The legend of artificial intelligence can be traced back to ancient Egypt, but with the development of electronic computers since 194 1, technology has finally created machine intelligence. The word "artificial intelligence" was first put forward at the Dartmouth Society in 1956. Since then, researchers have developed many theories and principles. The concept of artificial intelligence has also expanded. In its short history, the development of artificial intelligence is slower than expected, but it has been advancing. Since it appeared 40 years ago, many AI programs have appeared, which have also influenced the development of other technologies. An invention of 194 1 completely changed all aspects of information storage and processing. This invention, which appeared in the United States and Germany at the same time, is an electronic computer. The first computer occupied several large air-conditioned rooms. This is a nightmare for programmers: thousands of lines of code have to be set just to run a program. 1949, the improved computer that can store programs makes it easier to input programs, and the development of computer theory produces computer science, which finally promotes the emergence of artificial intelligence. The invention of computers that electronically process data provides a medium for the possible realization of artificial intelligence.

Although computers provided the necessary technical foundation for AI, it was not until the early 1950s that people noticed the connection between human intelligence and machines. Norbert wiener was one of the first Americans to study feedback theory. The most common example of feedback control is thermostat. It compares the collected room temperature with the required temperature and responds by turning the heater up or down. So as to control the ambient temperature. The importance of this research on feedback loop lies in: Wiener pointed out theoretically that all intelligent activities are the result of feedback mechanism and can be simulated by machines. This discovery had a great influence on the development of early AI.

At the end of 1955, NEWELL and SIMON made a program called Logic Theorist. This program is considered by many people to be the first AI program. It uses a tree model to represent each problem. Then choose the one that is most likely to get the correct conclusion to solve the problem. The influence of logic experts on the public and the field of AI research has made it an important milestone in the development of AI. 1956, john mccarthy, who is considered as the father of artificial intelligence, organized a community. Many experts and scholars interested in machine intelligence got together and had a discussion for a month. He invited them to attend the Dartmouth summer seminar on artificial intelligence in Vermont. Since then, this field has been named artificial intelligence. Although the Dartmouth Society was not very successful, it did focus on the founders of AI, which laid the foundation for future AI research.

Seven years after Dartmouth Conference, AI research began to develop rapidly. Although this field has not been clearly defined, some ideas in the meeting have been reconsidered and used. Carnegie Mellon University and Massachusetts Institute of Technology began to set up AI research centers. The research faces new challenges: the next step is to establish a system that can solve problems more effectively, such as reducing the search among logic experts; There is also the establishment of a self-study system.

1957 tested a new program, the first version of the universal problem solving machine (GPS). This program was developed by the same group of logic experts. GPS extends Wiener's feedback principle and can solve many common-sense problems. Two years later, IBM set up an AI research group. Herbert Grenett spent three years writing a program to solve geometric theorems.

When more and more programs appeared, McCarthy was busy making breakthroughs in the history of AI. 1958, McCarthy announced his new achievement: LISP language. LISP has been used to this day. LISP means list processing, which was quickly adopted by most AI developers.

1963, MIT received a $2.2 million grant from the U.S. government to study machine-aided identification. The grant came from the Advanced Research Projects Agency (ARPA) of the Ministry of National Defense, which ensured that the United States was ahead of the Soviet Union in technological progress. This plan has attracted computer scientists from all over the world and accelerated the development of AI research. artificial intelligence

It is a very attractive field for human beings to create a machine brain (artificial intelligence) parallel to the human brain with human wisdom, and human beings have struggled for many years to realize this dream. From the perspective of language researchers, it is quite difficult for machines to communicate freely with people, and it may even be an unanswered question. Human language and human intelligence are so complex that our research has not touched the edge of the extension of its guiding essence. A large number of programs appeared in the following years. One of them is called SHRDLU. SHRDLU is a part of the micro-world project, including micro-world research and programming (for example, there are only a few geometric shapes). At MIT, researchers led by Marvin Minsky found that computer programs can solve the spatial and logical problems of small-scale objects. Other students, such as those who appeared in the late 1960s, can solve algebra problems, and Mr. Wang can understand simple English sentences. The results of these projects.

