Chapter 8 Chapter 5 Game Master

Bridge, golf, croquet, chess, go, checkers, you name it, he could play.However, for Horan a long time ago, these games were not just games.He began to notice that some games had a special magic According to the agenda, the second presentation of the economic conference will begin after lunch on the first day and continue throughout the afternoon.The keynote speaker is John H. Holland from the University of Michigan, and the topic is the adaptive process of the global economy. Now that Arthur had finished his speech, he was curious about the next one, and not just because the title sounded interesting.That fall, Horan was another visiting scholar at the Santa Fe Institute, and the two of them were arranged to live in the same room.But Horan did not arrive in Santa Fe until late the night before, when Arthur seized the last chance to go to the monastery and practice repeatedly for the next day's speech.Arthur didn't see Hor Nan, he only knew that Hor Nan was a computer scientist, and according to the Institute, he was a very good person.

The Institute's claims appear to be correct.When everyone began to return to their seats, He Nan had already stood on the stage, ready to start.He was a dapper Midwesterner in his mid-sixties, with a broad, ruddy face that seemed to be smiling at all times, and a high-pitched voice that made him sound like an earnest graduate student.Arthur took an immediate liking to him. Horan began to speak.Within a few minutes, Arthur's drowsiness disappeared immediately, and he listened intently to Hor Nan's speech. eternal novelty Hornan begins by pointing out that economics is an excellent example of what the Santa Fe Institute calls a complex adaptive system.In the natural world, such systems include the brain, immune system, ecology, cells, embryos, and ant colonies.In the human world, this includes cultural and social systems like political parties or scientific communities.Once you learn how to discern, you will find these systems everywhere.Regardless of location, however, these systems seem to have some important common characteristics.

First, every system is a network of many agents acting in parallel.In the brain, the agent is the nerve cell; in the ecosystem, the agent is the species; in the cell, the agent is organelles such as the nucleus and mitochondria (mitochondria); is the cell, and so on.In an economic system, the actor may be an individual or a family; if the economic cycle is taken as an example, the actor may be a company; if international trade is taken as an example, the actor may even be the entire country.But in either case, each agent's environment is the product of its interactions with other agents in the system.Actors are constantly acting on, or reacting to, other actors, so that hardly a single thing in the environment is constant.

What's more, Hornan said, the control functions of complex adaptive systems are very decentralized.For example, there are no main neurons in our brains, nor are there main cells in the developing embryo.If there is any coherent behavior in the system, it must be due to competition and cooperation among actors.This is true even in economic systems.Ask any U.S. president who wants to revive the economy and he will tell you that no matter what Washington does about interest rates, taxes, and the money supply, the overall performance of the economic system still depends on millions of individuals Countless economic decisions are made.

Second, complex adaptive systems have many different levels of organization, and actors at each level are the basic units of actors at higher levels.Proteins, fats, and nucleic acids will form cells, a group of cells will form tissues, tissues will assemble to form organs, organs will form organisms, and many organisms will form an ecosystem.In the brain, one group of neurons forms the language center, another group of neurons forms the motor cortex, and another group forms the visual cortex.Similarly, a group of employees makes up a department, and many departments make up a company, which in turn makes up the economic sector, the national economy, and even the world economy.

Constant revision and reorganization Horan believes that it is particularly important that when a complex adaptive system has accumulated considerable experience, it will constantly modify or reorganize its basic units.Generations of organisms modify and reorganize their tissues through the process of evolution.When a person learns different knowledge, his brain will continue to strengthen or weaken countless connections between neurons.A company promotes high performers and rearranges its org chart to improve efficiency.Nations will sign new trade agreements and make new allies. On a deeper and fundamental level, all these processes of learning, evolving, and adapting are the same.One of the basic mechanisms of adaptation in any system is the modification and reorganization of basic units.

Third, all complex adaptive systems anticipate the future.Obviously, economists are not surprised by this.Since the downturn is expected to prolong, many people may put off buying a new car or taking a vacation, which in turn ensures a prolonged downturn.The anticipation of oil shortages may also lead to violent trading shocks in the crude oil market, regardless of whether an oil shortage will actually occur. But in fact, anticipation and forecasting mean more than human foresight.Every organism, from tiny bacteria onwards, has embedded in its genes codes of predictions: in certain environments, the organism specified by the genetic blueprint is likely to perform better.In the same way, every creature with a brain also has countless implicit predictions, which are translated into codes and stored in the learning database: in the case of ABC, it is more advantageous to adopt XYZ responses.

In general, every complex adaptive system is constantly making predictions based on its assumptions about how the outside world works.Moreover, these internal hypothetical models are not just passive blueprints, they are actually very active, just like subroutines in computer programs, which can come alive under certain conditions and execute (generate) the behavior of the system.In fact, you can think of this internal model as the basic unit of behavior.And they, like other units, can be tested, refined, and reorganized as the system gains experience. Equilibrium equals death Finally, Hornan said, complex adaptive systems often have many niches, allowing each actor to occupy a niche.Thus, in the economic world, there is a place for a computer programmer, a plumber, a steel mill, a pet store, and a place for a sloth and a butterfly in the rainforest.And, as each slot is filled, it opens up more slots for new parasites, new predators or prey, new symbiotic partners.So, the system is constantly opening up new opportunities.This also means that it is meaningless to discuss the equilibrium state of complex adaptive systems. Such systems will never reach an equilibrium state, but are always developing and transforming.

In fact, if such a system were ever to reach equilibrium, its state would be not just stable, but dead.Furthermore, it is pointless to imagine that the agents in the system cannot develop their adaptability or utility to the limit.Because the possibilities are too great, it is impossible to find the limit.The most an agent can do is to change or improve its own behavior as much as possible relative to the performance of other agents.In short, a property of complex adaptive systems is perpetual novelty. No wonder, says Horan, it is difficult to analyze complex adaptive systems using traditional mathematics.Traditional techniques like calculus or linear analysis are great for describing particles that are invariant in a fixed environment; but if you really want to gain insight into economic or complex adaptive systems, what you need is an emphasis on intrinsic models, the emergence of new fundamental units , as well as mathematical and computer simulation skills of frequent interactions between various actors.

While Horan was giving a speech, Arthur was writing vigorously.Arthur took notes faster as Horan continued to describe the various computer tricks he had developed over the past thirty years to make these concepts more precise and useful.He said: "It was unbelievable, I sat there dumbfounded all afternoon.Not just because Hornan's ideas coincide with the economics of increasing returns he's been working on for the past eight years, and not just because Hornan's ideas about niches are exactly what he and Kaufman have learned from autocatalysts over the past two weeks. The group got inspired; but Horan's whole unified, clear, flat way of seeing things will make you pat your forehead and say: Of course it is!Why didn't I think of it?Horan's thoughts can make people have an epiphany, and more thoughts start to explode in their minds.

