economicus.ru
 Economicus.Ru » Галерея экономистов » Герберт Саймон

Герберт Саймон
(1916-2001)
Herbert Simon
 
Источник:Forbes, 11/16/98, Vol. 162 Issue 11, p98, 3p, 2 diagrams, 3c
Ross, Philip E.
FLASH OF GENIUS
What gives people intuition? Can you teach it? Can you replicate it on a computer? Nobelist Herbert Simon has some insights on the subject.
WHAT MAKES a good doctor, lawyer or stock picker? What enables some students to pick up math with ease, whereas others--including many with brains--flounder? If we understood how people think, we could teach them better. We could also program more smarts into our computers. So says Herbert Simon, psychologist, author of The Sciences of the Artificial (MIT Press, 1996) and 30 other books--and one of this century's greatest thinkers on the subject of thought.
Simon won a Nobel Prize in economics in 1978 for theories of decision making that turn on the nature of human expertise. His central finding was that pattern recognition is critical. The more relevant patterns at your disposal, the better your decisions will be. This is hopeful news to proponents of artificial intelligence, since computers can surely be taught to recognize patterns. Indeed, successful computer programs that help banks score credit applicants, help doctors diagnose disease and help pilots land airplanes depend in some way on pattern recognition. What about creativity? Surely Mozart owed his phenomenal early success to more than mere pattern recognition.
"Mozart composed for 14 years before he wrote any music you'd regard as world-class," Simon says. "You can go and listen to it. You can tell juvenile Mozart from 18-year-old Mozart. It's the same in all fields. Bobby Fischer got the grand master title in chess in just under ten years, and so did the Polgar girl [Judit Polgar, now 22]." Brainpower matters--but so does experience.
Simon's inquiry into expertise--its limits, its specificity, its structured impulse to provide adequate solutions rather than ideal ones--led to his Nobel-winning theory of "bounded rationality." In recent years he has adapted the theory to practical concerns ranging from teaching computers to think to teaching children to do algebra.
FORBES sat down with Simon recently at his puzzle-and-book-strewn office at Pittsburgh's Carnegie-Mellon University. At 82, Simon is professor of computer science and psychology, far from retirement with a full teaching load.
FORBES: Your office is cluttered with brainteasers and puzzles, like that Tower of Hanoi problem. What can a psychologist learn from games?
Simon: Take chess mastery. Some have attributed chess skill mainly to analysis, but our research shows that pattern recognition is key. Give a master a quick glimpse of a position from a real game and he can reconstruct it almost without error; show him a random position and he does hardly better than a novice. He relies on a store of characteristic patterns.
Gary Kasparov's chess rating is 2,800, yet when he plays against teams of grand masters rated around 2,600 simultaneously, he still beats them. He loses only about 100 points of playing strength even though he thinks for seconds, not minutes.
This is true of other expertise. Your doctor probably has diagnosed you before you've finished reciting your symptoms.
You cite a study that found that business school students took hours to calculate solutions to problems that experienced businessmen found off the cuff.
Exactly. I think we need to pay much more explicit attention to teaching pattern recognition. I used to teach organizational theory in the business school here, and it was hard because although the students had lived in organizations all their lives, they hadn't thought of them as organizations. Most of them, in those days, had no business experience.
You calculated that experts must have access to something between 100,000 and 2 million memory patterns. How did you arrive at the number?
It takes at least ten years of hard work--say, 40 hours a week for 50 weeks a year--to begin to do world-class work. We found it takes eight seconds to learn a pattern for a day, and quite a lot longer to learn it permanently. That takes you to the million-pattern estimate, if you allow for certain inefficiencies in learning and also for forgetting.
You predicted in 1956 that a computer would beat the world chess champion within ten years.
I was off by 30 years.
When IBM's Deep Blue beat Kasparov, you insisted that the machine was thinking. A lot of philosophers quibbled with that.
That's because they define thinking as that which computers can't yet do. They keep raising the bar.
Why this great mystique about a word called thinking? What is supposed to be implied by putting it in quotation marks or italics or whatever the hell it is that people do? Why don't we use it like an ordinary term? Do people have mass? Do tractors have mass? Yeah, they both have mass. Does that undignify people?
I would simply define thinking in an operational way. If we saw a human doing what we'd call thinking, I'd like to apply the term in exactly the same way to computers.
Do you think people are afraid that smart software will steal their jobs? I don't know about any computer programs that purport to do magazine writing, but I'll bet there are some good CD-ROMs that teach college courses.
