.
As far as I know today, no computer understands that you can pull something with a string but you can’t push it.
Miles O’Brien: The interesting thing you pointed out is that the things that humans find hard, robots find easy; the things that humans find easy, robots find hard. Why is that?
Marvin Minsky: It’s really wonderful, but I think it’s that Herbert Simon and some of his students once tried to estimate how much knowledge a human expert has, and they sort of studied informally a number of fields. They concluded that if somebody is very good at some skill, it’s because they know about 20,000 fragments of knowledge or process or whatever.
So the sort of magic number of 10,000 or 20,000 came up. That’s like the number of words that a person knows and so forth. You can only learn a few things a minute of this sort, at best.
So advanced skill seems to be rather small in the number of things you know. A mathematician maybe knows 400 or 500 ways to solve certain kinds of problems but doesn’t know 100,000 ways.
However, it looks like if you take the knowledge you need to get through the day in the world of physical objects and things that happen and so forth, that maybe you have to know a few million little, tiny facts about or little representations of phenomena or little stories about how things happen and how things go wrong.
And so ordinary commonsense knowledge looks like it involves maybe a few million fragments of skill or knowledge, whereas being a world-class expert is just a few tens of thousands of things.
Miles O’Brien: Oh, the irony there!
Marvin Minsky: There is an irony. So children are smarter than adults, whatever, anyway you want to put it. And the field of artificial intelligence made a lot of progress in expertise in the first 20 years, from 1950 to 1970 or 1980, but in the next 30 years, from 1980 to now, they have only been slowly creeping up on the commonsense sorts of knowledge that a four or five-year-old knows. So that there’s still no machine that can solve everyday commonsensical problems.
Miles O’Brien: So it was a hockey stick that hit a plateau.
Marvin Minsky: I see it as sort of backwards evolution, that it was easy to do mathematical logic, which was the first beautiful computer programs, duplicated the discoveries that Bertrand Russell and Alfred Whitehead had made in 1900, which made logic into a complicated mathematical subject.
And in the late 1950s, some friends of ours, Herbert Simon and Allen Newell, wrote programs that could do that. And they got a letter of congratulations from Bertrand Russell, in fact.
So this advanced mathematics came easily, and then the high school type mathematics was a little later, and we are still not at the age of the four or five-year-old.
I have a little granddaughter; well, she is older now, but when she was about five, her mother said, you should talk to Charlotte because she is getting good at this.
So I said, how much is 15 and 15? She said, I think it’s 30. So I said, how did you do that so fast? And she said, well, everyone knows that 16 and 16 is 32. Now, some listeners might not know about that, because they don’t teach that 16, but the most important thing in my childhood and many of my friends was saying that 2 times 2 is 4 and 2 times 4 is 8 and 2 times 8 is 16, 32, 64, 128, 256, 512. Every smart mathematician knows that. Almost no ordinary citizen knows that.
And that simple kind of thing of not worrying about precision, but being estimating, what happens if you double something 20 times? Oh, it’s about a million. Every mathematician knows that, no citizen knows that.
Miles O’Brien: Yeah. Well, so why then this plateau? What happened? Was it… Did the smart people like you run out of good ideas? Did we run out of space on silicon wafers, or are we just grinding through a bigger task than we anticipated?
Marvin Minsky: I think the disaster was very simple. It was clear in 1980 that to make machines more smart, they would have to have a lot of commonsense knowledge. They would have to know a few million things about the world in order to get by solving everyday problems.
And one of our friends, named Douglas Lenat, had done some expert programs and he decided to attack that. So around 1985 he started a project to make a little encyclopedia of commonsense knowledge. And a lot of people said, that’s a good idea. Let’s wait and see how well he does, and that was why the field of artificial intelligence stopped in some sense.
Everybody–there is one project doing commonsense and everybody waiting to see, ten years went by, it made a little progress. You would think if something was important, there would be 50 people competing.
Miles O’Brien: Genome, the Genome project kind of thing, right?
Marvin Minsky: Right. That was fast.
Miles O’Brien: So it was that simple fact that one guy went off to do the commonsense and they said, okay, we’ll wait till that’s done and then we will be okay?