Arthur said: He Nan answered all kinds of questions that I have been asking myself for many years: What is adaptation?What is Emergence?There are still many doubts that I don't know myself.Arthur wasn't quite sure how all of this applied to economics, and in fact, as he looked around, he saw other economists looking either skeptical or confused. (At least one person is dreaming about Duke Zhou.) However, I believe that Horan's research is much more complicated than mine.He felt that He Nan's thoughts must be very important. sympathy Of course, the Santa Fe Institute thinks the same way.No matter how outlandish Hornan's ideas seemed to the scholars attending the economic conference, in fact, Hornan had long been a familiar and influential figure among Santa Fe's inner circle. Horan's first visit to Santa Fe was in 1985, when he was invited to participate in a symposium organized by Farmer and Packard on the topic of Evolution, Games, and Leaming. (It was also at this meeting that Farmer, Packard, and Kaufman reported the results of their simulation of the autocatalytic group.) Hornan's talk on emergence went well, but he remembered one audience Ask him some tough questions.The guy had gray hair, a focused and slightly sarcastic face, and his eyes shot straight through the black-rimmed glasses.I was a little rude when I answered the question.I don't know him, and if I knew who he is, I'd probably be scared to death!Horan said.This man is Gehrman. Regardless of whether Horan's answer was rude or not, Gelman obviously liked Horan's idea.Not long after, Gelman called Hornan and asked if he would like to join the then fledgling Santa Fe Advisory Council. Horan agreed.I liked this place as soon as I saw it, he said: I like what they talk about, the way they do things, and my immediate reaction was: I hope these guys like me because this place suits me! Gehrman obviously felt sorry for him.When he talked about Hor Nan, he used the word "brilliant", which was not an adjective he would casually add to anyone.But it wasn't often that Gehrman was eye-opening.In the early days, when the founders of the Santa Fe Institute, such as Ke Wen and Gellman, thought of complexity science, they almost completely revolved around the physical concepts they were already familiar with, such as emergence, collective behavior, self-organization, and so on.Moreover, they also thought that as long as the application of these concepts in economics and biology was added, it would be enough to form a very rich research project.Then Horan came along and published his analysis of adaptation, not to mention his computer model.Suddenly, Gelman and others found that their research program was missing a very important piece: What are these emergent structures actually doing?How do they respond to and adapt to their environment? During those months, they had been discussing whether the Institute should study not just complex systems, but complex adaptive systems.And Nan Horn's personal research project to understand the interrelated process of emergence and adaptation has become the main research project of the Institute.In the Complex Adaptive Systems Symposium organized by Cowen and Feldman in 1986, one of the first major conferences of the Institute (the first seminar Kaufman attended), he become more of a protagonist.The next day, Paines took Horan to a discussion with Rhett.Anderson also invited him to this big economic conference in September 1987. Horan readily attends all meetings.He's been obsessed with the concept of adaptation for twenty-five years.Until now, at the age of fifty-seven, he met Bole.It was great to be able to talk to people like Gellman and Anderson and be treated as equals!I can not believe it!Horan would have spent more time in Santa Fe if his wife hadn't been unable to leave Ann Arbor because of her work (she is the chief librarian of seven science libraries at the University of Michigan). However, Horan is an optimist.He's always been able to do what he wants, and he's always been amazed at his luck, so he has that genuine sense of humor that comes with being a happy man.It was almost impossible for anyone who met him not to like him. Arthur is one of them.That afternoon, after Horan's speech was over, he couldn't wait to go forward and introduce himself.In the days that followed, the two became instant friends.Horan found Arthur very pleasant.Few people can absorb the concept of adaptation so quickly, and then fully integrate it into their own concepts.Arthur was intrigued by the whole idea, and quickly grasped the gist. At the same time, Arthur also found that Horan was the most complex and fascinating intellectual among the people he knew in Santa Fe.Because of He Nan, he spent the rest of the economic meeting's agenda with extreme lack of sleep.On many late nights, he and Hor Nan sat around the kitchen table, drinking beer and discussing the mysteries of science. Chess game won without a fight He