The Emotion Machine
Posted by Jeriaska on August 24th, 2007
Marvin Minsky, Philip Rosedale, and Ben Goertzel at Transvision 2007
Marvin Minsky is Toshiba Professor of Media Arts and Sciences, and Professor of Electrical Engineering and Computer Science, at the Massachusetts Institute of Technology. His research has led to both theoretical and practical advances in artificial intelligence, cognitive psychology, neural networks, and the theory of Turing Machines and recursive functions. (In 1961 he solved Emil Post’s problem of “Tag”, and showed that any computer can be simulated by a machine with only two registers and two simple instructions.) He has made other contributions in the domains of graphics, symbolic mathematical computation, knowledge representation, computational semantics, machine perception, and both symbolic and connectionist learning. He has also been involved with advanced technologies for exploring space. At Transvision 2007 he gave a talk on some of the ideas put forth in his new book The Emotion Machine.
The following transcript of Marvin Minsky’s 2007 Transvision presentation has been approved by the author.
The Emotion Machine
Thanks. It’s a pleasure to be here among so many imaginative people. I think what I want to do is mostly talk about what is in my new book that evolved over twenty years from The Society of Mind. So, there are lots of problems we have to solve. And one of the problems is that on the evolutionary scale, human beings are pretty smart and they live longer than most other animals, but we still do not live long enough. We know so much as a society that it takes many years for an individual to come up to speed. And then they get sick and die. I think it is very important to pursue the directions that Aubrey and others are taking toward extending life, which has been talked about in two ways: either by repairing damage, maintaining our biological existence. The one I will be talking about, at least indirectly, is we really have to get out of these bodies, because they are not designed very well.
I love the expression “intelligent design.” In a wonderful chapter in Mark Twain’s Letters from the Earth someone asks, “How do you pray?” And Mark Twain’s character says, “Well, we get on our knees and we say, ‘We thank thee, oh mighty God, for syphilis, gonorrhea…’” It goes on for a page and a half. Maybe I’m confusing that with James Joyce, A Portrait of the Artist as a Young Man, which also has a minister telling you all the things that will happen to you if you don’t have faith. Mark Twain sums that all up. “Faith is believing what you know ain’t so.”
Anyway, we’re facing a lot of problems. I think the best solution is to upload, or download, whatever you want to call it. Copy your intelligence, personality, knowledge base, common sense procedures and all that into a computer. I think that it will probably require well under a gigabyte, which costs fifty cents these days. But we don’t know how to do it. And if we can get people into a more manageable form, less expensive and more productive, then we can solve a lot of these problems. This is just a little list of the problems that us earthpeople face. Epidemics, pollution, energy, terrorism, and so forth. Virtually all those problems are the result of having too large a population. And that’s politically unmentionable. So everyone is talking about making the people who exist live longer, but when you have 10 billion people using 140 watts of their own energy and countless kilowatts of other energy, we certainly have some nice problems to face. If you could get them into little chips, then 10 billion people probably only cost a few hundred dollars an hour to run. Probably less.
We are facing an emergency in that in 40 or 50 years there will be hardly any young people to do the work. And so we need AI and stuff like that. In the time we have, I am going to talk about the state of artificial intelligence and why it has not made much progress since 1980. Here is a wonderful example of a PhD thesis in 1991 that could have earned an A in a first year calculus course at MIT because it could solve integration problems in about the same time as a student. The fact that it took about the same time as a human has absolutely no significance, because that machine ran at about 20 kilohert. New machines will soon run at 20 gigahertz.
The reason AI didn’t make progress is sort of paradoxical, because it really did make progress. But almost every researcher, all but about a dozen, as far as I could see, decided that in order to make an intelligent machine, let’s imitate the physicists. Newton found three laws to explain all of mechanics. Maxwell found four laws to explain all of electricity and magnetism. Einstein squashed them into less. Then of course quantum mechanics introduced the Schroedinger Equation and a lot of new problems. Physics envy. Wouldn’t you like to discover the three laws of thought? So, what happened in artificial intelligence is that around 1980 it split into ten or twelve fields. A large number of people make learning machines, where a machine does something at random and if it leads to a good result you reward it. Rule based systems became a multi-billion dollar industry. Just make your machine have a list. If you are in this situation, do that. It was wonderful and profitable, and sucked most of the good researchers out of science.
Neural networks are wonderful at learning and you don’t have to program them. However, there are lots of things they cannot do. In particular, when a neural net knows how to do something, you can’t make another neural net that understands what the first one can do and describe it and reason about it. So there is no reflection in modern artificial intelligence. Well, of course the difference between people and other pretty smart animals is that we can think about what we’ve been thinking. That’s missing from modern artificial intelligence. Isn’t that shocking? It’s so obvious.
