Translated material
Georgy Dalidovich. Artificial intelligence notes: simulation games
In the modern understanding, the main thing for all directions for the development of artificial intelligence is its ability to extract information from the environment and use it for “self -learning”, in fact, reprogramming, in order to choose the optimal strategy of action. This question was disassembled in our previous article . It is important to note that the “environment” for artificial intelligence is that it can perceive with the help of its sensors receptors, therefore, depending on the type of task and purpose, it can be both ordinary reality and some abstract environment with formalized rules of behavior and laws.
There are several main areas of modern developments of artificial intelligence systems. Among them, programs intended for maintaining abstract games are widely known - in particular, for playing chess. A major success of this direction was the victory of the IBM "Deep Blue" computer , which beat the world champion Harry Kasparov in 1997. Obviously, the "environment" for "Deep Blue" is limited by 64 cells of the chessboard. However, in this field he achieved excellent results.
Machines that take on the intellectual functions of a person operator when managing complex systems are almost more important. There are a lot of similar systems: these are military developments, including forecasting the enemy’s actions, a comparative assessment of the importance and danger of goals and weapons; complexes of control of complex industries and energy systems; Research devices for the far cosmos, where direct management of the ground operator becomes ineffective due to a significant temporary delay in the signal; Programs for forecasting and choosing a strategy for actions in economic and social systems - for example, an “perfect broker” playing on the exchange - and many others.
But the human consciousness is usually not more difficult, really intellectual systems in their own way, but devices for simulating the appearance and actions of a person, although they are still more toys than prototypes of real machines.
In general, a machine imitation of actions that is only humanly characterized by a person has a centuries -old story, starting with a medieval “chess machine”, where a player hidden under a table and watching a board through a system of mirrors moved by a mannequin.
The last achievements of the "cyberimitators" of a person look impressive and affect almost all types of specific human activity - from its deep creative manifestations to purely external ones. As an example , the EMI program, which "wrote" the 42nd Symphony of Mozart in co-authorship with David Cope, performed in the concert hall of the University of California in Santa Cruz. (Let me remind you that the last (41st) symphony was written by Mozart more than 200 years ago.) At least none of the experts could absolutely confidently say that "this is not Mozart." At the annual meeting of the American Association of Scientists working in the field of artificial intelligence, the unexpected participant - Office Boy 2000 attracted the universal attention in Orlando (Florida). This cylindrical device on wheels was walking along the hotel hall and pestered to “colleagues” with numerous questions, fussed and expressed its emotions with random exclamations. However, in appearance, of course, it could not be confused with other participants.
The result of one of the most famous programs for imitation of the appearance and facial expressions of a person was a green -haired girl with an exotic name Ananova ( Ananova is not a surname, an emphasis on "o"). Although, of course, it does not “delve into” into the content of the read messages, but acts according to the laid scenario.
It is obvious that the moment when the autonomous and orienting environment is better than Office Boy, more realistic than Ananova, and not only a combination fragments like EMI, but a creature capable of real creativity will meet you, if not on the street, then at least at some exhibition.
The border between the "seeming" intellectual systems and devices or creatures that are really such, is blurry and vague. An attempt to determine that here that, probably, can only be implemented in the real interaction of the "subject" with the external environment or at least one single person.
In 1950, the English mathematician Turing determined the conditions of an imitation game that allows you to find out whether or not an automatic device is artificial intelligence, by his ability to pass the Turing Test.
Imagine an operator-experimenter connected by two terminals to a machine and another person. Moreover, our experimenter does not know which of them is tested artificial intelligence, and who is a person. His “human” partner should be honestly answered by the questions asked by the operator, and the car has the right and is obliged to “deceive”, trying to prove that it is the person. If she succeeds, then she should be recognized as well -developed artificial intelligence. Over the past decades, this test has been set repeatedly in many research centers. The machine managed to achieve success only in the case of rigidly limited areas and topics of "interrogations". Now it looks somewhat naive, but in the 60s such experiments were a reason for many speculation and sensational statements.
Türing believed that the car would ever expand its “world of Turing”, consisting of mutually exclusive and unambiguous events, to reflect reality and win an imitation game and, accordingly, would be recognized as “reasonable”. In his works, he listed nine possible arguments against the possibility of recognizing the car "reasonable" and if he did not refute, then at least he was witty commented on them.
1. The ability to think is the function of the immortal soul of a person. Since computers have no soul, computers cannot think.
Objection. Due to his own omnipotence, God can endow the computer with the soul if he wishes. Just as we grow our children to accommodate souls given by God, we can, fulfilling His will, create thinking computers.
2. It is terrible if the mechanisms could be equal to us in the ability to think!
Objection (comment). It would really be very bad (for us)!
3. Formal logic irrefutably proves that it is impossible to create a computer that could find the answer to any question.
