Click on the text "informatization collaborative innovation special committee" above to pay attention to us
Abstract: human constructs the concept to understand the world, and human cognition is closely related to human body. Computer is the extension of human mind, and various cognitive instruments (machines) invented by human beings are the extension of human senses, which are data forming functions. Therefore, we can embed the concept of human into the computer to make it conceptualize the world, and at the same time, we can connect the artificial perception system to digitize the world, so as to realize artificial intelligence. Such artificial intelligence is human like or quasi human, which can acquire knowledge of the outside world "like" human beings. Because the concept embedded in the computer is human, the artificial perception system, that is, the data formation function, is also based on the scientific theory of human, and the knowledge acquired by such artificial intelligence can also be understood by human.
Key words: artificial intelligence; knowledge formation function; conceptual world; perception system
introduction
We gain knowledge of the outside world through observation and thinking, but our ability of observation and rational thinking are limited. We use computers to accomplish heavy and even impossible computing tasks in principle. We use microscopes and telescopes to observe things that we cannot see with the naked eye Machines (including instruments) play an important role in our cognitive process. They can be regarded as the extension of human senses and mind. With the improvement of computer computing power, the proposal of efficient algorithms and the emergence of big data as calculation materials, the related technologies around artificial intelligence are profoundly changing the social life and production mode. Whether and to what extent intelligent machines can replace human mind has become a hot topic. As a branch of computer science, artificial intelligence is defined to study how to make computers and program them so that they can do what the human mind can do. For human intelligence, can we know the world, that is, acquire knowledge about the world. Can artificial intelligence (machines) made by human beings know the world? If it can understand the world, how to understand its cognitive behavior? If not, where is its boundary?
In the current research field of artificial intelligence, knowledge representation and reasoning, machine (deep) learning and other concepts are frequently used by people; in the practice field of artificial intelligence, machines can crush human chess players, and UAVs can process and judge the photos taken So, the natural question is, can AI, or machines, have or learn "knowledge"? Some people will say that this is a kind of usage or ambiguous usage, and machines can't have the same "knowledge" as owners; while in other people's opinion, machines can have intelligence in the sense of weak artificial intelligence, so machines can have intelligence but can't have knowledge, which is an unacceptable conclusion. This paper will demonstrate that machines exist in the cognitive world in the way of "quasi human" or "quasi human", which can operate human concepts and operate perfectly in the conceptualized and complete world. However, it is necessary to build a channel to connect with the world and develop a human like perception system, so that machines can perceive the world as human beings do.
How can AI be embedded in the world?
Knowing is a mental state of human being, which is a proposition attitude; someone knows a proposition, which constitutes his or her knowledge. Knowledge is considered to be justified true belief, which is a widely accepted definition, although it causes the problem of Getty's dilemma. Here "proof" is a process or behavior of giving reasons, "true" is a metaphysical concept, and "belief" is a certain internal state of proposition in human mind. "Truth" is the nature of proposition that can not be observed, and it is questioned as an element of knowledge; as for the proof, there is no unified standard to judge how a belief can be proved. However, knowledge is an internal cognitive attitude of related subjects, and there is no objection.
According to this definition of knowledge, today's computer, as a machine, although it has a strong ability, can't "own" knowledge. Because machines do not have the concept of "truth"; machines that do not understand the meaning of proof can not prove a proposition, although machines can give the logical relationship between propositions; especially, machines are difficult to have human beliefs. This view is reasonable, although it is anthropocentric.
Our cognition and our physical behavior interact in two ways. On the one hand, our cognition influences our actions, and thus our physical behaviors. Bruno would rather be burned than give up heliocentrism; I don't drink high sugar drinks because I believe that high sugar drinks will increase my blood sugar. On the other hand, our physical behavior affects our cognition. According to the theory of body related cognition, we recognize the world through the body embedded in the environment: the way and steps of human cognitive process are actually determined by the biological properties of the body, and the content of human cognition is also provided by the body. In metaphors on which we live, G. Lakoff and M. Johnson believe that people's subjective feelings of the body and the experience of the body in activities provide the basic content for language and thought; cognition is what happens when the body acts on the physical and cultural world.
The core of a computer is to execute a program composed of human instructions, that is, software; while hardware systems such as chips are only the physical basis for executing programs, that is, hardware, where software does not constitute the body in which computer programs are embedded into the world. So computers can't have the same belief as humans. And machines that perform non computing functions, such as cars and air conditioners, can't have faith.
