Chapter 35 Santa Fe Philosophy

Santa Fe Philosophy The economics program will begin at the Santa Fe Institute in September 1988, beginning with a second week-long economics seminar.So Arthur has lived in Santa Fe since June, and he will spend a summer preparing for it, and every minute is precious to him.He found that work got hectic in the fall, when conference attendees began reporting. “I get people coming to me every day. Like one guy who doesn’t know how to change a light bulb and asks if I’ll do it. The place is so small that I sometimes have to sort out questions like which office is available for smokers? or , how can you share an office with someone who wears shorts all the time and shows hairy thighs? The person who asked the question really can't accept sharing an office with someone who wears shorts. And I have to take full responsibility for the seminar Organizing. Part of organizing is going out and networking people, talking to them, asking for their opinions, spreading the word about the Santa Fe Economics Symposium.”

As Arthur discovered, being the boss meant not being out and playing with other kids all the time, but having to play an adult all the time.Despite the help of other staff at the Institute, Arthur found that 80 percent of his time was still spent on non-scientific tasks, which were not very interesting.He said he once returned to his rented house in Santa Fe and complained to his wife, Susan, how little time he had for research."She finally said, 'Oh, stop nagging, you've never been this happy in your life,'" Arthur said. "She was right." Indeed, she was right.Because despite all these clerical tasks, the remaining 20 percent is more than enough to make up for everything, Arthur said.By the fall of 1988, the Santa Fe Institute was a vibrant scene.This is not only because of the economics program, but also because late last fall, the long-awaited federal government funding actually came down through the National Science Foundation and the Department of Energy.Cowan was not able to persuade these institutions to meet his funding requirements in full. For example, the Institute still has no funds to hire long-term researchers, but these funds have committed to granting Santa Fe for three years starting in January 1988. $1.7 million.So before 1991, the Institute had financial security.The institute finally has enough money to seriously start working toward the goals it was created for.

A scientific committee chaired by Gell-Mann and Paines approved fifteen new workshops.Some seminars will approach complexity from a core physics perspective.The best example in this regard is the symposium on "Information Physics, Entropy and Complexity" that will be organized by Polish physicist WOjciech Zurek in Los Alamos.Zulek's idea was to start with well-defined concepts in computer science, such as information and computer complexity, and explore their relationship to quantum mechanics, thermodynamics, quantum radiation from black holes, and the (hypothetical) quantum origin of the universe. deep relationship between.

Other workshops will explore complexity from a biological perspective.Two of the best examples are two seminars on the immune system organized by Los Alamos biologist Allen Perrlson.Perelson had convened a very important immunology seminar at the Santa Fe Institute as early as June 1987, and has been chairing a small research project at Santa Fe.Perelson's idea is that the body's immune system has billions of highly sensitive cells that flow with the bloodstream, and when viruses or bacteria appear, they work together with antibodies to neutralize the invading virus or bacteria.The immune system is a complex adaptive system, in much the same way as the ecosystem and the organization of the brain.So Santa Fe's ideas and techniques should shed light on immune-related problems like AIDS, or multiple sclerosis, or arthritis.In turn, because so much is known about the molecular structure of the immune system in detail, projects devoted to the immune system should be able to put some of Santa Fe's advanced concepts into practice.

At the same time, the Science Committee also strongly advocates the recruitment of visiting scholars and postdoctoral fellows who have not participated in Santa Fe projects and seminars.That's the approach the institute has always followed: scoop up the best of the best and see what happens.Members of the scientific committee joke that the Santa Fe Institute itself is an emergent phenomenon.Actually, it's a joke they take pretty seriously. All of this is in line with Cowan's wishes.He is always eager to find more heroes with unspeakable fire burning in their souls.But Cowan thinks it's not just a matter of finding greatness.You can say that the Institute has a lot of talented people, but they don't know what you want to do.What the Institute is looking for is people who can hit each other with sparks: "Some of them seem to be dead-eyed in the contact, while others keep in touch with us from then on." If this is the case, then you are in fact A certain way of exercising a very coercive power: the power of knowledge.If you find that the other person's understanding of the concept of Santa Fe comes from the deepest part of his brain, and such thoughts are always hovering in his mind, then you have found the right person.You don't use the method of pulling people away from the flesh, but use the charm of knowledge to gather people's hearts.You're engaging them with brains, not carnival parties.

