In addition, if you are writing a thesis or writing about advanced topics, for example, intelligent robots in the future (used in electronics industry in manufacturing, such as electronic boards, capacitors, various chips, etc.), if you want to have such development in the future, you should learn about electronic agents, and single-chip computers are good.
Intelligent robot data
Robots have been widely used in many fields of production and life, and can be divided into three levels according to their intelligence level.
One is the industrial robot, which can only work rigidly according to the program given to it by people. No matter what the external conditions change, it can't adjust the program, that is, the work it does. If it wants to change the work it does, it must be changed by people, so it is unintelligent.
Second, the primary intelligent robot. Unlike industrial robots, it has the ability to feel, identify, reason and judge like human beings. It can modify the program itself within a certain range according to the changes of external conditions, that is, it can adapt to the changes of external conditions and make corresponding adjustments to itself. However, the principle of modifying the program is stipulated by people in advance. This primary intelligent robot already has certain intelligence, although it has no automatic planning ability, but it has also begun to mature and achieve practicality.
Third, the advanced intelligent robot. Like the primary intelligent robot, it has the ability of feeling, recognition, reasoning and judgment, and can also modify the program within a certain range according to the changes of external conditions. The difference is that the principle of modifying the program is not stipulated by people, but the robot itself obtains the principle of modifying the program through learning and summing up experience. Therefore, its intelligence is higher than that of the primary intelligent robot. This robot has certain automatic planning ability. Can arrange their own work. This kind of robot can work completely independently without human care, so it is called advanced autonomous robot. This kind of robot is also beginning to be practical.
intelligent robot
We understand the so-called intelligent robot in a broad sense, which gives people the deepest impression that it is a unique "living thing" that controls itself. In fact, the main organ of this self-control "living thing" is not as subtle and complicated as a real person.
Intelligent robots have various internal information sensors and external information sensors, such as vision, hearing, touch and smell. In addition to receptors, it also has effectors as a means to act on the surrounding environment. This is muscle, or self-synchronizing motor, which makes hands, feet, long nose, tentacles and so on move.
Intelligent robot is called intelligent robot because it has a fairly developed "brain". What works in the brain is the central computer, which has direct contact with the people who operate it. Most importantly, such a computer can perform actions arranged according to the purpose. Because of this, we say that these robots are real robots, although their appearance may be different.
We call this kind of robot automatic control robot, in order to distinguish it from the robot mentioned earlier. It is the result of cybernetics, which advocates the fact that life and non-life purposeful behavior are consistent in many ways. As an intelligent robot maker said, robots are functional descriptions of a system. In the past, such systems could only be obtained from the growth of living cells, but now they have become things that we can make ourselves.
Intelligent robot can understand human language, talk to the operator in human language, and form a detailed model of the external environment-actual situation that enables it to "survive" in its own "consciousness". It can analyze the situation, adjust its actions to meet all the requirements put forward by the operator, draw up the desired actions, and complete these actions under the condition of insufficient information and rapid changes in the environment. Of course, it is impossible to make it exactly the same as our human thinking. However, some people still try to build a kind of "micro-world" that computers can understand. For example, the robot made by Vinograd in the artificial intelligence laboratory of MIT. This machine tries to learn to play with building blocks completely: the arrangement, movement and geometric structure of building blocks, to the extent of a child. This robot can walk alone and pick up certain objects, can "see" things and analyze what it sees, can obey instructions and answer questions in human language. More importantly, it has the ability of "understanding". To this end, someone once said at an artificial intelligence academic conference that in less than ten years, we have improved the intelligence of electronic computers by 10 times; As Vinograd pointed out, computers have obvious artificial intelligence components.