remembered one conversation in particular.The reason why Horn came to this meeting was that he was eager to understand the important topics in economics. (He Nan once said to him: If you want to do interdisciplinary research and enter other people's fields, then at least you should take their problems seriously. They spent a lot of time on research.) That night, when they gathered around Sitting at the dining table, Horan asked him straightforwardly: Arthur, what is the real problem in economics? Arthur replied without thinking: like chess. chess?He Nan couldn't figure it out. Arthur took a sip of his beer, looking for the appropriate words to explain, but he didn't fully understand it himself.Economists always see the system as closed and simple: the system stabilizes quickly, at most two or three patterns of behavior emerge, and then, nothing happens anymore.They assume that the agents of the economy are so intelligent that they can immediately perceive what is the most beneficial course of action in any situation.Think about it, if you look at it from a chess point of view, what does this mean?In the mathematical theory of games, there is a theorem that tells you that any finite, two-player, zero-sum game, such as chess, will have an optimal solution.That is, there is a way of choosing moves that will give the best performance for both Reversi players. Of course, no one actually knows what this solution is or how to find it.But the ideal economic agent that economists talk about has an immediate answer.At the beginning of the game, when the two armies face each other on the chessboard, the two actors can list all the possibilities in their minds, and they can roll out all possible moves that may force the opponent.Then iterate until all possible moves have been considered and the ideal first move has been found.In this way, there is no need to actually play chess at all. Whoever seizes the theoretical advantage can immediately declare victory, because he knows that he will definitely win, and the opponent will immediately admit defeat, because he knows that he will definitely lose. He Nan, would anyone really play chess like this?Arthur asked. Horan laughed, knowing full well how absurd this was.As early as 1940, when the computer was just invented, researchers thought about designing a clever program that could play chess.The father of modern information theory, Claude Shannon of Bell Labs, estimated that the total number of all possible moves in chess is ten to the power of 120. This number is incomparably large. In our observable universe The number of all particles in is also not that large.No computer has the means to examine all possibilities, and certainly no human can.We human chess players must obey the rules of practical experience to decide which strategy is best to adopt in a certain situation.Even the most extraordinary chess players have to play chess step by step, as if they are going down a deep, deep cave, and only the dim light of a lantern in their hand can guide the way.Of course, chess players will improve day by day.Horan himself played chess, and he knew that the great players of the 1920s would never have risked a match against a current champion, such as Gary Kasparov.But even so, they had only advanced a few yards in the vast unknown.This is why Horan called chess an open system: it has infinite possibilities. Arthur said yes.Compared with the most ideal state, the form that human beings can really observe and then study is still some distance away.Unless you assume your agent is smarter than the average economist."But, that's how we solve our economic problems," he said.The U.S.-Japan trade problem is at least as complex as a chess game, yet economists start by approaching the problem by assuming it's a rational game. So, he told He Nan, simply put, this is an economic problem.How do we turn an unintelligent agent exploring infinite possibilities into science? aha!Horan said that whenever he saw a ray of light, he would say so.chess!He understands this metaphor. the infinite sky of possibility Horan likes to play games, all kinds of games.He played poker every month in Ann Arbor for thirty years.One of his earliest memories is of watching grown-ups play poker at his grandfather's house and wishing he was old enough to play cards.When he was in the first grade, his mother taught him to play chess. She is also a master bridge player.Horan's family members are all passionate about sailing, and mother and son often compete in regattas.Horan's father was a first-class gymnast and loved outdoor activities.The whole family was always changing some kind of game: bridge, golf, croquet, chess, go, checkers, you name it, they all played. However, for Horan a long time ago, these games were not just games.He began to notice that some games had a special magic that transcended winning and losing.During his freshman year of high school (around 1942 or 43, when the Hornans lived in