Another development in 1970s is expert system, which can predict the probability of a solution under certain conditions. At that time, because of the huge computer capacity, expert systems could extract rules from data. Expert system is widely used in the market. In the past ten years, expert systems have been used to predict the stock market, help doctors diagnose diseases and guide miners to determine the location of mineral deposits. The ability of expert system to store rules and information makes all this possible.

In 1970s, many new methods were used in the development of artificial intelligence, such as Minsky's structural theory. In addition, David image put forward a new theory of machine vision, for example, how to distinguish an image by basic information such as shadow, shape, color, boundary and texture. By analyzing this information, we can infer what the image may be. At the same time, another achievement is preface language. It was put forward by 1972. In 1980s, artificial intelligence developed faster and entered more commercial fields. 1986, the sales of AI-related software and hardware in the United States reached 425 million dollars. Expert system is especially needed because of its practicability. Companies like Digital Electric Company use XCON expert system to program VAX mainframes. Dupont, General Motors and Boeing also rely heavily on expert systems. In order to meet the needs of computer experts, some companies that produce expert systems to assist in making software, such as TEKNOWLEDGE and INTELLICORP, have been established. In order to find and correct the mistakes in the existing expert system, other expert systems have been designed. People began to feel the influence of computer and artificial intelligence technology. Computer technology no longer belongs to a small group of researchers in the laboratory. Personal computers and numerous technical magazines have brought computer technology to people. With a foundation like american association for artificial intelligence, there is a wave of researchers entering private companies because of the need of AI development. More than 150 companies, such as DEC (which employs more than 700 employees in the field of artificial intelligence research), spent 10 billion dollars on the internal artificial intelligence development team.

Other fields of artificial intelligence also entered the market in the 1980s. One of them is machine vision. The results of Minsky and MARR are now used in cameras and computers on the production line for quality control. Although these systems are still in the initial stage, they have been able to distinguish the shape of objects with black and white. By 1985, there were more than 100 companies producing machine vision systems in the United States, with sales of * * * reaching $80 million.

However, the 1980s were not all good years for the artificial intelligence industry. In 1986-87, the demand for artificial intelligence systems declined, and the industry lost nearly 500 million dollars. Two companies, such as TEKNOWLEDGE and INTELLICORP, lost more than $6 million, accounting for about one-third of the profits. The huge losses forced many research leaders to cut their funds. Another disappointment is the so-called smart truck supported by the Defense Advanced Research Projects Agency. The purpose of this project is to develop a robot that can complete many battlefield tasks. Due to the defects and hopeless success of the project, the Pentagon stopped funding the project.

Despite these setbacks, AI is still slowly recovering. Japan has developed new technologies, such as fuzzy logic pioneered by the United States, which can make decisions under uncertain conditions. There is also a neural network, which is regarded as a possible way to realize artificial intelligence. In short, AI was introduced into the market in 1980s and showed practical value. To be sure, it will be the key to the 265,438+0th century. The intelligent equipment of China's army stood the test of war in operation desert storm. Artificial intelligence technology has been used in missile systems, early warning displays and other advanced weapons. AI technology has also entered the family. The increase of intelligent computers has aroused public interest. Some application software for Mac and IBM compatible computers, such as voice and text recognition, are available. Using fuzzy logic and artificial intelligence technology simplifies the camera equipment. The greater demand for artificial intelligence related technologies has prompted new progress. Artificial intelligence has inevitably changed our lives and will continue to do so. A popular definition of artificial intelligence, which is also an earlier definition in this field, was put forward by john mccarthy at the Dartmouth meeting in 1956: artificial intelligence is to make the behavior of machines look like human intelligent behavior. But this definition seems to ignore the possibility of strong artificial intelligence (see below). Another definition means that artificial intelligence is the intelligence displayed by artificial machines. Generally speaking, the definition of artificial intelligence can be divided into four categories, that is, machines think like people, act like people, think rationally and act rationally. Here, "action" should be broadly understood as taking an action or a decision to take an action, not a physical action.