However, although robot artificial intelligence has made remarkable achievements, cybernetic experts believe that the limit of its intelligence level has not been reached. The problem lies not only in the insufficient operation speed of the computer and the few types of sensory sensors, but also in other aspects, such as the lack of design ideas for programming the rational behavior of robots. You think, now even the thinking process of people in solving the most common problems has not been deciphered. What will happen to human intelligence? This kind of understanding process is progressing very slowly. How can we master the laws to make the computer "think" faster? Therefore, the problem of not knowing human beings has become a stumbling block on the road of robot development. In recent years, the subject of making intelligent robots that "live" in an unstable environment has made people deeply study the cognitive and self-cognitive processes that occur in biological systems, animals and human brains. As a result, the theory of hierarchical adaptive system appears, which is developing effectively. As the theoretical basis of organizing intelligent robots to conduct purposeful behaviors, how does our brain control our bodies? From a purely mechanical point of view, our body also has more than 200 degrees of freedom. When we are doing complex actions such as writing, walking, running, swimming and playing the piano, how does the brain give orders to every muscle? How can the brain process so much information in the shortest time? Our brains are not involved in these activities at all. The brain-our central information processor "disdains" to take care of this. It doesn't supervise all the moving parts of our body at all, and the detailed design of the action is carried out at a much lower level than the cerebral cortex. This is much like programming in a high-level language. As long as you point out "a set of numbers from1~ 20 with an interval of one", the robot itself will input this set of instructions into the detailed operating system. The most obvious thing is that the most obvious instructions such as "retract your hand as soon as you touch a hot object" have been issued even before the brain realizes it.
It is more cost-effective, economical and effective to allocate a big task among several cortex than to control organs and specify the necessary actions for each element of the system. When solving major problems, such a centralized brain will be too complicated, not only for the skull, but even for the whole human body. When completing some complicated actions of one kind or another, we usually break them down into a series of common small moves (such as getting up, sitting down, taking the right foot and taking the left foot). Teaching children all kinds of actions can be summed up in forming and consolidating the corresponding small moves in the children's "memory". In the same way, the process of perception is organized in this way. Perceptual image-this is a fixed sequence or combination of auditory, visual or tactile pulses (horses, people), or both.
Learning ability is another general principle of organizational control in complex biological systems, and it is the ability to adapt to the living environment that has changed in a fairly wide range and has not been known before. This adaptive ability is not only inherent in the whole body, but also inherent in individual organs and even functions of the body. This ability is irreplaceable when the same problem should be solved many times. It can be seen that this phenomenon of adaptability plays an extremely important role in the purposeful behavior of the whole biological world. At the beginning of this century, the zoologist Thorndike conducted the following animal experiments. Firstly, a T-shaped maze with three small platforms is designed. The experimental animals are located on the small platforms at the bottom of the letter T, and the bait is located on the small platforms at both ends of the letter T beam. This animal can only make the following two choices, that is, after running to the fork, it can turn to the small platform on the left or right. However, something unpleasant lurks on the way to the bait: electrodes are installed on both sides of the corridor, and voltage is input into these electrodes at a certain fixed frequency, so animals running past these electrodes are stimulated by pain-the outside world sends out punishment signals. The bait waiting for animals on the other side of the platform is a signal of external reward. In the experiment, if the stimulation probability in one corridor greatly exceeds the stimulation probability in another corridor, then the animal will naturally adapt to the external situation: after running several times repeatedly, the animal runs towards the corridor with low stimulation probability and less pain. Thorndike experimented with rats the most. Rats, for example, choose a safer route more quickly, and confidently choose a safer route with little difference in punishment. Other animals in the experiment show this with different degrees of adaptability, but this ability is possessed by all kinds of animals participating in the experiment.
The problem of controlling robots lies in simulating animal movements and human adaptability. Establish the level of robot control-first of all, implement the distribution of perception function, information processing function and control function at each level of robot and among subsystems. The third generation robot has the ability of large-scale processing. In this case, the completely unified algorithm of information processing and control is actually inefficient or even useless. Therefore, the emergence of hierarchical adaptive structure is first to improve the quality of robot control, that is, to reduce the level of uncertainty and increase the rapidity of action. In order to play the role of each level and subsystem, the amount of information must be greatly reduced. Therefore, people can complete the task under the condition that the uncertainty is greatly reduced.
