Computing Machines and Mind

“Computing Machines and Mind” (eng. Computing Machinery and Intelligence ) is a fundamental work in the field of artificial intelligence ^[⇨] , written by English scientist Alan Turing and published in 1950 in the Mind magazine , giving the general public an idea that This is now called the Turing test ^[⇨] .

Turing discusses the question “Can machines think?”. Since the words “machines” and “think” cannot be defined unequivocally, Turing suggests replacing “a question with another one that is closely related to the first one, but expressed not in such ambiguous words ^[1] .” To do this, the author, first, finds an unequivocal replacement for the word "think." Secondly, it explains what kind of “machines” he considers ^[⇨] . On this basis, he formulates a new question related to the original one: whether the machine can perform actions that are indistinguishable from deliberate actions. According to Turing, this question can be answered in the affirmative, for which the author shows the inconsistency of opposing views ^[⇨] , and also describes the way to create one of these machines ^[⇨] .

Turing Test

The standard interpretation of the Turing test, in which the examiner communicates with the person and the machine to determine who is who

Instead of determining whether machines can think, Turing suggests the question of whether machines can win a game called Imitation Game. It involves three participants: a car, a person, and an examiner (who is also a person). The examiner sits in a separate room from which he can communicate with both the machine and the person. The answers must be submitted in text form and transmitted via TTY or with the help of an intermediary. Both the machine and the person are trying to convince the examiner that they are human. If the examiner cannot confidently say who is who, it is considered that the car has won the game. This is a description of the simplest version of the test. There are other variants of the Turing test.

As Stephen Harnad noted, the question began to sound like "Can machines do what we (as thinking creatures) can do?" ^[2] ". In other words, Turing no longer asks “Can machines think?”, He asks if the machine can perform actions that are indistinguishable from deliberate actions. This formulation of the question allows us to avoid difficult philosophical problems by definition of the verb “think” and focus on the tasks of creating and increasing productivity, which makes the ability to think possible.

Some have decided that the Turing question only sounds like, “Can a machine that communicates via a teletype completely deceive a person that she is a person?”. However, Turing is not talking about fooling people, but about reproducing the cognitive abilities of a person ^[3] .

Digital computers

Turing also notes the need to determine which "cars" are meant. Naturally, he excludes people from the class of cars. The clones also would not provide an interesting example of "constructing a thinking machine." Turing proposes to focus on the possibility of "digital computers" that manipulate binary numbers 1 and 0, overwriting them into memory through simple rules. He gives two reasons for this:

• First, as early as 1950 digital computers existed.
• Secondly, such computers are “universal ^[1] ”.

Turing's research in the theory of algorithms proved that a digital computer can simulate any discrete machine, having sufficient amounts of memory and time. (This is the main point of the thesis of Church - Turing and the universal Turing machine.) Therefore, if “any” digital machine can act as it thinks, then “each” sufficiently powerful digital machine can also. Turing writes that "all digital computers are in some sense equivalent ^[1] ."

This allows you to ask the original question even more correctly. Now Turing decides the question differently: “Let us fix our attention on a digital computer B. Really, changing the computer so as to have enough memory, which is equivalent to increasing the speed of its actions, and providing it with a suitable program, computer B can be made so that it is satisfactory played the role of computer A in the imitation game, and the role of person B? ^[1] ". This question, according to the author, has become a direct question of software engineering.

In addition, Turing argues that it is necessary "not to ask whether all computers would succeed in the game and whether all existing computers would succeed, but imaginary computers could succeed in it ^[1] ." This is most important in order to consider the possibility of achieving a “thinking machine,” regardless of whether resources are currently available for it or not.

Opposite views

After defining the question, Turing returns to the answer to it: he considers 9 basic opposing views, which include all the main arguments against artificial intelligence that were available before the first publication of the article.

