Intelligent machines

Some philosophers tried to prove that it is impossible to create artificial intelligence, i.e. that machines can never act intellectually. Some of them even used their erudition to urge everyone to stop researching artificial intelligence, citing the following arguments.

The artificial intelligence created within the framework of the cult of computer-centrism does not give even the slightest chance that with its help it will be possible to achieve long-term results ... the time has come to direct the efforts of researchers of artificial intelligence (and significant funds allocated for their support) in areas excellent from this computerized approach.

Obviously, the answer to the question of whether it is possible or impossible to create artificial intelligence depends on how the very concept of artificial intelligence is defined. Essentially, the creation of artificial intelligence is a struggle to develop the best possible agent program in this particular architecture.

When using such a formulation, the creation of artificial intelligence is possible by definition, since for any digital architecture consisting of k bits of memory, there are exactly ^2k agents programs, and in order to find the best of them, it is enough just to check them all sequentially. Such an approach may not be feasible with large values ​​of k, but philosophers operate on theoretical rather than practical constructions.

The above definition of artificial intelligence is quite suitable for solving the technical problem of finding an acceptable agent program in the presence of some given architecture. Therefore, it would be quite possible to finish this section right now by responding positively to the question given in its title.

But philosophers are also interested in the problem of comparing two computational architectures - man and machine. In addition, philosophy has a long tradition of finding an answer to the question: Can machines think? Unfortunately, this question itself is defined incorrectly. In order to understand why this is so, consider the questions below.

• Can cars fly?
• Can cars do swimming?

Most people will agree that the answer to the first question is positive, since the planes are designed precisely to fly, but will answer the second question in the negative; ships and submarines move above and below the water, but this is not what we call swimming. However, neither the above questions nor the answers have any impact on the daily activities of aeronautics and shipping specialists, as well as on the users of the machines they operate.

In addition, the answers to these questions can in no way help to better design or expand the capabilities of airplanes and submarines, and to a greater extent concern the choice of the right ways of using words. The word "swim" in English gradually adopted the meaning of "moving in the water due to movements of the body parts", and the word "to fly" (fly) does not impose such restrictions on the choice of methods of movement. The practical ability to use the services of "thinking machines" is given to a person for only 50 years, and this is not enough to make the interpretation of the word "thinking" spread to machines.

Alan Turing in his famous article "Computing Machinery and Intelligence" indicated that instead of searching for the answer to the question whether machines can think, we should be interested in whether machines can pass the behavioral intelligence test, which later became known as the Turing test.

This test consists in the fact that the program takes part in a conversation (made up of messages that are transmitted online) for 5 minutes with some interlocutor. Then the interviewer must determine whether this conversation was conducted with the program or with another person; The program successfully passes the test if it manages to deceive the interlocutor in 30% of cases. Turing predicted that by 2000, a computer with a memory capacity of 10 ⁹ units could be programmed so successfully that it passed this test, but it turned out to be wrong.

Some people even managed to get their nose on for 5 minutes; for example, the Eliza program and the chatbot (robot interlocutor) Mgonz, acting on the Internet, more than once deceived people who could not understand that they were communicating with the program, and Alice could even fool the judge in competitions for the Loebner prize (Loebner Prize ) 2001. But not a single program was able to approach the 30% criterion in the fight against specially trained judges, and in the field of artificial intelligence itself, Turing is receiving less and less attention.