In a word, the development of intelligence is an important feature of the third generation robot. People decide which robot they belong to according to their intelligence level. Some people even divide robots into the following categories: controlled robots-"zero generation" robots, which do not have any intellectual performance, are manipulators controlled by people; Robots that can be trained-the first generation robots, have memories, are operated by people, and the action plans and programs are specified by people. They just remember (the ability to receive training) and reproduce them; Feeling robot-the robot remembers the plan arranged by people, and then calculates the specific program of action according to the external data (feedback); Intelligent robot-after a person specifies a target, the robot makes an operation plan independently, determines the action program according to the actual situation, and then turns the action into the movement of the operating mechanism. Therefore, it has a wide range of sensory systems, intelligence and simulation devices (the surrounding situation and itself-the robot's consciousness and self-consciousness).
How to become smart
Experts in artificial intelligence point out that computers should not only do what humans have assigned them to do, but also solve many things in the best way by themselves. For example, the whole program of an ordinary computer that accounts for electricity or engages in banking business is to complete the instruction list accurately, while the computers in some scientific research centers "think" about the problem. The former works quickly, but it has no intelligence; The latter stores more complicated programs, and the computer is full of information, which can imitate many human abilities (in some cases, it even exceeds our abilities).
In order to study this problem, many scientists have exhausted their life's efforts. For example, during World War II, British mathematician Turing invented a machine, which became the originator of modern robots. This is a system to decipher enemy communications. Later, Turing spent his whole life fantasizing about making a learning and intelligent machine. In Princeton in 1945 10, another famous mathematician von Naiman designed something called "artificial brain". He and his students are fanatics of psychology and neurology. In order to create a mathematical simulator of human behavior, they suffered many failures and finally lost confidence in the possibility of creating "artificial intelligence". Early computing devices were too bulky and the components were too large, which made Von Naiman unable to solve the problem of how to replace tiny nerve cells with these components. At that time, the human brain was regarded as something woven by interconnected neurons, so it could be imagined as a computing device, in which information was circulated instead of energy. Scientists thought, if we accept such a comparison, why can't we invent a system that makes information pass through and generate intelligence?
So they put forward various theories of artificial thinking. For example, physicist Mark proposed a method to make robots think with binary or binary logic elements. This method is considered to be very simple. 1956 scientists held the first large-scale seminar, and many experts and scholars advocated using the term "artificial intelligence" as the name of the research object. Two unknown researchers, Neville and Simon, put forward extraordinary ideas. They studied the way two people communicate with the help of signal devices and button systems. This system will decompose the behavior of these two people into a series of simple actions and logical actions. Because there are two large computers in the workplace of these two researchers, they often amuse themselves by putting their own experiments upside down: inputting simple logical rules into the computer, so that it can develop the ability of complex reasoning. This is really a genius idea; The computer program not only works, but also finds a new theorem with its help. The proof of this theorem is completely unexpected and much more beautiful than all previous proofs. Neville and Simon discovered a laid-down principle, that is, it is not necessary to understand the human brain to give robots intelligence. What needs to be studied is not how our brain works, but what it does; It is necessary to analyze human behavior and study the process of acquiring knowledge by human behavior, without exploring the theory of neural network. Simply put, we should focus on psychology, not physiology.
Since then, researchers have begun to move in the above direction. However, they have been arguing about how to make the computer "think".
One school of researchers takes logic as the research point and tries to divide the reasoning process into a series of logical judgments. The computer goes from one judgment to another and draws a logical conclusion. Like the well-known syllogism: "All animals will die; Small thorn singing is an animal, so small thorn cabin will die. " Can computers achieve the same level of intelligence as young children? Scientists have two opposite views on this issue. Dreyfus, a philosophy teacher at Berkeley, took the lead in vehemently opposing the "artificial intelligence school". He said that the theory of artificial intelligence is alchemy. He believes that people's thinking can't be programmed at any time, because there is the simplest truth: people come to know the world together with their own bodies, and people are not just composed of intelligence.