Ohio), he and a few buddies would often invent new games in the basements of friends' houses.Inspired by the daily headlines in the newspapers, their masterpiece is a war game that takes up more than half of the basement.They designed games with tanks and cannons, as well as shell launch meters and range meters, and they even invented some methods to cover up the game graphics to simulate smoke screens.The game got complicated, says Horan: I still remember using the mimeograph machine in my dad's office to print a lot of war game graphics. (He Nan's father opened several soybean processing factories, and the business has been booming.) Horan said: The three of us like playing chess very much.Chess is a game with simple rules, but surprisingly, no two positions are ever exactly alike, as there are infinite possibilities.So we try to invent games with the same characteristics. And since then, he's been inventing games, no matter which method he uses, he says with a smile.I like to watch the situation evolve and I say, hey, did that really evolve from these assumptions?Because if my design is correct, if the underlying rules of the game's theme are gradually evolving instead of me controlling the overall development of the situation, then the ending will surprise me.And if I'm not surprised at all, then I'm not too happy, because I know it means that everything is going as I originally set it. Of course, this stuff is now called Emergence.However, long before Horan heard the term, his interest in emergent phenomena had sparked his lifelong love of science and mathematics, and no amount of scientific knowledge could satisfy him.He still remembers that when he was studying, I went to the library and borrowed all books about science and technology.Before the second grade of middle school, I had decided to become a physicist.What attracts him the most is not that science allows you to simplify everything in the universe into a few simple laws, but that science can tell you how a few simple laws can produce extremely rich behaviors.Horan said: I really enjoyed it.From one perspective, science and mathematics are the acme of reduction theory, but if you look at it as a whole, there are infinite and unexpected possibilities.On the one hand, science makes the universe comprehensible, but on the other hand, it makes the universe forever a mystery. Participate in the whirlwind project In the fall of 1946, Horan entered the Massachusetts Institute of Technology as an undergraduate student, and he soon discovered the same surprising quality in the computer.I don't know why, but I've always been fascinated by the thought process, and the fact that you can feed a little bit of information into a computer and tell it to do a bunch of things.It seems to me that you put so little in and get so much out. Unfortunately, the computer knowledge that He Nanneng learned at the beginning was almost lacklustre except for some sporadic second-hand information in the electrical engineering class.At that time, electronic computers were still a novelty, and most of the materials were included in confidential documents, and of course there were no computer courses to study.But one day, when Horan was browsing books in the library as usual, he saw a batch of loose-leaf lecture notes covered with simple thesis covers. Discussion content of a symposium held in the Electrical Engineering Department of the University of Pennsylvania in 1946.In order to calculate the range table of the cannon during the war, the University of Pennsylvania developed the first digital computer in the United States︱ENIAC.These notes are famous.It was the first time that digital computing was discussed in detail in a lecture, from what we now call computer architecture all the way to software.Among them, the speech also mentioned new concepts such as information and information processing, and explained a new mathematical art: programming.Horan immediately bought a transcript of the speech himself, and read it several times from beginning to end. In the fall of 1949, when Horan began his senior year and was looking for a thesis topic, he discovered the Whirlwind Project.It was an MIT research project to create a real|time computer fast enough to track air traffic.The Navy subsidizes the program with $1 million a year, a staggering amount at the time.Project Cyclone employed some seventy engineers and technicians and was the largest and most innovative computer project of its time.The Whirlwind was also the first computer to use magnetic core memory and an interactive display screen, thus leading to the development of computer networking and multiprocessing (multiprocessing, executing more than one program at a time).Since this was the first real-time computer, it also paved the way for future computer applications in air traffic control, industrial process control, and banking. But when Horan first heard about Project Whirlwind, it was just an experiment.I