1. The theological argument : It says that thinking is a function of the immortal soul of man, so no machine is able to think. Turing replied: “Trying to construct such machines, we must not unceremoniously usurp His power of the gift of the soul, just as we do not do this by producing children. In both cases, we are rather His instruments, creating containers for the souls He created ^[1] . ”
2. The head of the sand argument : “The consequences of machine thinking would be too terrible. Let us hope that machines are not capable of anything of the kind. ”This opinion is most common among intellectuals, since they regard man as superior to other creatures because of his ability to think. Turing does not consider this argument sufficiently justified to deserve a refutation.
3. Mathematical Objection : This objection uses mathematical theorems such as Godel's incompleteness theorem to show that there is a limit to the questions that a computer based on logic can answer. Turing says that people themselves often give incorrect answers to questions, so the joy about the weakness of the machine can not be justified. (The same argument was cited by the philosopher John Lucas (English) Russian. ^[4] in 1961 and the physicist Roger Penrose ^[5] in 1989.)
4. Argument from Consciousness : This argument is expressed in Professor Jefferson’s Lister Speech in 1949: “Until the machine writes a sonnet or a concert inspired by feelings, rather than a random combination of symbols, we cannot agree that the machine is equal to the brain ^{[6 ]} . ”Turing replies that in order to make sure that a person thinks, it is necessary to become that person. Therefore, to make sure that the car thinks, you can only become a machine. He adds: “I do not want to say that there is nothing mysterious in the mind. <...> But I don’t think that all these secrets must be revealed without fail before we can answer the question of interest to us in this article ^[1] . ”
5. Argument from various limited abilities : This argument looks like: “Well, I agree that you can create machines that do everything you mentioned, but you never make the machine do X.” Turing gives examples:

   To be kind, inventive, beautiful, friendly ... be proactive, have a sense of humor, distinguish good from evil, make mistakes ... fall in love, enjoy strawberries with whipped cream ... fall in love with someone, learn from experience ... use words correctly, think about yourself ... show the same diverse behavior as a person, create something new

   Turing notes that these are usually allegations. All of them depend on naive assumptions, what future machines may be, and which are “hidden arguments from consciousness”. He offers solutions to some of them:
   1. The machine can not be wrong . Simply program a machine that makes mistakes.
   2. The machine can not be the subject of their own thoughts . A program that can report its internal states and processes, it’s also a debugger program, can certainly be written. Turing argues that "a machine can certainly be its own subject content."
   3. The machine can not be diverse behavior . He notes that a computer with a sufficiently large memory may have an astronomical number of actions.
6. Lady Lovelace's argument : One of the most popular objections is that the car is not capable of being original. According to Ada Lovelace, this is due to the inability of the machines to learn independently.

   The analytic machine does not claim to create something new. A machine can do all that we can prescribe. She can follow the analysis, but she cannot predict any analytical dependencies or truths. The functions of the machine are to help us get what we already know.

   Turing says that Lovelace's objection can be reduced to the statement that the machine “cannot surprise us”, to which one can directly answer that the machines surprise people very often. In particular, because the consequences of certain facts cannot be accurately determined. Turing also notes that Lady Lovelace’s information about the machines did not allow her to imagine that the memory of the human brain is very similar to that of a computer.
7. Argument from the continuity of the nervous system : Modern neurological studies show that information in the nervous system is not represented in digital form. “The slightest error in the intensity of a nerve impulse in one of the neurons can cause a significant difference in the output impulse. With this in mind, it can be argued that it is impossible to imitate the behavior of the nervous system using any discrete machine. ” ^[1] Turing acknowledges this, but argues that any continuous system can be modeled with a sufficiently high degree of accuracy in a digital computer. (The philosopher Hubert Dreyfus put forward this argument against “biological speculation” in 1972 ^[7] .)
8. Argument from informal behavior : This argument says that any system acting according to the laws will be predictable, therefore it cannot be called truly intellectual. Turing notes that there is confusion between the “rules of conduct” and the “laws of behavior”. If the scale of the machine’s actions is large enough (like that of a person), its behavior will be very difficult to predict. He argues that if the law cannot be seen in human behavior, this does not mean that it does not exist. He writes: “The only way to learn these laws is through scientific observation, and under no circumstances can we say:“ We searched enough. There are no such laws ^[1] . ”In 1972, Hubert Dreyfus declared that actions and decisions human problems are not based on formal rules, instead instinct and awareness are used, which can never be expressed through rules ^[7] . Subsequent studies of artificial intelligence in robotics and computational intelligence try to reveal more complex rules. that drive our “informal” behavior and unconscious perception and mobility skills.
9. Argument from extrasensory perception : In 1950, extrasensory perception was an active area of ​​research, and Turing also allocated an important place for him, arguing that all the conditions of the tests would have been carried out without the influence of telepathy.

Learning machines

The last section of the article Turing begins with an assessment of the possibility of developing thinking machines in terms of engineering and programming. For the imitation game, in his opinion, the required memory capacity of the equipment of those years seemed quite feasible, and there was no need to increase the speed of operations. More important was the task to create a machine program for this. “Trying to imitate the mind of an adult, we have to think a lot about the process by which human intelligence has reached its present state ^[1] .” The author identifies three components here:

1. initial state of mind;
2. education;
3. an experience that cannot be called education.

To avoid programming such a state, Turing proposes to write a program that would imitate the mind of a child, and a program that performs education. The author’s calculation is that the mechanism in the child’s brain is simple, and a device like this can be easily programmed, although not on the first attempt. The proposed process of education is partly based on the method of punishment and rewards.