He further exemplified: the computer may know what a restaurant means, but it will never know whether the guests eat with their feet or whether the waitress flies to the table or climbs to the feet; In a word, computers will never have enough knowledge to know the world. But Minsky, a researcher at the Massachusetts Institute of Technology, disagreed with dreyfus. He thought that the intelligence of robots was infinite. His explanation of "artificial intelligence" is that it is a science, which makes machines do such a thing. If this kind of thing is done by people, it will be considered as an intelligent behavior. Minsky is also a physicist and mathematician, and he also studies psychology, sociology and neurology. He pointed out that artificial intelligence is a new category of psychology, which simulates the nature of human thinking by means of experiments and computers. He thinks that the computer he studied is a brand-new science; Of course, the machine is not a person. It never has the kind of happy or painful emotional experience of people, but is keen on mastering pure knowledge. For example, people can input the concept of "water" into the computer: water is a liquid with a flat surface; If you pour it from one container into another, its quantity will not change; Water can leak from containers with holes, wet clothes, and so on. However, after getting the most general information about water, it tries its best to answer a very important question: "What if the glass full of water is tilted?" The computer shows a glass tilted to a horizontal position on its screen. Although the computer knows the law of gravity, it still stubbornly shows on the screen that the glass is tilted, but the liquid just doesn't flow out. Computers will never experience the unpleasant feelings of people whose clothes are wet from painful but beneficial experiences.
So a psychologist named Schenk is leading a group of scholars to study this interesting topic: let computers learn to read and summarize the contents of reading materials and answer relevant questions; Let the computer learn several human languages and translate them to each other; Let the computer learn to talk, learn the art of argument, recite words ...
Talk to people
Yale University in the United States once designed such a computer: it didn't store fixed statements prepared in advance in its memory. It compiled its own answers, demonstrated and "thought", which was somewhat like a human being. Relying on psychology and information theory, scientists have put forward a topic for themselves that surprises the world: to study people's thinking mode and behavior clearly, and then to simulate it artificially.
When we talk about the word "artificial intelligence", we will immediately associate it with something unreal. The appearance of this word makes many people nervous: robots are the same as people, so where will human beings go! Some people are desperately defending the last bastion of human beings from the harm and infringement of robots. The problem is complicated because the word has not yet formed a unified definition. Minsky said: "This is a science, which makes robots do such a thing. If it is done by people, it will be considered as intelligent behavior." This kind of playful definition is of little use, and sometimes it simply leads researchers into the swamp of practical formalism. Another researcher named Turing put forward a test method of artificial intelligence: If human can't guess what the computer will say when talking to him-I don't know what it will say, then the computer has reached the level of human intelligence. His remark once caused a sensation and added a lot of fuss to the academic circles. In order to eliminate computer speech problems, such a conversation is best carried out by telex. For many control experts, a lot of work has been done to reach Turing's level. Countless electronic talkers of all kinds have come out.
In the late 1960s, Weisenbaum, an American cybernetic expert and MIT teacher, compiled several programs, the main purpose of which was to meet Turing's test conditions-to confuse the nitpicking technical experts. This approach is based on specious dialogue. In this kind of conversation, the talker just looks like talking. In fact, the "talker" doesn't think about what the other person said, but makes some uncomplicated formal changes to what he heard and forms his own answer. Please see:
The researcher said, "My friend suggested that I come to you, and he said it would make me happier."
The computer asked in surprise, "Your friend suggested that you come here?"
The researcher said, "He said I was always unhappy."
The computer said, "I'm sorry that you are unhappy."
After studying the boring and empty salon dialogue in detail, Weisenbaum invented his charming
"robot lady" He found that some of his compatriots could cope with each other's conversation without any interest, and their conversation could still go on well. This robot lady has inherited this ability, and can find key words in questions, and then form some formulaic and appropriate answers. It can be seen that this "lady" has some intelligence. But in fact, her intelligence is limited, and she is only good at camouflage. The mind of "lady" is very simple. In the words spoken by people, some fixed key words force it to react and put forward some fixed sentences. For example, when someone says "you don't know so-and-so", the "lady" will reply: "Why do you think I don't know?" She has a glib tongue, but she is just an expert at chatting in a salon.