know there's a whirlwind project.The project isn't finished yet, and the computer is still being built, but it's ready to use.Somehow he just felt he had to get involved, so he started knocking on doors.He found a Czech astronomer named Zednek Kopal in the Department of Electrical Engineering who had taught him numerical analysis in the past.I persuaded him to chair my thesis committee, and I convinced the physics department to let the electrical engineering faculty chair my thesis committee, and then I went to the Cyclone Project people to let me refer to their operating manual; the operating manual is a confidential document. That period was probably the happiest I had been at MIT, he said.Kopal suggested that the subject of his dissertation be: write a program for the Whirlwind computer that could solve Laplace's equation.Laplace's equations describe many different physical phenomena, including the distribution of an electric field around any charged object and the vibration of a taut drumhead.Horan immediately proceeded. This is by no means the easiest senior thesis at MIT.At the time, no one had heard of a programming language like C, Pascal, or Evangelization, and in fact the concept of a programming language wasn't invented until 1950.Therefore, Horan had to write his program in mechanical language, that is, he had to encode computer instructions into numbers, and it was not ordinary decimal numbers, but hexadecimal numbers.He took longer to work on his dissertation than he had estimated, and ended up having to ask the school for twice as much time to complete it. However, he enjoyed it.I liked the logical nature of the process, he recalls, programming as characteristic of mathematics, where you take one step and it takes you to the next.In addition, writing programs for the Whirlwind Project let him discover that computers are not just fast computers. In the hidden column of numbers, he can imagine vibrating drums, swirling electric fields, or whatever he wants. s things.In flowing bits, he can create an imaginary universe, all that is required is to encode the appropriate laws into a computer code, and the rest will unfold naturally. IBM's big bet Since Horan's thesis was set as an exercise on paper from the beginning, he never had the opportunity to actually execute his program on the whirlwind computer.But this paper brought him another fruitful harvest, making him one of the few people in the United States who knew programming at that time.As a result, as soon as he graduated in 1950, he was immediately recruited by IBM. The timing couldn't have been better.At the time, IBM was designing the first commercial computer, the Defense Computer, at its Poughkeepsie, New York, factory, which was later renamed the IBM 701.For IBM at the time, the computer represented a big, mixed bet, and many conservatives believed that developing computers was a waste of money and that investing in improved punch machines was better.In fact, throughout 1950, IBM's product planning department insisted that the demand for computers in the United States would not exceed eighteen. IBM continued to develop defense computers only because an upstart named Tom Junior, who was IBM's aging president Thomas B. Watson, Sr. ) son and heir of course. However, the 21-year-old He Nan knew nothing about these inside stories, he only knew that he had arrived in a wonderful fairyland.Here I am, at such a young age, to get the most important position.I'm one of the very few people who know what the 701 computer is doing. IBM's program convenor assigned Horan to a seven-person logic-engineering team to design the instruction set and general organization of the new machine.The god of luck is favored again, and Horan can take advantage of this ideal opportunity to practice programming skills.After the first phase is completed, we have a computer prototype, and then we have to test the prototype in various ways.So engineers worked during the day, taking the machines apart and putting them back together at night whenever possible.A few of us started at eleven o'clock in the evening and ran our program all night to see if it worked. To a certain extent, it does work.Of course, by today's standards, the 701 computer looks like a relic from the Stone Age. Its control panel is full of various dials and switches. .This machine claims to have a memory capacity of 4,000 bytes (the memory capacity of computers sold on the market today is thousands of times that), and it only takes 30 microseconds to calculate the result of multiplying two numbers.This machine also has a lot of problems. On average, there will be errors within 30 minutes at most, so we have to run each program twice.Horan said.To make matters worse, the 701 stores data by creating spots of light on the surface of a special cathode ray tube, so Horan and other programmers had to tweak the logic to avoid constantly writing data to memory Otherwise, the surface charge of this