Most people in this field make money by writing statistical learning programs, and for lots of problems you can solve these things. I’ve got lots of criticisms of those, but that would take a long time. Here is my theory in a nutshell of how the human mind works. The human brain has hundreds of brain centers. If you look in the index to Kandel and Scwartz, which is a twenty-pound book, you’ll see the names of 400 brain centers, and if you look them up, you’ll see some very, very, very vague ideas about what those brain centers do. Like the hippocampus has something to do with short-term memory. In fact, there are many other parts of the brain that are involved in short-term memory as well. So, what I think is going on is, if you think of the brain as 400 computers, each of them is specialized to do certain kinds of things. We know almost nothing about what those things are, except in the case of the visual cortex and the cerebellum, where there are fairly good theories. Although there is no good theory of how you recognize a telephone or a microphone in modern neuroscience. But there is quite a lot of knowledge about low-level processes.
So, here is my idea. What the new parts of the brain do, the frontal lobe and parietal lobe and the parts that make us smart, they have new kinds of pattern recognition that don’t recognize things like telephones and microphones, but this part of your brain has a lot of things that recognize patterns of activity in other parts of the brain, and in particular they recognize when you are trying to solve a problem and not getting anywhere. And these very powerful gadgets that I call critics, which no neurologist has thought of, so no one has designed experiments to look for them. I mention this because there is a nice criticism of my book on the Amazon page for The Emotion Machine. A neuroscientist says a few nice things, and then he says, “Where is the evidence for these theories?” He’s missed the point. The neuroscientists in general have not tolerated high level theories of how the mind works, and so they haven’t suggested that there are things like critics. He should be saying, “I love this theory, I’m going to go look for the evidence for that.” Of course, I only talk to freshmen as a rule, because even graduate students are too rigid.
So the idea is that the brain is full of these critics, and what the critics do is say, “I see what sort of trouble he’s in. I will switch him into using a neural net,” or using a genetic program, or using a symbolic-based goal-seeking analysis, and so forth. The nice thing about this theory is that all of a sudden, you see it explains emotions, too. There is nothing special about emotions. Emotions are just ways to think. The book starts with this dialog: “I’m in love with a wonderful person. I can’t think about anything else. My sweetheart is unbelievably beautiful,” and blah, blah, blah. If you look at what he said, he hasn’t said anything about her. He’s said things about himself. He says “indescribable,” meaning I can’t figure out why I’m attracted. “Unbelievably perfect.” No sensible person would believe this. So you see, we can think of emotions not as extra mysterious gadgets added to rational thinking, but the opposite in many cases. What happen when you get angry? You turn off your deliberative processes and social diplomacy. You show your teeth if you are an animal, or make fun of their mother, and this is just a way to intimidate them to make them go away. So there is nothing special about emotions. Generally they are simpler than the other kinds of thinking for which we don’t have a hundred different words. We have hundreds of words for emotions and tens of words for different ways to think.
So this is a graphic description of the thing. We think of the mind as a thousand resources, or whatever. When you are angry you turn on certain ones. When you are facing a hard problem and you are stuck, maybe you split it into parts, maybe you try trial-and-error, maybe you make an analogy or ask someone for help, and all of those involve reconfiguring your brain to deal with the problem in a different way. And that’s why the book is called The Emotion Machine. It’s to fool readers into realizing that the problem isn’t emotions, the problem is with the usual dumbbell distinctions between rational or intuitive, etc.
So if you read the book you will run into criticisms of other approaches to AI, a couple of theories about how to deal with them, but mostly very detailed discussions of why we need to understand our human ways to think. And every person has several dozen of them that are common, and for me the research program for the next 20 years would be to try to separate these out and analyze them, and see how they are interconnected, making diagrams of what are our ways to think and how do we recognize them. One central point of the book is that if you understand something in only one way, then you don’t understand it. Because if that way does not work, you are stuck, and have nowhere to go. So, my theory is that the brain has evolved so that when anything happens, you represent it in six or seven very different ways using different realms of thought: social, physical, mechanical, and there is a whole chapter on those.
And then, following Sigmund Freud, actually, my conjecture is that this could not have evolved at once. It had to have occurred in evolutionary epochs of various sort, and as far as I can see you need at least six layers of this stuff. Each of these layers is produced by genes, which are pretty similar, just as all of your vertebrae use the same low-level plans, but each of them is different in various ways, and has different shaped ribs connected, and different nerves and so forth. That’s the main idea.
Well, when we understand this stuff, then our brain scientists will be able to look at your brain and will have higher resolution scanners, will be able to see all of these processes, make copies of them in something like this. Here’s a $5 2-gigabyte gadget. So I think that’s the other path toward immortality, and as Aubrey de Grey said, there’s a trade off in the next couple hundred of years between being able to upload things and being able to preserve the biological parts, and we will eventually shift between these. If we could do this next year, then all of our global climate problems would go away. Because if people were just an inch tall instead of six-feet tall, that’s a factor of 100 or so. If you take the cube of that, you see it gives you a million times less stuff. The environment would be fine. It is fairly urgent not only to preserve your own irreplaceable and priceless self, but to preserve all of our irreplaceable and priceless selves. And let’s get going.