Marvin Minsky: What happened was, one guy went off to work on commonsense, the others said, let’s see what happens, but that’s a lot of work. Let’s find a shortcut. So some other people said, let’s make a system that will learn this stuff, we won’t just program it. Yet other people said, let’s do it mathematically. Let’s collect statistics on all the things that happen in the world.
There are 50,000 people, to this day, writing programs to collect knowledge statistically. It’s the most popular thing. I think it’s useful for many applications, but intellectually it’s a sort of dead end, because it doesn’t find out why these things happened, it just says, there are correlations.
But anyway, this must happen in many fields. Some theory comes up and people admire it and they all work on that, and so you have these plateaus. There was a book by Thomas Kuhn about how science develops, and his view was that some idea sweeps through and everybody works on that for a whole generation, and then things change.
So at least in my field of artificial intelligence, there hasn’t been much steady progress. There has been, oh, here is something that really works, everybody works on that, maybe self-organizing systems. And then somebody says, oh, look, I got this to work, and everybody flocks over and copies that. And so we haven’t had a lot of independent people developing different ideas and putting them together.
Miles O’Brien: Well, as the child that read Wells and the young adult who became enthralled with Asimov, you must be a little disappointed?
Marvin Minsky: I am a little disappointed that most people look for the magic bullet: what’s the trick that will make machines more intelligent? And it seems to me that we know from brain science, if you look at the brain, it’s like 40 different computers. In fact, if you look in a big book on neuroscience, you will find maybe 300 or 400 descriptions of different parts of the brain that do different things.
So I think this isn’t a case of finding the Unified Theory like Einstein, Maxwell is a good example. All of electricity and magnetism was described by four laws, or Newton before him. Mechanics was very complicate, and Newton discovered three simple laws that explained how the planets and everything like that worked. And a little later, somebody discovered a couple of laws that explains how heat and cold work.
And so I think in the field of psychology, most people are looking for the magic bullet. What are the five laws of intelligence? And I think the answer is: that’s a waste of time. The brain has evolved maybe 400 or 500 ways of doing things, and the interesting question is, what are they and how do they work together? But most of my friends in this field just say, I think this is the important one. So it’s just a matter of trying to find a few simple laws or trying to find how to put together a lot of not very good laws.
Miles O’Brien: We sort of need your researchers to act like a neural net a little more.
Marvin Minsky: Themselves.
Miles O’Brien: Themselves. They are not, are they? It’s like herding cats, isn’t it?
Marvin Minsky: Well, the scientific community does… I think physics was so wonderfully successful that everybody is jealous of it. Newton made these three laws: something that’s moving, keeps moving, energy is conserved. Oh, it was so simple, but it was thousands of years before that, that people just collected all this data and they would predict what happens, and they predict pretty well and things got–I don’t expect for psychology that that will ever happen, because evolution doesn’t, isn’t based on a single principle, it’s based on hundreds of tricks. When a bunch of animals happen to find a new trick, get a few new genes, then they can move and survive in another environment, and it’s a big mess.
Miles O’Brien: Okay. So tell me then, is it possible, after all these years, are you willing to surrender? Is it possible that we can’t make a robot that really does all we want it to do?
Marvin Minsky: I think we will be able to make robots intelligent and do all the sorts of things that people do and more, at the point when we get a community of people who know how to work together to develop different methods to solve different kinds of problems and put them together. Right now we are sort of stuck, because each person is trying to solve the whole problem by themselves by finding one wonderful trick that does everything.
I don’t know when this will happen, and I am a little worried that it won’t happen here. There is much less basic research in the United States now than there was in 1950 and 1960.
Miles O’Brien: You can say that again. So where is it, is it going to happen in Japan or China or somewhere else?
Marvin Minsky: It’s hard to predict where the center of gravity of science will move. Today when a student graduates from college, it’s hard to find any job at all, at least in the United States, and quite a bit of the world, because of the current economic hurricane.
When I started to work in research in the 1950s and 1960s, this was after World War II and there was lots of interest in basic research. I think the war was won by things like radar and the Manhattan project and the bomb and so forth, and everybody knew that you could change the world with a little bit of science. Now they think you can change the world by changing the entry for getting a mortgage.