part of the cathode ray tube will be increased, which will affect the surrounding data. Horan laughed: It's amazing that we can actually make this machine work.In fact, he felt that the flaws were not concealed.The machine has grown like a giant to us, and we thought it would be great to have time to experiment with our stuff on a fast machine. teach a computer to play chess There are so many things to experiment with.In those early, heady days of computers, when new ideas about information, control, automata, and so on were surging, who knew where the limit was?Almost every new venture may open up a new world.Beyond that, for pioneers like Horan, who tended to philosophize, this vast, unwieldy library of wires and vacuum tubes opened up new avenues for thinking.The computer may not be as powerful as the giant brain described in the Sunday newspaper supplement, in fact, from the structure and working details of the computer, it is not like a brain at all.However, if we look at it at a deeper level, maybe computers, like the human brain, are information processing devices.In this way, we can also understand thinking in the form of information processing. Of course, no one knew how to call such ideas artificial intelligence or cognitive science at the time.Still, the unprecedented attempt to program a computer has forced people to think more carefully about what it means to solve a problem.The computer is basically like an alien, you have to teach it everything: what is data?How is the data converted?What steps are needed to get from here to there?These questions quickly point to questions that have puzzled philosophers for centuries: What is knowledge?How can knowledge be acquired from sensory impressions?How is knowledge represented in the mind?How to revise the original knowledge with the help of experience?How can knowledge be used for reasoning?How do you turn decisions into actions? The answer was unclear at the time.In fact, it is still unclear now, but the questions are asked more clearly and precisely than in the past.The IBM computer development team, which has suddenly become the center of computer geniuses in the country, is bearing the brunt.Horan likes to recall that there was a group of people who would meet in the evening every two weeks or so to discuss poker games and Go.Among those who attended was a summer intern named John McCarthy, a graduate student at Caltech who went on to become one of the founding gurus of artificial intelligence. The other was Arthur Samuel, a soft-voiced electrical engineer in his forties. IBM specifically recruited him from the University of Illinois to assist in research on how to make reliable vacuum tubes.Samuel often accompanied Horan through the long nights of running the program.To be honest, Samuel has lost interest in vacuum tubes for a long time. He has been writing a computer program to play checkers for the past five years. Not only can he play chess, but after accumulating experience, he will learn to play better and better.It now appears that Samuel's chess-playing program is one of the milestones in the research of artificial intelligence. When he finally corrected the program in 1967, his chess-playing software has already caught up with the world champion.However, even in the era of the 701 computer, this program is already very good.Horan was very impressed, especially the computer chess player had the ability to adjust his tactics according to his opponent's strategy.In effect, the program devises a model of the opponent and uses this model to predict the best response.Although Horan couldn't explain it clearly at the time, he felt that this part of the computer player's functions captured the fundamental principles of learning and adaptation. how the brain learns Later, other things occupied He Nan's mind, and he temporarily put these thoughts behind him.At the time, he was busy modeling the inner workings of the brain for his own research project.This project originated in the spring of 1952. JCR Licklider, a professor of psychology at the Massachusetts Institute of Technology, visited the Porgy Puxi Laboratory and agreed to give a lecture on the most popular topic in the field of psychology at that time. Topic: The latest theory of learning and memory by neurophysiologist Donald O. Hebb of McGill University in Montreal. Ricklider explained: The problem is this, through a microscope, most of the brain appears to be a chaotic scene, each nerve cell is free to extend thousands of axons and dendrites, and thousands of other nerve cells. The axons and dendrites are haphazardly connected.However, this tightly connected neural network is obviously not composed randomly. It is no accident that a healthy brain can produce feelings, thoughts and actions consistently.Moreover, the brain is obviously not static. It can correct its behavior through experience, find ways to adapt to different situations, and it can learn.The question is, how exactly does the brain learn? 