However, not everyone agrees that machines cannot have beliefs. For example, in the view of McCarthy, an AI expert, almost all machines that can solve problems can be said to have faith, even simple machines like thermostats can be said to have faith, "my thermostat has three beliefs - too hot here, too cold here, and the temperature is appropriate here.". (U.S. John Searle, heart, brain and science, Shanghai Translation Press, 2006, pp. 22, 23.) This, of course, is what Searle can't agree with, and what he can't agree with: machines don't have minds. What about faith?
Machine is the extension of human senses, and its cognition is also anthropomorphic. In fact, human beings have been unconsciously "recognizing themselves" without Socrates' admonition. People know their own limitations and try to overcome them. Machines are developed by human beings to understand and overcome their own limitations. Different from other human inventions, computers were invented by human beings to help people reduce human brain work. They were invented to make up for the limitations of the brain. German philosopher E. Kapp put forward the theory of organ projection - the materialized form of tool as technology is the projection or extension of human organs to nature. In this sense, whether it is the predecessor of computer, such as Babbage's difference machine, or the modern high-speed running large computer, they are the extension of human brain or mind. Of course, people worry that the computer, as the extension of human mind, can develop to "own" the independent mind and fight against or even eliminate human beings.
Today's machines are capable of computation as well as deduction and induction. Computing is the function that computers are initially given. Today, people won't be surprised that machines can work out complicated mathematical problems, such as the work of making computers work out 1000 decimal places or even harder after the PI. We think this is the most common thing, and if someone can do this calculation, he or she will be regarded as a genius instead. For mathematical equations, we need to find their "deductive solution" or "analytical solution", and we also have many theorems in this respect; however, for complex mathematical problems, we are often difficult or even impossible to give such "analytical solution". So "computational solution" came into being. The computer helps us to get the solution. A computational solution is often an "approximate solution", such an approximate solution is satisfactory relative to a certain demand. In engineering, we need to obtain such a satisfactory solution in a limited time, instead of expecting mathematicians to seek the reasoning of analytical solution endlessly. In today's scientific research, "calculation", "theory" and "experiment" are three pillars.
The computer can do deductive reasoning - get the conclusion from the premise. The computer can "prove" the mechanization theorem. In this field, computers can "prove" mathematical theorems that human beings cannot or cannot obtain, such as the four-color theorem. In the field of mechanical mathematical theorem proving, there are three aspects: "proof test", "proof discovery" and "conjecture generation". Proof test is to give a reasoning d to draw a conclusion P from a large number of premises to judge whether D is reasonable; proof discovery: give a large number of premises and a hypothetical conclusion P, judge whether P is a logical successor, if so, give a formal proof; conjecture generation: give a large number of premises and an interesting conclusion P derived from premise logic, and give a proof. The difficulty of these three aspects is gradually increasing. In the first two aspects, people have made many outstanding achievements by using computers. Machines can do inductive reasoning. The so-called data mining work is that people use computers to do inductive reasoning based on big data; mining work based on massive data is impossible or difficult for a single person to complete. In the complex game and decision-making problems, as follows, the machine can give the action strategy through learning and do much better than the human. At this time, the machine is making the strategy selection based on inductive reasoning. Machines do inductive reasoning in a fully conceptual world, not in the real world. If machines can do such inductive reasoning in the real world, then machines are scientists. Because machines can do reasoning and decision-making under these complex conditions, the behavior of computers is more and more "intelligent".
How to understand these abilities of computer? It is generally agreed that the machine is able to do these tasks by relying on human designed programs. A program is a series of human instructions. It is a kind of human intelligence to be able to design a program. Human beings solve the problems that we can't do or can't do in principle by letting machines execute the programs designed by human beings. If we think that machines have "a little intelligence", then with the passage of time, the intelligence of machines will become stronger and stronger. In this way, it is possible for humans to create a super intelligent machine, which can complete all human intelligent behaviors. An intelligent explosion will occur. Super intelligent machine is the last invention of human beings, because design machine is one of human intelligence, which can design machine and better than human beings. This is what I. J. good put forward in 1965. Intelligent explosion is the so-called technological singularity.
For what these computers can do, people may not