Objection. Is it possible to find a person who could answer any question? Computers are not stupid than we are. Although no one can give answers to all questions, why not endow the car with the ability to self -reproduction and the creation of more and more advanced devices in subsequent generations, so that the approach to truth becomes arbitrarily more accurate?
4. Although computers can produce thoughts or judgments, they cannot feel what they have produced. They cannot experience pleasure from success, regret about their mistakes and disappointment, when they do not receive the desired.
Objection. The only way to make sure that the computer has a computer is to become it. An experiment could be more practical to create a computer, which was able to stare to explain the course of his reasoning, motives and indicate obstacles that have to be overcome - as well as analyze emotionally rich information - for example, poetry. Obviously, such a car can no longer simply "seem" thinking. She will have to be like that.
5. The computer cannot be kind, resourceful, friendly, to possess the initiative, a sense of humor, to distinguish the right and wrong, make mistakes, fall in love, enjoy strawberries and encourage someone to do this, use the words properly and be the subject of their own thoughts.
Objection. Why not? Although such a computer has not yet been created, it is still possible in the future.
6. A computer can never create something original or amazing. He does only what is laid down in the program.
Objection. How do you know that the so -called "original" human work is not just a sprout of what was laid down with upbringing and education, or is it simply the application of general principles? And why do you think that the correct answer of the computer to a complex question that is not consistent with the gross and erroneous initial assumption of a person is not “original” and “amazing”?
7. Nerve cells have a smooth scale of possible electrical activity. You can feel: “slightly”, “a little”, “something like”, while the computer distinguishes only two possible logical states-yes or no.
Objection. Using the "random number generator", you can force the computer to imitate the unstable and probabilistic behavior of the nervous system, so that the observer will not be able to notice the difference.
8. It is impossible to decompose all your life into rules. For example, if the rules of the road requires stop in red light and continue to move with green, what will you do if the traffic light has deteriorated and red and green lights burn at the same time? Perhaps you should add a new rule that prescribes to stop in this case, but it will not exhaust all possible options. Since the behavior of the computer is controlled by the rules, there will be obviously situations when it is helpless or make a mistake.
Objection. Although life is not exhausted by a set of rules, it is possible to determine a number of the most general principles of behavior inherent in the human psyche and teach them a machine.
9. A person has a psychic perception, but the computer does not.
Objection. It is possible that the random number generator will be able to exert the same influence on the behavior of the machine as an extrasensory perception on human behavior. In extreme cases, during the test, you can try to exclude its influence by the appropriate screenings.
Is it not true, now, after half a century, many of the thoughts of Turing and his hypothetical opponent look nothing more than funny. And the idea itself with the placement of the tester and the subject in isolated rooms with terminals already seems naive. Firstly, the machines perceive and reproduce oral speech quite tolerably and can even imitate the expression of emotions on their synthesized “face”, and secondly, the test conditions are performed with literal accuracy in any chat or forum, which has already become part of reality for many of us. The Internet is the artificial environment in which the imperfection of artificial intelligence receptors does not affect its perception of the virtual world.
Of course, the Turing test does not fully meet modern ideas about artificial intelligence. But the main thing, probably, is not in the naive optimism of Turing about the capabilities of his car, but in those restrictions on its capabilities that he saw better than others and which he did not bring in the list of typical objections of the primitive “anthropocentric”. In particular, he formulated the concept of “oracle turing”, capable of working with “unprofitable numbers” (uncomputable numbers), which is written only “that it obviously cannot be a machine”. The "inexpressible numbers" themselves arise as a result of the "unsolvable" situation and are the cause of the Turing machine stop. It was on the problem of these "unsuccessful numbers", which, in modern jargon, are able to "hang out" the system, and worked Turing for several decades.
Primitive people about simultaneously ceased to be just herd animals and created the first primitive system of beliefs. Oracle or God was a necessary condition for a certain level of development of intelligence - both for the Turing machine and in human history. This has a causal relationship, and not just a formal coincidence. A thin and elusive line distinguishes a car from a person, as, indeed, one person from another. The machine will become something more than just a machine when he creates a “supernatural” way of solving insoluble situations and finding answers to “unsuccessful” questions. But not every person is capable of this, which, however, does not deprive him of the right to be considered a person. However, we have already found out that to be considered and to appear does not always mean the same thing.
The main conclusion is that the creation of artificial intelligence is a logical stage in the general evolutionary process, and its "artificiality" is not of fundamental importance. In the process of natural selection and competition of artificial and natural intelligence, the dominant will become precisely the component of the global consciousness that can most effectively adapt to environmental conditions, including various virtual realities. Probably, this component will be precisely cybernetic, but it will no longer be “artificial”, since the process of natural selection will pass, and will include all the achievements of the biological component, opening the way to the real immortality of an individual person with the preservation of personal memory.
Alan Turing was concise by formulating the thesis that "science is a differential equation, and religion is a boundary condition." It is important to observe this "boundary condition." The following article of the cycle will be devoted to the complex personality and tragic fate of Alan Turing.