In my early years, I went to Bell Labs for a summer. I had a wonderful job, because there was a great scientist named Claude Shannon and he sort of adopted me. And when I got to Bell Telephone Laboratories, someone said, by the way, you shouldn’t work on anything that will take less than 30 years.
That’s what the research people were doing. They sort of planned to work on things that were very hard and would take a long time to understand. And that was the goal, find something really hard and work on it.
Now the goal is, find something you can do in 15 months, so you can get your second wave of investor capital. There is no job a student can get today that says, would you like to do research for five years? It’s gone.
So the United States is in deep trouble and I don’t know what’s going to happen, but I am sure it will suddenly change.
Now there are a lot of billionaires, but it seems to me most of the billionaires are wasting their money by doing the things they wish governments would do. There should be more ways to give people the way to do research for five or ten years, and think about a hard problem.
Miles O’Brien: Why do we collectively want humanoid robots, what is it about that that is intriguing, do you think, to human beings?
Marvin Minsky: Well, I am not sure most people like the idea of robots, but there are several reasons why it’s important to get them. And two reasons are: we don’t have ways to take care of young people who are in trouble, and we don’t have ways to take care of old people.
When I was a child, I would run into some person who knew a lot and spend a lot of time with them. I went to a wonderful private school, and if a student had a hard problem, somebody would come and spend hours with me, and we would work on something, and there aren’t enough teachers to do that. A class of 20 is too big.
So I think for children to develop, we need friends who know more than we do, and there aren’t enough of them. And same for old people who are becoming handicapped and there’s no one to take care of them. And, as far as I can tell, if you look at what’s happened in the last 50 years, the life expectancy in developed countries has gone from 40 years to 80 years, and I expect that with all of the progress in medicine going on, people will soon live 120 years. Only two people have done that in recorded history so far. And then they’ll live 150 years, and we’ll need robots to take care of them, because they won’t… Oh, there are lots of reasons.
Miles O’Brien: But wait a minute, shouldn’t human beings be taking care of our kids and our old people? Do you want to be taken care of by a robot?
Marvin Minsky: Well, if we had two children for every adult, the planet would break. We are going to have 9 or 10 billion people soon and there won’t be any place for anyone to live. So we can’t get more people, but we could have little robots that do big things.
Miles O’Brien: So that vision is what drives you?
Marvin Minsky: Oh, I don’t know. I think I am trying to understand intelligence and have it become smarter is what really interests me, and it’s always nice to think of an excuse which will get other people to support than research.
Miles O’Brien: Making machines smart in a real way, smart in a human way, is a very challenging thing. It’s like the holy grail, isn’t it?
Marvin Minsky: It’s interesting how little happened in psychology from the — if you read Aristotle. Aristotle, like there is a book called ‘Rhetoric’, it’s about thinking. It’s full of theories about thinking, and Aristotle’s theories of how thinking works are pretty much the same as the technical theories of the 1930s. So 2,000 years sort of disappeared.
The first great child psychologist in history is Jean Piaget in the 1930s. So there’s a strange thing: why was psychology so advanced and why did it never make progress? And it seems to me that that’s the interesting thing about people, that they can think and some people think better than others, and wouldn’t it be wonderful if everybody could think better, and we should have more research on that.
Miles O’Brien: So do you think the plateau will go away? Do you anticipate a spur that will get us to this goal anytime soon?
Marvin Minsky: I think probably most countries will peak at one science or another and then deteriorate. History is full of these inexplicable waves. But maybe in Europe, some group will develop a new grade psychology institute and do more of this work. The United States seems to be sort of stuck right now, but we will see what happens.
As I mentioned, there was a great psychologist named Jean Piaget who started to do experiments with children in around 1930, and by 1950 he had written about ten books on how children think. There had been nothing like that in the last 2,000 years. Before Piaget, as far as I can see, most people thought of children are just ignorant adults. Piaget had the idea that they didn’t have certain ways to think till certain ages.