Strong artificial intelligence (bottom-up AI)

From the viewpoint of strong artificial intelligence, it is possible to create intelligent machines that can really reason and solve problems-solving such machines can be regarded as conscious and self-conscious. There are two kinds of strong artificial intelligence:

Humanoid artificial intelligence, that is, machine thinking and reasoning, is just like human thinking.

Non-human artificial intelligence, that is, machines produce completely different perceptions and consciousness from people and use completely different reasoning methods from people.

Weak artificial intelligence (top-down AI)

From the point of view of weak artificial intelligence, it is impossible to create an intelligent machine that can really reason and solve problems. These machines just look intelligent, but they don't really have intelligence and won't have autonomous consciousness.

The mainstream scientific research focuses on the field of weak artificial intelligence, and it is generally believed that this research field has made considerable achievements. The research of strong artificial intelligence is at a standstill.

Philosophical Debate on Strong Artificial Intelligence

The term "strong artificial intelligence" was originally coined by John Rogers Searle for computers and other information processing machines. It is defined as:

"From the viewpoint of strong artificial intelligence, computers are not only tools to study people's thinking; On the contrary, the computer itself is thinking as long as it runs the appropriate program. " Mind, brain and program. Behavior and Brain Science, Volume 3, 1980) This refers to guiding computers to engage in intelligent activities. The meaning of intelligence here is vague and uncertain. The following is an example. When you use a computer to solve problems, you must know clear procedures. However, even when people don't know the procedure, many people try to solve the problem skillfully according to the Heu-risk method. For example, recognizing words, graphics and sounds, the so-called cognitive model is an example. Moreover, the ability is improved due to learning, inductive reasoning and analogy-based reasoning. It is also an example. In addition, although the solution procedure is clear, it will take a long time to implement. For such problems, people can find quite good solutions in a short time, such as competitive competitions. In addition, the computer can't understand its meaning without giving sufficient logical and correct information, and people can only grasp its meaning by giving insufficient and incorrect information according to appropriate supplementary information. Natural language is an example. Processing natural language by computer is called natural language processing.

The debate about strong artificial intelligence is different from the debate about monism and dualism in a broader sense. The point of the debate is: if the only working principle of a machine is to convert coded data, is it thinking? Shearer thinks this is impossible. He cited an example of a Chinese room to illustrate that if a machine only converts data, and the data itself is a coded expression of something, then it is impossible for a machine to understand the data it processes without understanding the corresponding relationship between this code and this actual thing. Based on this argument, Hiller believes that even if the machine passes the Turing test, it does not necessarily mean that the machine really has thinking and consciousness like people.

Some philosophers hold different views. DANIEL C. DENNETT thinks that man is just a machine with a soul in the book Interpretation of Consciousness. Why do we think that people can have intelligence, but ordinary machines can't? He believes that it is possible for a data conversion machine like the above to have thinking and consciousness.

Some philosophers believe that if weak artificial intelligence is realizable, so is strong artificial intelligence. For example, simon blackburn said in his introductory philosophy textbook that thinking that a person looks "smart" doesn't mean that he is really smart. I can never know whether another person is really as smart as me, or whether she/he just looks smart. Based on this argument, since weak artificial intelligence thinks that machines can look smart, it cannot completely deny that machines are really smart. Blackburn thinks this is a subjective problem.

It should be pointed out that weak artificial intelligence and strong artificial intelligence are not completely opposite, that is to say, even if strong artificial intelligence is possible, weak artificial intelligence is still meaningful. At least, what computers can do now, such as arithmetic operations, was considered very intelligent more than 100 years ago.