It turned out that horses were better at pulling plows than people were at pushing them. So what's new? This doesn't happen instantly, and people can go and do other things.
If you look at what's happening in laboratory science, instruments are taking over the data gathering and the data interpretation. The diagnosis, so to speak, has been taken over in spades. If we succeed in deciphering the genome, it's not going to be because we hired 200,000 slaves and put them in a dungeon, but because we got the right computer programs.
Innovators may always be in demand, but most professionals sell knowledge they've acquired from others. Won't software take over that function?
It may indeed. But I hate to speculate about just when. These things don't happen suddenly, and people have time to adapt. The good career bet has always been to exercise your curiosity and get some general tools that you can apply anywhere. My degrees were in political science, and I haven't had any professional occasion to use political science for 60 years.
You've demystified human intuition as mere pattern recognition. How do you explain the way we sometimes instantly see a solution to a problem that had resisted attack for a long time?
Insight, you mean. I've modeled that with my colleague Craig Kaplan in an experiment called the Mutilated Checkerboard. We showed subjects a 64-square checkerboard and covered it with 32 dominoes--one for every two squares. Then we cut off the upper-left and lower-right corners of the board and asked the subjects to either cover it with 31 dominoes or prove that it can't be done. Many couldn't, even after working for hours. When they finally did, they did so in a flash.
Representation is everything. For example, if you don't color the board red and black, the problem becomes much harder.
That's because the colors give a big hint--they point to a pattern in the way the dominoes cover alternate squares. You're filling in the blanks for them, the way a computer programmer specifies every last step in an algorithm.
Well, sure. I mean, that's what we are--computers.
Can you apply these ideas to school instruction?
I have been much involved since about 1980 in an experiment, which is no longer an experiment, for teaching algebra and geometry in middle schools in China. We're in maybe 200 schools, and it's spreading.
Take a look at a typical algebra book. In the first chapter it tells you what you can do to an equation. You can add the same number to both sides; you can subtract the same number from both sides; you can multiply both sides by the same number; you can divide both sides by the same number; you can't divide by zero. And maybe the teacher explains, or maybe the teacher doesn't, and maybe you understand, or maybe you don't, that if you do those things and only those things, the X that will satisfy the equation will be the same as it was before.
In the second chapter you learn to solve equations by applying those rules. But you show me the algebra book that explains how you choose which rule to apply when.
Isn't that the kind of pattern recognition you just have to gain through experience?
No. There are perfectly good rules. I can write down three "if/then" rules: If there is a number on the left-hand side, subtract it from both sides. If there's a letter on the right-hand side, subtract it from both sides. If you've done those two things and the X on your left-hand side has a coefficient other than one, divide by that coefficient. So you have X equals the answer.
But notice that algebra teaching has concentrated on the rules, the legalities. What do legalities do for you? They tell you what you must not do. They don't tell you what to do. The teacher gets up, writes a proof on the board and writes Q.E.D. You have followed every step and checked it; the proof was right. And then you go up and ask, "But what made you think of using those steps?" That hadn't been taught.
Did you test the curriculum out against the traditional way of teaching math?
Under the experiments we ran, [students] could learn about half again as fast. You can get such advantages from other tutoring schemes, but I think they all have this idea implicit in them.
My colleague John Anderson has devoted most of his research for the past ten years to computer tutors in geometry and algebra. He has them in somewhere between half a dozen and a dozen cities in the U.S., and they're performing very well.
What's the trick with that Tower of Hanoi? You have to move the rings, one at a time, from one pin to another without ever stacking a larger one on a smaller one, right?
Right. The problem explores human memory, above all the seven-item limit that makes telephone numbers hard to retain. Attack the problem with short-term memory and you can do three or, at most, four disks. The number of moves doubles with each added disk. The smart strategy is to try to get the largest disk to the far spike, and if you can't, then remove what's blocking you. If you can't do that, then remove what's blocking you there, and so on--a recursive process.
Новости портала
Рекомендуем посетить
Allbest.ru
Награды
Лауреат конкурса

Номинант конкурса
Как найти и купить книги
Возможность изучить дистанционно 9 языков
 Copyright © 2002-2005 Институт "Экономическая школа".
Rambler's Top100