People argue about, do all children go through these stages, or how necessary are they. But it seems to me that psychology just suddenly developed in the 20th century. In the 1950s a new kind of psychology developed called cognitive science, which was more based on theories of how machines work. And that’s still growing and it’s a fairly subtle science.
Miles O’Brien: So the field of robotics right now is not progressing rapidly in your opinion?
Marvin Minsky: Yes, the field of robotics is such that robots can still barely walk. People are focusing on little things like how to make a robot grasp things better and so forth, but very few people are working on, how can a robot discover new ways to do things?
For example, instead of teaching a robot to walk, we should be teaching robots to learn how to keep from falling. I see it as almost the opposite of what people are doing, that if I hurt my knee, I can still walk, but I do what’s called limping. I don’t know of any robot today, if you interfere with how it works, can figure out another way to do it. So we should be working on the higher levels of robotic thinking and have it figure out the lower ones.
But what most people today are doing is saying, first let’s get the robot so that it can do the simple things and then we will make it do the harder ones. I think we should just turn it opposite.
Miles O’Brien: So we are doing it backwards?
Marvin Minsky: I think we are doing it backwards, yes.
Miles O’Brien: Does the world need robots?
Marvin Minsky: I think the world is going to desperately need robots, because eventually people are going to want to live 200 and 300 and 500 years. So that’s a great thing. I think it’s terrible to spend 50 years just barely learning a profession and then you have to die. It’s really quite wasteful. Some people say that’s good, because then we have room for new ideas. But no reason you can’t do both.
However, when people start to live for 300 years, then we can’t allow them really to have more than two children under the age of 150. Population has to stop. So we do have to make robotic assistance that can do all the things that young workers do today. It’s the only way to live an endless, luxurious, intelligent life.
Miles O’Brien: And of course if you take that a little farther, a little further, and I know that Kurzweil is somebody who thinks highly of you, suddenly we all merge into, we become the robot in a sense. Is that what you see?
Marvin Minsky: Well, Arthur Clarke, who is a great science fiction writer that I loved, died recently, wrote a book about what would happen to our planet where people could live as long as they want, because their brain has been made into digital memory and you can make copies and edit it.
And so, in his picture, there would be billions of people, but there would only be a few million at a time using up space on the Earth. So someone would come out of cold storage. They would live for a million years doing, learning, and so forth. Then they would edit their memories and decide what to keep and go back into cold storage. And so this was a novel about eternal life where people could keep growing and developing. They wouldn’t get stuck.
Of course, it didn’t work out very well and finally he had to introduce a young person who upset the whole thing and started it over. But anyway —
Miles O’Brien: Good story!
Marvin Minsky: It’s a good story, and we should have goals like that and figure out, how can we make new ways of living and last longer and learn more and don’t get stuck.
Miles O’Brien: We were at Cynthia Breazeal’s laboratory today and she is involved, as you well know, in giving robots the emotive cues that a human being might have; expressions and so forth, in order to connect with human beings in more meaningful ways for human beings.
I also spoke to Sherry Turkle today and she has some real strong misgivings about this whole notion, this sense that as humans invest more and more in relationships with machines, in a sense we are collectively deceiving ourselves. Do you go along with that?
Marvin Minsky: There are endless possibilities for what humans’ minds could do. We seem sort of stuck, in the sense that I don’t find people writing plays that are much better than Shakespeare these days. Now, that’s 400 years or whatever, so don’t you think something is wrong, or Aeschylus or Euripides? The scientists today are pretty good, but they don’t seem a lot better than Newton or Gauss or Leibniz. So people who say, let’s not upset the applecart, let’s leave things as they are, I think they are not greedy enough.
Miles O’Brien: Really? What do you mean by that?
Marvin Minsky: They say, well, I know what a good life is, let’s just let everybody lead what I think is a good life. So the people who are against progress are actually against everything, in my view.
Now, we don’t have to have everybody do the same thing, but people who say, let’s not make intelligent machines or let’s not make whatever, whatever, whatever, you can’t stay in the same place. The universe is 16 billion, I don’t know what the current age of the universe is, but the Earth is only going to be here another two or three billion years, because we know the sun is going to burn us up. So we really–it’s nice to say, let’s keep doing the good life, but there’s really an emergency out there and we better find some way to provide ourselves a real future.