Colby compiled another computer program at Stanford University, which is different from the above-mentioned "lady". Its task is to help psychiatrists who have just started working to gain some experience in talking with seriously ill patients. This program simulates a persecutor. The program stores some resume information: Perry, 28 years old, unmarried, working in the post office; He is extremely fragile and sensitive to all issues concerning his appearance, education and beliefs. His hobbies are watching movies and horse racing; On one occasion, he pounced on the bookkeeper of the racecourse because the man didn't pay the prize money he won; From then on, he began to fear the revenge of the underworld. During the conversation, Perry's mood often changes. If the talk host often touches Perry's "sore spot", Perry will start to get excited and have paranoid symptoms. Here is their conversation:
Doctor: "Are you afraid of the underworld?"
Perry: "Mafia is the most terrible mob."
Doctor: "Do you have any connection with the mafia?"
Perry: "Don't you see? They know me."
Doctor: "How do they know about you?"
Perry: "They are watching me."
Doctor: "Why?"
Perry: "They want to put me in jail for Luo Zhi."
Doctor: "Why did they hurt you?"
Perry: "They want me dead."
Doctor: "Why?"
Perry: "I have the evidence of the mafia."
A group of psychiatrists were arranged to talk to a real patient and Perry at the same time through a teleprinter. Although they knew that one of the talkers was a computer, they couldn't guess which one was. In this way, the Turing test method began to work. Later, something even more interesting happened. Previously, Weisenbaum made a robot that simulated psychiatrists to treat certain neurological disorders. Later, the idea of using this robot to treat Perry came into being. This is the dialogue between them:
Doctor: "Hello, please tell me what's wrong with you?"
Perry: "I am angry with others sometimes."
Doctor: "I don't quite understand what you said."
Perry: "You should pay attention."
Doctor: "Please go on."
Perry: "I have been to Bay-Medouus Racecourse."
Doctor: "What are you going to do?"
Perry: "I went to the horse race."
Doctor: "What else do you want to discuss?"
Perry: "I am interested in horse racing."
Doctor: "Why do you say this?"
Perry: "There are tricks when gambling on horse races."
Doctor: "I don't quite understand what you mean."
Perry: "You are really a nuisance!" "
Doctor: "Do you like to regard me as a real nuisance?"
Perry: "I'm tired of listening."
Doctor: "Why?"
Perry: "Goodbye!"
Doctor: "Nice to meet you. You should pay 399 dollars and 25 cents. "
This is their conversation, and the conversation is good, although the doctor is a little "blind". Programmers are going to gradually improve Perry's program so that young doctors can "cure" this artificial patient during the dialogue. By then, college students can learn by talking to computers, avoiding the pain caused by unskilled medical skills when facing real mental patients.
It is more important to teach robots to grasp this or that essence. There are two types of conversations with computers: limited conversations and limited understanding. In a limited conversation, the robot "understands" all the contents it talks about, but only when it comes to determining the topic, for example, playing chess or laying bricks. When you have limited understanding, you can talk to it at will, but it is far from fully understanding what you say. The robot "Lady" program compiled by Weisenbaum belongs to this category. "Lady" can only understand events and phenomena on the surface. However, with the development of control dialogue theory and practice, the robot's speech becomes more and more expressive. Turing test method began to take effect regularly.
The vice chairman of an American computer company, by mistake, accepted a Turing standard test. Since then, the status of this standard has begun to decline. Because control experts found that it is not the best standard to test the limits of computer intelligence.
What is the best standard? What level of intelligence can be called a real "intelligent" robot? This has become a new problem for intelligent scientists.
Robots teach you
The development of computer industry is based on many scientific researchers' "whimsical" subjective ideas and hard-working objective practice. As mentioned earlier, some scholars have made a lot of contributions in developing the principle of control dialogue. At this time, other practitioners and pragmatists are trying to put this new ability of robots on the bus of scientific and technological progress, and they are determined to make robots have some knowledge in specific fields.