一九四九年，希伯已在他的名著行為組織（The Organization of Behavior）中，提出答案。他的基本想法是，假設腦子經常在突觸（synapse）上作些細微的改變。突觸是軸突和樹突的連接點，神經衝動經由突觸，從一個神經細胞傳遞到另一個神經細胞。希伯的假設很大膽，因為當時他還沒有確實的證據。但是，他辯稱，這些突觸的變化正是所有學習與記憶的基礎。例如，從眼睛而來的感官衝動會強化沿路所有的突觸，因此在神經網路上留下痕跡。當衝動來自耳朵或腦部其他的精神活動時，也會發生同樣的情況。希伯說，結果，原本隨機啟動的網路會迅速自我組織，經驗會經由正回饋而不斷累積；也就是說，強壯、經常被使用的突觸會愈長愈壯，而微弱、很少被使用的突觸會日益萎縮。常用的突觸強大到某個程度，記憶就被鎖定。這些記憶會轉而在腦中廣泛分布，每個記憶都對應於一個複雜的突觸形態，其中包含了數以千計或甚至百萬計的神經元。希伯是最早描述這種記憶形態、並稱之為結合論（connectionism）的少數人之一。 但是，還不止於此。里克萊德繼續解釋希伯的第二個假設：選擇性的強化突觸，會使腦子自我組織成一個個細胞集合也就是許多組神經元。流動的神經衝動會在其中自我強化，並繼續流動。希伯認為這些細胞集合是腦部的資訊基本單位，每一個細胞集合都對應於一個聲調、一線光、或一部分的想法。但是，這些集合在生理上並未彼此分開，而是重疊，每一個神經元都同時屬於好幾個細胞集合。因此，刺激一個細胞集合起反應，不可避免的會引發另外一個集合的反應，於是這些基本單位很快就會自我組織成較大的概念和複雜的行為。簡而言之，細胞集合是思想的基本量子（quantum）或基本單位。 人類第一次電腦模擬 賀南坐在觀眾席上，聽得目瞪口呆。有別於哈佛的史金納（BF Skinner）等行為學家極力推動的呆板的刺激／反應觀點，希伯討論的是心靈內部的活動。結合論的豐富性和永恆的驚奇使賀南激動不已，他迫不及待要作一些相關的研究。希伯的理論已打開了一扇探索思想本質的窗，他希望好好探究這個問題，他想看看細胞集合如何從一片混沌中自我組織和成長茁壯，他想看看細胞集合彼此互動的情形，看看它們如何融合經驗，而逐漸演化；他也想看看心靈本身的突現，更想看看所有這一切在沒有外界引導之下如何自然發生。 里克萊德的演講一結束，賀南就去找他在七○一電腦小組的主管羅徹斯特（Nathaniel Rochester），他說：好了，我們已經有電腦，我們來寫個神經網路模擬程式吧！ 於是，他們就這麼做了。他寫了一個程式，我寫了另一個程式，兩個程式形式不太一樣，我們把程式叫做：觀念啟迪者，我們不是自大！ 事實上，即使在四十年後的今天，當神經網路模擬程式早就變成人工智慧研究的標準工具時，當年的IBM觀念啟迪者仍然成就非凡。基本的概念在今日看起來仍很熟悉。在程式中，賀南和羅徹斯特把人工神經模擬為節點（node），也就是能夠對自己內部狀態有一些記憶的小電腦。他們把人工的突觸模擬為各種節點之間的抽象連結點，每一個連結點對應於突觸的強度，都會有一些重量。當網路獲得經驗時，他們也調整強度，以模擬希伯的學習法則。賀南、羅徹斯特和合作的研究人員，納入了許多今天大多數的神經網路模擬所沒有納入的神經生理學細節，包括像每一個模擬的神經元多快起反應，以及如果太常反應，多久會疲乏。 自然，他們在過程中碰到很多困難，不只是因為他們的程式是有史以來第一次模擬神經網路，也是人類第一次用電腦來模擬真實世界，而不只是計算數字或分析數據。賀南非常讚揚IBM的耐心，他和他的同事在電腦上花了數不清的小時來模擬網路，甚至還動用公款，出差到蒙特婁問希伯本人的意見。 但是最後，他們模擬成功了。賀南談起來，仍然難掩興奮之情。你可以用相同的神經元底質（substrate）開始，然後看著細胞集合形成，發生很多突現的現象。賀南、羅徹斯特和他們的同事在完成這個研究幾年後，在一九五六年發表了他們的研究結果，這是賀南第一篇在學術期刊上發表的論文。 非比尋常的哲學家 現在看來，希伯的理論和賀南自己的網路模擬可能對他未來三十年的思想形成，有舉足輕重的影響。但是，當時最直接的結果卻是驅使他離開IBM。 問題出在電腦模擬有一些必然的限制，而在七○一電腦上作模擬，限制尤其多。真正的神經網路上的細胞集合，有一萬個神經元分布在腦部的大部分區域，每一個神經元又有一萬個左右的突觸。但是，賀南等人在七○一電腦上所能模擬的網路只有一千個神經元，每個神經元只有十六個連結點，無論他們用各種程式設計技巧，想辦法加快速度，都只能得到這麼多。賀南說：我愈作實驗，愈覺得我們能測試的與我真正想看到的，差距實在太大了。 替代方案是以數學方法來分析網路。但是，結果也很困難。他嘗試的每個方法都踢到鐵板，成熟的希伯網路不是靠他在麻省理工學院學到的數學就可以應付的，儘管他比其他物理系學生都多修了很多數學課。對我而言，更精通數學似乎是更深入了解神經網路的關鍵。He said.所以，一九五二年秋天，帶著IBM的祝福和一紙每個月繼續為IBM作一百個小時顧問工作的合約，賀南來到安娜堡的密西根大學攻讀數學博士。 幸運又降臨他身上了。當然，密西根無論如何都是個不壞的選擇，不只是因為密大數學系在全美排名首屈一指，而且還有個美式足球隊對賀南而言，這是個重要的考慮因素。 但是，真正的幸運是，賀南在密西根大學碰到了勃克斯（Arthur Burks），一位非比尋常的哲學家。勃克斯專攻皮爾斯（Charles Peirce）的實用主義哲學，他在一九四一年拿到博士學位的時候，因為戰時的情況，不可能找到哲學教職，所以第二年，他在賓州大學的摩爾學院修了為期十週的電機課程，成為戰時工程師。結果，這是個愉快的選擇。一九四三年，他被摩爾學院網羅，加入最高機密的ENIAC電腦研究計畫。他在摩爾學院認識了馮諾曼，馮諾曼當時經常從普林斯頓高等研究院跑來指導這個計畫。在馮諾曼指導下，勃克斯也參與設計ENIAC電腦的第二代EDVAC。EDVAC是第一個能以程式的形式將指令電子化儲存的電腦。一九四六年，馮諾曼、勃克斯和數學家高士譚（Herman Goldstine）共同發表的論文電子計算工具之邏輯設計的初步討論（Preliminary Discussion of the Logical Design of an Electronic Computing Instrument），被視為現代電腦科學的基石之一。在這篇論文中，他們三位以精確的邏輯形式為程式的概念下定義，並且顯示藉著從電腦的記憶單位中取得指令、在中央處理單位中執行指令、然後再回頭把結果儲存在記憶體中這幾個步驟不斷循環之下，一般的電腦如何執行程式。這個馮諾曼架構一直到今天幾乎還是所有電腦的基礎。 當賀南於一九五○年代中期在密西根大學碰到勃克斯的時候，勃克斯的樣子瘦削而優雅，很像他一度嚮往的傳教士。勃克斯也是個熱心的朋友和絕佳的指導老師，他很快把賀南引進他的電腦邏輯小組。這個小組的理論學家專門研究電腦語言，並求證關於轉換網路的定理，而且試圖從最嚴謹而根本的層次來了解這個新機器。 勃克斯也邀請賀南參與一項由他協助籌畫的博士班研究計畫，主要是廣泛的探討電腦和資訊處理的隱含意義即所謂的通訊科學（communication science），後來的正式名稱叫電腦通訊科學。但是當時，勃克斯覺得他只是繼續馮諾曼未完成的志業，馮諾曼在一九五四年因癌症去世。馮諾曼認為電腦有兩種應用方式，一種是作為一般電腦，另一種就是作為自動機的一般理論基礎。勃克斯認為像這樣的研究計畫很適合那些不按牌理出牌的學生，賀南顯然就是其中之一。 賀南欣然同意。他們的想法是一方面開一些很困難的生物學、語言學、心理學課程，同時也提供很多標準課程，例如資訊理論。他們找不同領域的教授來講課，因此學生能夠把這些學問和電腦模型連起來。修過課的學生對於這些領域的基本理論會有深入的了解，例如主要的問題是什麼？為什麼這個問題這麼困難？電腦可以幫什麼忙？而不會只學到皮毛而已。賀南說。 玻璃珠遊戲 賀南樂於加入的其中一個原因是，他對數學已經完全失去興趣了。密西根大學數學系就像二次大戰後大多數的數學系一樣，服膺法國布巴奇學派（Bourbaki School）的理想，要求數學研究必須具備非人性的純淨和抽象。根據布巴奇的標準，甚至以世俗的繪圖方式說明定理背後的概念，都被視為魯鈍。他們的想法是要證明數學不需要任何詮釋，賀南說。但這不是他學數學的目的，他是想藉著數學來了解這個世界。 所以，當勃克斯建議賀南轉到通訊科學研究計畫時，他毫不遲疑就答應了，放棄了幾乎完成的數學論文，重新開始。也就是說，我的論文會更接近我想作的研究，他說，也就是神經網路的研究。（他後來決定的論文題目邏輯網路循環（Cycles in Logical Nets）是關於網路開關情形的分析。他在這篇論文中證明的許多定理，居然正是年輕的醫科學生考夫曼四年後在柏克萊奮力想證明的定理。）當賀南在一九五九年拿到博士學位時，他是通訊科學計畫第一個出爐的博士。 但是，這些都沒有使賀南忽略了當初他到密西根的目的。事實上，勃克斯的通訊科學計畫正是這類議題可以蓬勃發展的環境，包括：突現是什麼？思考是什麼？思考如何產生？有些什麼法則？系統適應的真正意義是什麼？賀南記下關於這些問題的一些想法，然後有系統的存檔在貼著Glasperlenspiel一號檔案、Glasperlenspiel二號檔案的檔案夾中。 Glas什麼？他笑著說：Das Glasperlenspiel!這是赫曼赫塞（Herman Hesse）的最後一部小說，於一九四三年流亡於瑞士時出版。賀南有一天在室友從圖書館借回來的書堆中發現了這本書，德文原文的意思是玻璃珠遊戲，英文版書名則稱遊戲高手（Master of the Game）。小說的場景是未來的世界，描繪一種原本是音樂家玩的遊戲：先在一種特別的玻璃珠算盤上設定主旋律，然後藉著撥上撥下玻璃珠而把各種對位旋律和變奏編織進去，經過一段時間，遊戲就會演變為極其複雜的樂器，並由一群有力的教士、知識分子所控制。最厲害的是你可以任選不同的主旋律組合，一點占星學、一點中國歷史、一點數學，然後想辦法把它們發展成好像音樂的主旋律。 當然，赫塞並沒有明講確實是怎麼做的，但是賀南並不在乎，玻璃珠遊戲比他過去所知道的任何事物都能捕捉到他所要追求的東西，這也正是西洋棋、科學、電腦或腦子之所以令他目眩神迷的地方。這個遊戲代表了他這一輩子都在追求的奧祕：我希望能夠從萬物中擷取不同的主題，然後看看把它們整合在一起時，會發生什麼事。 以數學掀起遺傳革命 另外有一本書也給了賀南很多啟發。有一天他在數學系圖書館瀏覽書籍時發現了費雪（RA Fisher, 1890︱1962）在一九二九年出版的遺傳學巨著天擇的遺傳理論（The Genetical Theory of Natural Selection）. 起先，賀南很著迷。從中學時代起，我就很喜歡閱讀關於遺傳和演化的書，他說。每一代生物都會再重新組合遺傳自父母的基因，你可以計算像藍眼睛或黑頭髮這些特質，會在後代出現多少次，這些想法都令他大感興趣。我一直想：哇！真是巧妙！讀這本書使我第一次了解在遺傳學領域除了代數之外，還可以運用很多其他的數學技巧。的確，費雪就用了很多複雜的數學概念，像微分、積分及或然率理論等。他的書以嚴謹的數學分析說明天擇如何改變基因分布，也因此為關於演化變遷的新達爾文理論奠定基礎。二十五年以後，這仍然是當代最先進的理論。 所以，賀南狼吞虎嚥的讀完這本書。我真是大開眼界，原來可以把我在數學課上學到的微積分、微分方程、以及其他的數學方法用來掀起一場遺傳學革命。一旦看到這點，我知道我無法放手了，一定要做一點事情。所以，我腦中一直盤旋著這些想法，不時把一些想法記下來。然而，儘管賀南很讚賞費雪的數學技巧，費雪應用數學的方式卻有些叫他困惑。事實上，他愈深入思索，就愈感困惑。 舉例來說吧，費雪對天擇的整個分析都著眼於一次分析一個基因的演化，彷彿有機體中的基因完全各自獨立，互不相干。