Miles O’Brien: So in building robots that are smart, that they have relationship with us, we don’t in any way dehumanize ourselves, in fact, you would say to the contrary, it actually takes human beings to a higher level ultimately?
Marvin Minsky: Imagine the chimpanzee is saying, I heard that if we change some genes, we would become much smarter. For heaven’s sake, that would de-chimpansize us. What a terrible idea!
Miles O’Brien: De-chimpansize.
Marvin Minsky: De-humanizing is the most important thing we could do.
Miles O’Brien: Anyway, so, when is it going to happen? When am I going to have my robot butler? I want him!
Marvin Minsky: Well, if Kurzweil has his way, it could happen by 2030, but I don’t think it will. However, I think if we had a lot of money and tenured fellowships, then in 30 or 40 years we could have very intelligent machines.
Miles O’Brien: It’s still 30 or 40 years away?
Marvin Minsky: Well, 30 or 40 years is three generations of graduate students. It’s pretty short. I have been through seven of them, and the last three, I think, were too slow.
Miles O’Brien: But I bet you would have predicted by now you would have your robot butler.
Marvin Minsky: I predicted in the 1950s that by 2000, we would have 2001, as Stanley Kubrick put it. I was surprised that, to me, researching artificial intelligence almost stopped around 1980, because people moved from, this sounds technical, but in the 1970s, people were working on semantics in language, on the meanings of words. Some other people were working on the grammar of language, which I think is really trivial, it’s not very important, but representing the meanings is important.
But what happened in the 1980s, among my friends in artificial intelligence, is that people started to try to make machines that used statistics to make predictions rather than to try to understand what’s really happening and the causes and nature of things. So a lot of the research I liked just disappeared and got replaced by a different kind, and I hope there’s a reversal soon.
Miles O’Brien: Because people weren’t as motivated to build these robots as you, is that it, not enough motivation there?
Marvin Minsky: I don’t think it was so much motivation as commercial pressure, because there were fewer opportunities for somebody to look to a career in research. So most young people started companies to make enough money, and when they got enough money, then they would be able to do the research. That was the dream of most of my friends who got out of the university, they started companies, they made money, and their plan was, when they got enough support then they would do research. But they never did. They kept making more money.
Miles O’Brien: So all the potential robot makers out there are making iPads or whatever.
Marvin Minsky: I love iPads, but —
Miles O’Brien: We all do, but the point is, the talent is not there.
Marvin Minsky: I know almost no individuals in the United States who are working on advanced ideas in artificial intelligence in robots. It’s all gone.
Miles O’Brien: Well we’re sitting in a place right now where people are working on it, right?
Marvin Minsky: No, they think so, but they are making the robots appear to be human, but they are not making them smart.
Miles O’Brien: There’s a distinction there, isn’t there?
Marvin Minsky: There is a lot of interest in social facilitation, but —
Miles O’Brien: But they are no smarter than they were?
Marvin Minsky: No, not very much.
Miles O’Brien: It was really a fascinating talk, as you can tell, and I consider myself very fortunate to have had an opportunity to meet and talk with him.
I wish he were still around–I would have liked to ask him what he made of the recent advances in machine learning. Considering he wasn’t too happy with using statistics for artificial intelligence, he probably wouldn’t like the black box approach of neural nets nowadays.
I don’t know… even with all the recent advances, that robot butler still looks like it’s a few years away, don’t you think?
Before you go, a quick favor. I would appreciate it if you would review this podcast, it would just help us out–yea or nah, any kind of reaction is good–just don’t have your robot fill out your form. We want your opinion–not your robot’s opinion.
And, while you’re at it, would you please sign up for my newsletter? I promise we don’t have robots here working in cobot fashion to spam you. Just one email a week is what we’re talking about and it will keep you up to date on the world of science and technology and all things that interest us here at Miles O’Brien Productions.
Thanks for your time, thanks for listening! I’m Miles O’Brien and this is Miles To Go.
Banner image credit: Alex Knight | Unsplash.