Chapter 33 Chapter 7 Peasant Economy in the Glass House

On Tuesday, September 22, 1987, the day Holland and Arthur came to participate in Langdon's artificial life seminar, Holland and Arthur left the artificial life seminar in Los Alamos and drove to Get off the platform and return to Santa Fe.Along the way, they occasionally stopped to enjoy Xiang Wan's scenery.To their east, the Sangri de Cristo Mountains rose majestically seven thousand feet from the Rio Grande Valley.They've been driving for a solid hour, talking about "boid": Here's Craig Reynolds from the Symbolics Corporation in Los Angeles A computer simulation presented at the workshop.

Arthur was fascinated by the simulation.The program, Reynolds claims, was intended to capture the essence of flocking behavior of birds, or flocks of sheep, or flocks of fish.It seemed to Arthur that he had succeeded in doing so.Reynolds' basic idea was to place an automatic, bird-like actor, "The Bird," in an on-screen environment full of walls and obstacles.Every Bird follows three simple rules of conduct: 1. It tries its best to keep the minimum distance from other obstacles, including other "Byrd". 2.It tries to maintain the same speed as its neighboring Bird.

3.It tried its best to move toward the center of aggregation of its neighboring "Peder" group. Strikingly, none of these rules say: "Gather in groups".Quite the contrary: the rules are entirely local, dictated only to what each individual "bird" can do and see from its neighbours.Therefore, if the phenomenon of clustering can really be produced because of this, then this kind of motivation can only come from the lowest level, and it can only be a sudden phenomenon.But every time it can indeed produce the phenomenon of clustering.Reynolds started the simulation by scatter- ing the Birds randomly across the computer screen, and they would spontaneously gather themselves into clusters that circle obstacles in a fluid, very natural form. fly.Sometimes, the flocks can even break up into smaller flocks, bypass the obstacles on both sides, and regroup on the other side of the obstacle, as if Bird had been doing it on purpose.Once, a "Bed" unfortunately bumped into a pillar, flapped its wings and circled for a while, as if dizzy, when the "Bed" group started to move, it immediately followed and rejoined the group.

According to Reynolds, this last part of the process proves that Bird's behavior really does emerge.Neither its rules of conduct nor other computer codes tell any particular "Bird" that it should take such an action.So as soon as Arthur and Holland got into the car, they began to ponder this question: To what extent was Bird's behavior internalized, and how much was it really unexpected emergent behavior? Holland stood his ground.He had seen too many examples of simulating "emergent" behaviors where instructions had been programmed into the program from the start. "I said to Brian (Arthur), you have to be careful. Maybe all the simulations on display here, including the example of hitting a pillar, are obviously programmed in, and the programmed rules don't Any ability to learn new things. I'd like to at least be able to put other things into this simulation, change its environment, and see if it's capable of producing reasonable behavior."

Arthur could not argue eloquently with this point of view."But to me, I don't know how you define 'true' emergent behavior," he said. In a sense, everything that happens in the universe, including life itself, is already internalized to be able to dominate Rules for quark behavior.So, what exactly is emergence?How do you recognize it when you face it? "This goes directly to the heart of artificial life." Since Holland and everyone else could not answer the question, neither he nor Arthur could draw a firm conclusion.But in retrospect, Arthur says, the discussion between them did plant a seed in his sleep-deprived mind. In early October 1987, exhausted but brimming with joy, Arthur returned to Stanford from his work as a visiting scholar at the Santa Fe Institute.When he returned, after a good night's sleep, he began to think carefully about all he had learned and heard in Santa Fe. "Holland's Genetic Algorithm. The classifier system and the concept of 'Bard' left a deep impression on me. These new concepts, new ideas, and the infinite possibilities opened up in front of me made me Thinking about it for a long time. My instinct tells me that these concepts are the answer. But the point is, what is the problem with economics?"

"My initial interest was in how the economies of third world countries changed and developed. So, around November 1987, I called Holland and said, I had an idea of ​​how to apply these concepts to economics. I think you could do a little simulation of the development of a peasant economy in a virtual glass house in a university office, actually on a computer. But it has to be all small actors, these little Actors should be able to learn to be smart without being programmed, and they must be able to interact with each other." "And then, in this dreamlike vision, you walk into the office one morning and say, 'Hey, look at these guys! Two or three weeks ago they were bartering, and now they have a joint stock company. ' The next day, you walk into the office and say, 'Oh, they know they're going to form a central bank.' A few days later, all your colleagues gather around, and you're watching: 'Wow! They've got a union!' What else do they want to do next?’ or half of them have become communists."

"I couldn't articulate the idea at the time," Arthur said. But he knew that the glass-house economic simulation would be very different from conventional economic simulations.In conventional economic simulations, the computer simply puts together a different set of equations.In his glass house economy, the economic actors are not mathematical variables, but actors, entities caught in a web of interactions and accidental events.These entities make mistakes and are able to learn from experience.They have their own histories and, like humans, are not governed by mathematical formulas.Of course, from practical considerations, they are much simpler than real humans.But if Reynolds can indeed produce very realistic flocking behavior with three simple rules, then we can at least imagine that perhaps a computer full of well-designed adaptive agents could produce very realistic economic behavior.

Arthur said: "I vaguely wondered if I could use Holland's classifier system to make these agents. I know how to do it. John (Holland) has no idea how to do it." directly applicable, but he was also very motivated.” So the two agreed that it would be a priority research topic when the economics program at the Santa Fe Institute began next year.