事實上，在費雪的分析中，基因的作用完全是線性的。賀南說：我知道這絕對錯誤。除非有數十個或上百個形成眼睛結構的基因共同合作，單獨一個綠眼睛的基因絕對起不了什麼作用。就賀南所了解，每一個基因都必須在團隊中運作，任何理論如果沒有把這點考慮在內，都錯失了整個故事的關鍵。這也正是希伯在精神領域的研究中所一再強調的，希伯的細胞集合和基因有一點相像，細胞集合是思考的基本單位，但是如果單獨存在，細胞集合幾乎沒有任何價值。無論是要傳達一個音調、一束光、或命令肌肉抽動，唯一的方法是細胞集合彼此連結成更大的概念和更複雜的行為。 此外，還有一件事令賀南不解。費雪一直談到演化會達到穩定的均衡每個特定的物種都會發展為最理想的大小、牙齒銳利得恰如其分，總而言之，即達到能生存和繁殖的最佳狀態。費雲的論點和經濟學家對經濟均衡的定義如出一轍：一旦物種達到最佳狀態，任何的突變都會降低它自己的適應性，因此，天擇無法再形成改變的壓力。費雪的論點大半在強調：好了，因為以下的流程，這個系統會達到哈地︱威恩伯格的均衡（Hardy︱Weinberg equilibrium）但是，這聽起來這不像我心目中的演化論。 他回過頭去，重新閱讀達爾文和希伯的理論。不，費雪的均衡觀念一點也不像演化論，費雪的論調似乎是要達到某種純淨而永恆的完美。但是，在達爾文的理論中，隨著時間演進，物種的發展會愈來愈寬廣，愈來愈多樣，費雪的數學分析沒有提到這點。希伯探討的是學習，而不是演化，但是依稀可以看到相同的脈絡：當心靈從外界累積愈來愈多的經驗時，它會變得更豐富、更靈巧、更令人訝異。 演化是無盡的旅程 對賀南而言，演化和學習就像遊戲一樣，兩者都有一個和環境對抗的作用體，試圖贏得繼續往下發展所需要的一切。就演化而言，報酬就是生存，以及能把基因遺傳給下一代的機會；就學習而言，報酬是像食物、愉悅的感覺、情緒的滿足等等。在這兩種情況下，報酬正為作用體提供了適當的回饋，讓它們能改進自我表現。如果作用體想要具備適應能力，就必須維持能得到好報酬的策略，而放棄無效的策略。 賀南禁不住想到薩穆爾的西洋跳棋遊戲軟體，這個軟體正充分利用了這種回饋作用：當電腦棋手累積經驗，而且對對手了解更多之後，它會不斷更新戰術。現在，賀南才了解薩穆爾把重心放在遊戲是多麼有先見之明，這種遊戲的比喻似乎適用於任何的適應性系統。在經濟中，報酬就是金錢；在政治中，報酬就是選票。在某種層次上，所有的適應性系統基本上都一樣，也就是說，它們基本上都像西洋棋或西洋跳棋一樣，可能性的空間都大到超乎想像之外。作用體可以藉著學習，把遊戲玩得愈來愈好，但是如果想要找到最理想的狀態，找到遊戲的穩定平衡點，就會和我們下西洋棋一樣，只能在無限大的可能性中大海撈針。 難怪對他而言，均衡的概念和演化格格不入，甚至不像他十四歲時在地下室玩的戰爭遊戲。均衡暗指終點，但是對賀南而言，演化的本質在於旅程，在於無盡的驚奇。我愈來愈清楚我想要了解及我所好奇的是什麼東西，均衡絕對不是其中的重要部分。 在撰寫博士論文期間，賀南只能把這些想法暫擱一旁，但是他一拿到博士學位，就立定志向要把他的想法轉化成完整而嚴謹的適應性理論，而當時勃克斯已經邀請他繼續留在電腦邏輯小組作博士後研究。我相信如果我把遺傳適應看成長期的適應過程，把神經系統看成短期的適應過程，那麼兩者的一般性理論架構應該相同。為了要釐清自己腦中的想法，他甚至在一九六一年七月發表了一份四十八頁的技術報告，題目是適應性系統的非正式邏輯理論。 他也注意到有些同事頻頻皺眉。他們倒不見得有敵意，只是有些人認為他這套適應性理論的玩意兒聽起來荒誕不經，賀南為什麼不花時間作一些比較有收穫的研究呢？ 問題是，這真的只是胡思亂想嗎？賀南說。不過他欣然承認，如果換做是他，他也會懷疑。我作的研究沒有辦法照一般人熟悉的領域來歸類。它既不完全是關於硬體，也不純然是軟體的研究，在當時，這當然也還不能叫人工智慧，所以你沒有辦法以任何的標準尺度來下判斷。 找尋一組最理想的基因 他最不需要多費唇舌說服的人就是勃克斯。勃克斯說：我支持賀南。有一派邏輯學家很不以為然，覺得賀南的研究不是電腦邏輯小組該作的研究，他們的想法比較傳統。但是我告訴他們，這正是我們需要的研究，就爭取經費補助的角度而言，這研究的重要性和他們的研究不相上下。身為這個研究計畫的創辦人，勃克斯的話深具分量。抱著懷疑態度的人逐漸離開了這個研究小組。一九六四年，在勃克斯大力支持下，賀南得到了終身職。那幾年多虧了勃克斯當我的擋箭牌。賀南說。 的確，勃克斯的支持使賀南能特別著力於適應性理論的研究。一九六二年，他拋開了其他研究計畫，全力研究適應性理論。他特別決定要以不止一個基因為基礎，來探討天擇的問題。他這麼做不僅僅因為費雪在著作中假設基因互不相關，是困擾他多時的問題，同時，以多數基因為分析基礎也是避免均衡問題的關鍵。 賀南說，平心而論，當你討論互不相關的基因時，均衡說的確很有道理。例如，假定有一個物種有一千個基因，差不多就會像海藻一樣複雜。再假定，為了單純化，每一個基因只有兩種特性綠色相對於褐色，皺摺的葉片相對於平滑的葉片等等。天擇的過程要經過多少次試驗才能找到一組最理想的基因，能賦予海藻最佳的適應能力？ 如果你假定所有的基因真的各自獨立，每個基因你只需要試驗兩次就能發現哪一種特性比較好，然後一千個基因中的每一個基因你都要各試驗兩次，所以總共需要做二千次試驗，這並不算多。事實上，比較起來，這算是個小數目。因此，你可以預期這海藻很快就會達到最佳的適應能力，這時候，這個物種確實達到了演化的均衡狀態。 但是，現在當你假定海藻的一千個基因並非各自獨立時，看看會發生什麼事情。為了要確定真的找到最佳的適應能力，天擇過程必須檢查每個可能的基因組合，因為每一種組合都有不同的適應能力。當你檢查過所有組合時，數目不再是二乘以一○○○，而是二自乘一千次，也就是二的一千次方，或是十的三百次方，這數目大得連西洋棋的可能棋步都顯得微不足道。演化過程根本不可能試驗這麼多種可能性，而且不管電腦多發達，都辦不到！賀南說。這數字大得即使我們假設宇宙中的每個粒子都是一具超級電腦，從大霹靂之後就不斷在計算這些數字，所得到的解答離實際都還有一段距離。而且要記住，這還只是海藻而已，人類和其他哺乳類的基因差不多是海藻的一百倍，而大多數的基因都有不止兩種特性。 所以，這個系統又置身於可能性的無垠穹蒼，毫無找到一個最佳位置的希望。演化唯一能做的只是尋求改善，而不是尋求完美。但是，這正是他一直想解答的問題：如何做呢？要了解多基因的演化過程，顯然不只是把費雪的單基因方程式改為多基因方程式那麼簡單。賀南想了解的是，演化過程如何深入可能性的無垠穹蒼、發掘有用的基因組合，而不需要搜尋其中的每一寸空間？ 世界為什麼如此架構？ 事實上，主流的人工智慧學者早就很熟悉一種類似可能性一發不可收拾的現象。例如，在匹茲堡的卡內基技術學院（現在叫卡內基美崙大學），紐威爾（Allen Newell）和西蒙（Herbert Simon）從一九五○年代中期開始，就在進行一個畫時代的研究研究人類如何解決問題。他們要求實驗對象在絞盡腦汁玩拼圖和遊戲的時候，說出他們的想法。紐威爾和西蒙的結論是，問題解決是在有廣大可能性的問題空間中一步接著一步做心智搜尋。每一個步驟都由經驗法則所引導：如果情況是這樣，那麼該採取那個步驟。紐威爾和西蒙把理論架構為一般問題解答者（General Problem Solver）的程式，而且以這個程式來解原先的拼圖和遊戲，顯示他們的問題空間方式能充分複製人類的推理風格。他們的觀念已經成為人工智慧領域的金科玉律，而一般問題解答者也成為人工智慧發展史上最有影響力的電腦程式。 但是賀南仍然存疑。他倒不是認為紐威爾和西蒙關於問題空間和經驗法則的觀點錯誤，事實上，賀南拿到博士學位後不久，就建議邀請他們兩位到密西根大學來開人工智慧的課，從此他和紐威爾就成為好友和知識上的諍友。不，只不過是他們的觀點無法幫助他解決生物演化的問題。演化論的整個觀點中沒有經驗法則、沒有任何引導，一代接一代的物種是藉著突變和兩性基因的隨機組合，來探測可能性的廣大空間；簡單的說，它們靠的是嘗試和錯誤。而且，交替的世代並不是按部就班的探索著各種遺傳組合的可能性，而是各種嘗試齊頭並進，每一個個體都有一組稍微不同的基因，嘗試稍微不同的可能組合。 但是，儘管有這些差異，儘管演化花的時間比較長，演化過程所產生的創意和驚奇與人類的精神活動並無二致。對賀南而言，這意味適應性的真正大一統原理隱藏在更深的層次中，但是，到底在哪裏呢？ 起初，他只是直覺的認為：某些基因組能一起運作得很好，並形成統一而相互強化的整體。例如一群能指揮細胞如何從葡萄糖分子汲取能量的基因，或是一群控制細胞分裂的基因，或是指導如何組成身體組織的基因群。這種情形很類似希伯關於腦部學習的理論：一組相互共鳴的細胞集合可能會形成像汽車的相關觀念，或是像舉起手臂的協調性動作。 但是賀南愈思考這個統一、相互強化的基因群的概念，就愈覺得其中奧妙無窮。類似的例子幾乎隨處可見，電腦程式中的副常式、官僚體系中的部門、西洋棋局戰略中的小戰術皆是。更重要的是，你可以在每一個組織層次有同樣的發現。如果某個群體有足夠的一致性和穩定性，那麼就通常成為更大群體的基本單位（building block）。細胞組成組織，組織組成器官，器官組成有機體，有機體組成生態系。賀南想，這正是突現的觀念：一個層次的基本單位組合後，形成更高層次的基本單位。這是世界的基本組織法則，幾乎每個複雜適應性系統都出現這種特性。 但是，為什麼呢？這種階層式、由基本單位層層上推的結構，簡直像空氣一樣平常而普遍，因此我們從來不會多加思索。但是，當你深入思考這個問題的時候，卻發現需要有個解釋：世界為什麼如此架構呢？ 先分化，後征服 事實上，原因很多。電腦程式設計師都會把問題分成幾個副常式來處理，因為簡單的小問題要比繁雜的大問題容易解決，這正符合先分化、後征服的古老智慧。像鯨魚和紅木等巨大生物是由不計其數的小細胞所組成，正因為這些細胞比較早出現。當五億七千萬年前，巨大的動植物開始在地球出現的時候，要從已經存在的單細胞生物經由天擇過程而形成巨大生物，顯然要比從頭產生點點滴滴的原形質容易多了。通用汽車（General Motors）把公司分成不計其數的部門，是因為最高主管不想讓五十萬名員工都直接向他報告，否則他的時間根本不夠分配。事實上，就像西蒙在一九四○和五○年代對商業組織的系列研究中指出，設計精良的階層組織是既能完成工作、又不會讓一個人被會議和備忘錄所淹沒的最好方法。 然而賀南愈想愈覺得最重要的理由隱藏在更深的層次：階層式、由基本單位組成的結構能改變一個系統學習、演化和適應的能力。想想看包含像紅色、汽車和馬路等概念的認知基本單位吧！一旦像這樣的一組基本概念藉著經驗累積而精煉、修正錯誤，這組概念通常能調整並重組成許多新的概念，例如：路邊一輛紅色的汽車，這當然要比一切從頭開始有效率多了。 這個事實提示了一個適應的全新機制，適應性系統不是按部就班的在可能性的廣大空間中搜尋，而是重組基本單位，採取大躍進式的突破。 賀南最喜歡以警方繪相師來說明這個觀念。在過去尚未應用電腦的年代，警方如果需要根據目擊證人的描述來畫出嫌疑犯的相貌，他們的做法是，先把臉部分成十個基本單位：髮型、前額、眼睛、鼻子等，然後繪相師在許多紙片上描繪出各個部分的不同形狀，以供選擇，例如十種不同的鼻子、十種不同的髮型，所以總共有一百張紙片。有了這些基本形貌的紙片，繪相師會根據目擊者的描述，把適當的紙片拼湊在一起，很快就能畫出嫌疑犯的肖像。當然，繪相師不可能畫出所有可想像的容貌，但是這麼做雖不中亦不遠矣，因為藉著替換這一百張紙片，繪相師能畫出一百億張不同的臉孔，即使在廣大的可能性空間中，這都有相當的代表性。所以，如果我能找到發現基本單位的方法，這些組合都將為我所用，而不是變成我的阻礙。我能夠利用少數的基本單位，來描述許多複雜的事物。賀南說。 而這正是解開多基因謎團之鑰。演化的過程之所以會有割愛及嘗試錯誤，並非為了要創造出一個