Finding such talent is harder than ever, but they do exist.And there are more and more such talents flocking to Santa Fe, causing the small monastery to be often overcrowded.Indeed, this situation is unimaginable: various seminars are held in the chapel for many years, three or four people are often crowded in the office that was originally only enough for one person, and colleagues are endlessly scribbling on the blackboard. What, what are they arguing about, the free discussions in the corridors and under the big trees are constantly forming and reorganizing, and the vitality and camaraderie among people are like an electric current that infects everyone.This is just as Stuart Kaufman said: "My vision of the world is refreshed twice a day."

Everyone feels the same way.Arthur said: "Typically, every morning, most people will creep into the office and you'll hear the slight beeping of the computer terminal and the clicking of the keyboard. Then someone will be poking at your door. Have you ever done this? Have you thought about that? Can you talk to a client for half an hour? Then we'll go to lunch together, often to the Kanyang Road restaurant we call the 'Faculty' club' and we became so regulars that the waitresses there don't even bring us menus anymore. We always say: 'give me a number five' so they give it to us without even asking What to eat."

The conversations between them were endless, most of them very good and inviting.Arthur said what he remembers most are the impromptu freelance sessions that started anytime, anywhere.Those discussions always start near noon, or in the afternoon. "Three, four, five times a week. Someone would come out in the hallway and say, 'Hey, let's talk about X,' and five or six people would gather in the chapel, or something like that. Often they gather in the small conference room next to the kitchen. The small conference room is very dimly lit, but it is next to the coffee room and the Coke machine. The room is Indian style, with a picture of Einstein hanging on the wall , Einstein in a turban smiling at us."

"We'd sit around the table. Stubert (Kaufman) might be leaning on a mantelpiece. Someone might write a question on the blackboard. We'd start asking a million questions about it, all very Well-intentioned debate. You never speak bad words to each other, but the questions you ask are quite sharp, because what you talk about is the most fundamental problem, not the technical problem of economics research, not how you solve this or that fixed-point theorem, or In physics, questions like why materials superconduct at minus 253 degrees are questions about where science is going. Questions like, how do you deal with limited rationality? When things really get How should economics progress when it is as complex as chess? What do you think of economics that is always evolving and never reaches a point of equilibrium? If you applied computer experiments to economics, what would you do? "

"I think that's what makes Santa Fe what Santa Fe is: the answers we're trying to find and the technical tools we're borrowing happen to be shaping the concept of Santa Fe economics." Arthur remembers one series of discussions particularly well because it refined his thinking.He said Arrow and Hann of Cambridge were there at the time, so it must have been during their visit in October-November 1988. "Me, Holland, Arrow, Hahn, maybe Kaufman, and one or two others get together. We go back and forth about what economists can do about bounded rationality." That is, if economics What would become of economic theory when it no longer assumed that people could spontaneously and computer-like reason about the outcome of any economic problem, even one as complex as the game of chess?