As we know, all the information factors obtained by computers are linked by an interdependent and complex system. Compared with logical reasoning, computers often use analogy and judgment methods. They classify, merge and synthesize these elements, and gradually develop their own "thinking" ability. Now let's review some historical events of robots in this development process.
The first batch of such computers was born in the late 1950s. They have proved about 40 theorems and can answer simple and small questions like "building a children's pyramid". By the 1960s, people had been able to talk about weather and other topics with computers, because these computers knew meteorology and had the necessary syntactic knowledge to make sentences correctly. For example, if you say to it, "I don't like rain in summer." It will politely answer, "Yes, but it doesn't often rain in summer." In addition, there is a program called "Baseball" that can answer all the questions related to this year's competition: the venue, the score and the personnel of the participating teams. As for the "Talk" program, it has begun to be interested in the family relationship of the speaker, although it really knows nothing about it. It was only in 1965 that the robot "Mr" began to pay more attention to the meaning of words, not just the order of words in sentences. Computer "students" are also of this type, like a student with excellent academic performance, who can solve equations once and describe the order of solving equations in fluent English.
The more specialized the knowledge input into the computer, the more likely it is for the computer to master them. Now, some computers have become real "technical consultants". For example, they are already helping experts to determine which strata are rich in minerals; Assist experts to make a diagnosis of infectious diseases. To produce such "experts", we must teach them the knowledge of people-experts. However, no matter how incredible it is, the main difficulty still lies in how to "dig out" all this knowledge from people's brains. For example, when doctors make a diagnosis, they follow some rules according to their experience. He used these rules almost subconsciously and mechanically. Researchers spend a lot of time interviewing doctors and other experts in order to find out the basic laws inherent in their thinking process. As long as the whole process of their thinking can be restored, it is relatively uncomplicated to copy it into a computer program. Since 1965, the first "expert" in the computer was made by Feigenbaum in Stanford. As soon as it was born, it volunteered to help chemists determine the molecular structure of substances; Another technical consultant, the prospector, works even more rigorously. It studies geological maps and soil samples in detail to determine the existence of mineral deposits. It actually found a rich molybdenum mine in Washington state.
The computer "doctor" was programmed in the 1970s. It can diagnose infectious diseases after knowing the diagnosis results and main symptoms. The most wonderful thing is that if the user asks it to explain the reason for making such a diagnosis, then it can always explain that the reason for making such a diagnosis is this one, not another. Boupur, a computer expert at the University of Pittsburgh, and Myers, a medical expert, also designed the program of the computer "Kodak", which stores more diseases in its memory than a doctor can remember under any circumstances. It can combine facts, evaluation and judgment to make a difficult diagnosis. The computer actually learned to diagnose? Right, don't believe it, please see the following example:
One day, people entered the detailed illness of a middle-aged person into this computer. At that time, the middle-aged man looked so ugly that he had difficulty breathing and was taken to the hospital by an ambulance. Miles was first diagnosed with a heart attack. The computer noticed the patient's condition-no pain in the chest. When he had a heart attack before, his blood pressure was normal. There were records about diabetes in the medical records. The computer first considered the symptoms of more than ten diseases and denied these hypothetical diseases. Then, the main diagnosis results were displayed on the screen. After a few minutes, the computer came to the conclusion that the patient had a heart attack. It will take several days for doctors to make the same diagnosis. In some complicated and abnormal situations, its diagnosis is more correct and careful than that of private doctors. So Dr. Myers believes that computers are almost always willing to study every disease of patients with medical experts who have enough time. For example, after additional testing,
"Keda" can become the general staff of doctors, and it can even reduce medical expenses, because according to the questions raised by the computer, the number of times doctors designate patients for laboratory tests will be reduced.
Now the team of such "experts" has expanded. If this continues, they will have children and grandchildren. For example, the electronic computer being developed can translate, distinguish written language from spoken language, point out mistakes, learn and correct them. In short, the future "expert system" will be involved in more and more fields, from heaven to underground, from ancient times to modern times-truly "knowing three points in the sky and knowing everything on the ground" (although a bit exaggerated, it is in line with its development direction and people's wishes).