In this case, the machine should be arranged in such a way that the arrival of a “punishment” signal in it would lead to a sharp decrease in the probability of repeating those reactions of the machine that immediately preceded this signal, while the “encouragement” signal would, on the contrary, increase the probability of those reactions that preceded him (which "caused him").

To increase the complexity of the “child-machine”, Turing proposes to “build” a system of logical conclusions into it, which would not necessarily satisfy the principles of strict logicians, for example, “type hierarchy”.

An important feature of such a learning machine is that the teacher can only predict her behavior with some probability. Departure from absolutely deterministic behavior, apparently, is a manifestation of intelligence. Another important learning outcome is that mistakes will be made in a natural way, rather than “coaching” in order to confuse the examiner of the imitation game.

Recognition and criticism of work

Since the publication of the article, “it has become one of the most reprinted, cited, referenced, incorrectly cited, paraphrased, and generally noticeable philosophical articles ever published. It influenced many intellectual disciplines - artificial intelligence, robotics, epistemology, philosophy of mind - and helped shape public opinion as it is now, about the boundaries and possibilities of the inhuman, man-made, artificial “intelligence” ^[8] . ”

During the 1950s and 1960s, noteworthy arguments against the possibility of creating a machine capable of thinking were relatively rare. Even the existing objections did not look convincing enough either from an evolutionary or logical point of view and did not have a deterrent effect on research in the field of artificial intelligence.

In 1972, Hubert Dreyfus published a book, What Computers Cannot Do, in which mind manifestations in existing artificial intelligence systems were sharply criticized ^[7] . In his opinion, the models lacked that huge stock of non-formalized knowledge about the world that any person possesses, as well as the ability inherent in common sense, to rely on some or other components of this knowledge. Dreyfus did not deny the fundamental possibility of creating an artificial physical system capable of thinking, but he was very critical of Turing's idea that this could be achieved by manipulating symbols with the help of recursively applied rules.

However, these objections by artificial intelligence experts and philosophers were not accepted and did not affect the further development of research in the field. Overcoming the problems described by Dreyfus was considered possible in the future, after creating more powerful machines and better programs.

But in the late 1970s - early 1980s, an increase in the speed and memory capacity of computers increased their “mental abilities” only slightly. To obtain almost reliable results, it was necessary to spend much more time than the biological systems required for the same tasks. Such slow modeling processes were alarming for some professionals working in the field of artificial intelligence ^[7] .

In 1980, John Searle in his article “The Mind of the Brain - a Computer Program?” Expressed a fundamentally new critical concept that challenged the very fundamental assumption of the classical research program on artificial intelligence, namely, the idea that correct manipulation of structured symbols by recursive application of the rules , taking into account their structure, may constitute the essence of the conscious mind.

Searle explained his reasoning in an experiment called the "Chinese Room." Its meaning lies in the fact that a machine capable of passing the Turing test manipulates symbols, but cannot give them any meaning. Он ставит вопрос, почему вообще компьютерное моделирование человеческого мышления считается полностью ему идентичным и почему в этом случае может возникнуть разумное поведение.

Никто не думает, что компьютерная модель пищеварения способна что-то переварить на самом деле, но там, где речь идет о мышлении, люди охотно верят в такие чудеса, потому что забывают о том, что разум — это такое же биологическое явление, как и пищеварение ^[9] .

В отличие от Тьюринга, Сирл не считал, что мышление сводится к программам, в то же время, не отрицал самой возможности создания искусственной мыслящей системы. «Китайская комната», предложенная Сирлом, подняла много критики, уточнений и обсуждений, которые все же ничего не разъяснили в поднятых вопросах и не привели к объединению различных мнений ^[8] .
To demonstrate the created thinking machines in 1991, businessman Hugh Löbner (English) Russian. founded and subsidized an annual competition to determine and award a prize to a computer program that is more satisfactorily pass Turing test. However, for all the time of the contest, the programs remained fairly innocent and did not show a great desire for progress. Regarding these attempts to pass the Turing test, professor of physics Mark Halpern (Eng.) Russian. in his article "Disappointing with Turing test" says:

Конечно, невозможность пройти тест Тьюринга является эмпирическим фактом, который завтра может измениться на противоположный; что более серьезно — так это то, что для большего и большего числа наблюдателей становится ясно, что даже если это произойдет, этот успех не будет означать то, что Тьюринг и его последователи имели ввиду: даже осмысленные ответы на вопросы испытателя не доказывают присутствия активного интеллекта в устройстве, через который эти ответы проходят ^[8] .