They discuss the issue almost every day in small conference rooms.Arthur remembered Hahn pointing out that economics draws upon perfect rationality because it is a benchmark.If people were perfectly rational, theorists could say with absolute certainty how those people would react.But what about total irrationality?Hann asked curiously. "Brian (Arthur), you're Irish. You might know," he asked. Arthur laughed, and Hann went on seriously. There was only one way to achieve perfect rationality, and there were countless ways to achieve partial rationality.So for humans, which approach is correct? "How do you define the scale of rationality?" How to define the scale of rationality? "This is Hann's metaphor, which made me deaf. I thought about it for a long time, bit a lot of pencil ends, and had many discussions." Arthur said.He and others, like observing how the image on a photograph is displayed in the dish of the photo development, slowly found the answer: the way to define the scale of rationality is to let it go, let the actor define rationality for himself scale. Arthur said: "You would take the John Holland approach. You would model all these actors as a classifier system or a neural network, or some other form of adaptive learning system, and then let the label follow the actor Constantly changing with lessons learned. So all actors start out completely dumb. That is, they make random, wrong decisions. But as they get feedback from each other, they get smarter and smarter." Maybe they will really become very smart, maybe not, it all depends on their experience.Arthur realized that these adaptive AI agents are exactly what you want to use to build a real dynamic theory of economics.If you put them in a stable, predictable economic environment, it might become apparent that they make exactly the kind of highly rational decisions that neoclassical economic theory predicts—not only Not only because they have access to comprehensive information and infinitely rapid reasoning, but also because stability gives them plenty of time to figure out the trick. But these same actors can still function if they are exposed to simulated economic changes and turmoil.But maybe it's not so perfect.They will stumble, they will fail, and they will make all kinds of mistakes in the first place, just like humans display.But because they are built with learning algorithms, they can gradually learn how to act rationally.Likewise, if you put these agents in a competitive environment, like a game of chess, where they have to take actions against each other, you can see how they make choices.And if you put these actors in a simulated economic environment that simulates prosperity, you see how they explore an infinite space of possibilities.In fact, wherever you place them, they will try to do something.Neoclassical economics theory cannot explain the dynamic phenomena and changes in the economy, but the model full of adaptive actors is completely different, and the dynamic mechanism of the latter is internally controlled in advance. This, Arthur realized, clearly paralleled his vision of the glass house economy.This is exactly what he realized when he read "The Eighth Day of Creation" ten years ago.It's just that he sees it more clearly now.This is the seductive "Santa Fe idea": as opposed to the neoclassical view of economics that emphasizes diminishing returns, stagnant equilibrium, and perfect rationalization, Santa Fe emphasizes increasing rates of return, bounded rationality, and evolution and Motivation to learn.Instead of basing their theories on mathematically manipulable assumptions, they sought to create psychologically plausible economic models.They see the economy not as some Newtonian machine, but as something organic, adaptable, surprising, alive.Instead of speaking of the world as some inert thing buried deep in the permafrost, they learn how to speak of the world as a dynamic, ever-changing system balanced on the edge of chaos. "Of course, it's not a new idea in economics," Arthur said.The great economist Josoph Schumpeter may not have known the term "on the brink of chaos," but he called for an evolutionary approach to economics in the thirties.Richard Nelson and Sidney Winter of Yale University have been fanning the evolutionary movement in economics since the mid-1970s, with some success.There are other researchers who have studied learning effects in the field of economics.Arthur said: "But in these early learning simulations, the agent is assumed to have formed a model that can give some correct feedback to the external environment, and the learning is just to make this feedback model by adjusting a few connection points. become sharper. And what we need is something more realistic. What we need is to let the emergence come from the 'built-in model', and the agent establishes some kind of feedback mechanism from the inside of the mind during the learning process. We have There are many methods that can be used to analyze this process, including Holland's classifier system and genetic algorithms. Also, Richard Palmer has just finished a book on neural networks. David Lane and I know how Doing systematic analysis mathematically on the basis of probability." Urmleaf and Keniowski are experts in the study of guessing learning.We also collected data from a complete set of psychological studies.These methods really paved the way for us to model adaptations and refine their algorithms. Adds Arthur: "In fact, the intellectual impact that was crucial to us in the first year in general was learning to use computers, specifically learning about computing from RAND, not learning about condensed matter physics, paying Incremental rate, it's not about learning computer science, it's about mastering learning and adaptability. When we explored this concept with Arrow, Hann, and others, it was obviously exciting to all of us that we could use this completely different approach Come study economics.”  … Economists at Santa Fe, while exhilarated by this economic prospect, are vaguely disturbed.The reason, Arthur says, is that he didn't start touching on certain issues until much later. "Economics, as it is usually practiced, operates on the basis of purely deductive models. Any economic situation is first deduced using mathematical formulae, in which economic actors are assumed to be Rigorous analytical reasoning to solve economic problems. Then came Holland, neural network researchers, and other functional theorists of computer learning. They all talked about agents operating on the basis of inductive models according to fragments Induction allows us to infer the presence of a cat when we glimpse a tail that is disappearing around a bend, and induction allows us to infer the presence of a cat when passing a zoo. Even if we have never seen a red-crested parrot before, we know it is a bird when we first see it.Induction is what enables us to survive in a chaotic, unpredictable, and often incomprehensible world. "If you're airdropped into Japan for a negotiation and you've never been to Japan before, you don't know anything about how the Japanese think or behave or work, you don't fully understand what's going on around you, what you do Most of them will not fit the local cultural background and customs. But over time, you notice that some of the things you do succeed. Gradually, you and your company somehow learn to adapt to the environment and understand the local environment. behavior.” (Of course, whether Japanese companies actually buy your products is another matter.) Imagine being in a competitive environment such as playing chess, where the players gain some insight into the intentions and capabilities of their opponents. piece of information.To come up with countermeasures, they do use logical, deductive reasoning.But this method can only infer the number of the next few steps at most.Chess players operate more often by induction.They try to deal with the situation with assumptions, analogies, past experience, and rules derived from actual operation.No matter what method is used, as long as it can win, it doesn't matter even if they don't know the reason.Therefore, induction cannot rely solely on precise, inferential logic. Arthur admits that even he was puzzled by it at the time. "Until I came to Santa Fe, I thought you had to define the economic problem before you could talk about it. If you can't define the problem clearly, what can you do with it? You can't use logic, of course. Solved the problem." "But Holland told us that's not the case. When we talked to Holland and read his academic papers, we began to realize that the paradigms he was talking about were problems whose content was not well defined, whose environment changed over time. We said to him: 'John, how can you learn in such an environment?'” Holland's answer, in essence, is that you learn in such an environment because you have to: "Evolution doesn't care if the problem is clearly defined." respond.They do not have to make assumptions about where this payment comes from.In fact, that's what his classifier system is all about.Algorithmically, these systems are tightly defined, yet they can operate in environments that are not well defined at all.Since classifier rules are assumptions about the world, not "facts," they may contradict each other.And, because the system is always probing those assumptions and distinguishing which ones are useful and rewarding, it can learn even from fragmented information, in ever-changing, unpredictable environments. "But it's not doing the optimal thing," the economist complained.Economists are convinced that a rational agent uses its "function" to the maximum. "Maximization relative to what?" Holland asked.Let's talk about your ill-defined criteria: In any real-world setting, the space of possibilities is so large that it's impossible for any single actor to find, or even tell what is optimal.What's more, the environment may change unforeseen. "This whole idea of ​​induction fascinates me," said Arthur. "You can imagine that the agent is confronted with an ill-defined problem, an ill-defined environment, and a change in which there is no direction at all, and you do economics in this context. Of course, if you think about it for a moment, you will Recognize that this is what life is all about. People often have to make decisions in situations of ambiguity that they don't even understand. You're going through the mud, changing your mind, copying other people's Experience, trying past successes. In fact, economists have talked about this behavior before. But we now need to find a way to analyze it precisely and put it into the core of the theory.” Arthur remembers an important debate at that time that got to the heart of the problem. "That was a long discussion between October and November," he said. "Aro, Hahn, Holland, and I, maybe five or six people. We just realized that if you do economics this way, if that's the Santa Fe way, then in economics Perhaps there would be no equilibrium at all. The economy would be like a biological environment: forever evolving, changing, always exploring new areas of development.” "Now our concern is that it seems impossible to study economics under these conditions. Because economics implies knowledge of equilibrium. We are accustomed to study problems by looking at butterflies, that is, by pinning butterflies to cardboard , put them in balance, and then look at them carefully, instead of letting them fly free around you. So Hahn said: 'If things don't repeat themselves, if things are not in equilibrium, what should we economists do? Say? How do you predict things? How do you form a science?'” Holland took this issue very seriously and thought about it for a long time.Let's look at meteorology, he said to them.The weather is never the same, never exactly the same weather.We basically cannot predict the weather for more than a week in advance, but we can understand and explain various weather phenomena in the sky, and we can recognize important meteorological features such as fronts, air currents, and high-pressure circles.We can understand weather dynamics and how they interact to produce different weather conditions in local areas.In a word, although we cannot make complete predictions about the weather, meteorology is still a true science.The essence of science is understanding and interpretation, and this is what Santa Fe hopes to contribute to economics and other social sciences.Just as meteorology understands and explains fronts, they understand and explain dynamic social phenomena, he said. "Holland's answer was a revelation to me, and it thrilled me. I've been thinking for nearly a decade about the fact that large parts of the economy are out of equilibrium, and I don't know how to do economics without equilibrium. Research. John (Holland)'s discussion immediately opened up the knots in my mind and made me enlightened." Arthur said, indeed, these conversations in the fall of 1988 made me really realize how far-reaching changes the Santa Fe concept would bring to economics. "Many people, myself included, naively assumed that what we would get from physicists and people doing computer learning would be new algorithms, new problem-solving methods, and new technical frameworks. But the results were huge. Not the same. Often what we gain is a new attitude, a new way of looking at a problem, a whole new worldview.”