In this post, I reflect on an article I enjoyed reading lately, written by Alan Turin in 1950. I learned that this is one of his most important works, as it sets the foundations of what is known today as “Artificial Intelligence” and “Machine Learning”. While reading the paper, I could feel the depth, rigor, and genius in Turing’s ideas. Below are some highlights. This summary is in no way a comprehensive account of Turing’s paper. For a full appreciation please read the article.
The central question in the paper, “Can machines think?”, is re-phrased by Turing in the form of a game he calls the “Imitation Game”, which is also the title of a very good movie about the life of Alan Turing. The concept behind the game is described as follows: “It is played with three people, a man (A), a woman (B), and an interrogator (C) who may be of either sex. The interrogator stays in a room apart front the other two. The object of the game for the interrogator is to determine which of the other two is the man and which is the woman. He knows them by labels X and Y, and at the end of the game he says either “X is A and Y is B” or “X is B and Y is A.”
To test whether a machine can think Turing asks: “What will happen when a machine takes the part of A in this game?” Will the interrogator decide wrongly as often when the game is played like this as he does when the game is played between a man and a woman?”. In other words, the goal is to test, in a systematic way, the machine’s ability to produce answers indistinguishable from those of a human. This is also known as the Turing test.
Even though known as the paper’s main contribution, this test represents only a small portion of what’s discussed by Alan Turing. He starts by investigating the “worthiness” of the new question and builds the right context for it. For instance, he stresses that a machine would satisfactorily pass the imitation game if and only if the answers it provides resemble the ones that would naturally be given by a man or a woman. No other measures of performance, such as physical beauty or artistic creativity, should be used, he argues, by the interrogator while conducting the imitation game.
The paper then explains why “machines” in the imitation game should be restricted to “digital computers”, defined by Turing as consisting of three main functions: (i) Store, (ii) Executive Unit, (iii) Control. Interestingly, this same structure is very close and relevant to modern-day computers. Turing then contrasts between the “deterministic” version of the digital computer, and the “digital computer with a random element”. He explains how this latter is more likely to mimic human behavior, in that it can make the answers provided by a machine less predictable, and therefore, more interesting in a way. He explains how a machine with a random element can mimick free will, even though this term will not recur a lot in the paper. He highlights that modern digital computers are electrical in the same way that the nervous system is.
To be more specific and pragmatic about the central question posed in the paper, Turing proposes: “Let us fix our attention on one particular digital computer C. Is it true that by modifying this computer to have an adequate storage, suitably increasing its speed of action, and providing it with an appropriate program, C can be made to play satisfactorily the part of A in the imitation game, the part of B being taken by a man?”
Turing then provides seven arguments that can be used to motivate contrary views to disprove the substitution made between the original question, “Can machines think?”, and the question in the previous paragraph.
He first exposes the theological argument some might have that “thinking” is solely a function of human beings. The second argument, not of a very different nature, relates to the fear some people can feel, emanating from the possibility that someday machines will take over. Turing expresses that these two arguments are of little substance. I find these two arguments unnecessary in a paper of this importance.
I find the third argument, “the mathematical objection”, quite interesting. Godel’s theorem (1931) shows that in any sufficiently powerful (the technical term is “consistent”) logical system (a “machine”, say), statements can be formulated that can neither be proved nor disproved. In the context of the imitation game, Godel’s theorem can be understood as stating that there are certain things a machine cannot do, i.e. answers the machine cannot provide. Turing accepts this mathematical result, but also explains that, for most purposes, this result is of little significance. He explains that even human beings lack the ability of being perfect in their answers, and so such expectation should not be required for machines.
The argument from “consciousness” related to machines is explained to be too ideal and practically very hard and restrictive to account for, especially since “consciousness”, and its related mysteries, have not been adequately understood: “I do not wish to give the impression that I think there is no mystery about consciousness. There is, for instance, something of a paradox connected with any attempt to localise it. But I do not think these mysteries necessarily need to be solved before we can answer the question with which we are concerned in this paper.”
Then the paper lists arguments related to actions machines, even at their best levels, would not be able to do.
“Be kind, resourceful, beautiful, friendly, have initiative, have a sense of humor, tell right from wrong, make mistakes, fall in love, enjoy strawberries and cream, make some one fall in love with it, learn from experience, use words properly, be the subject of its own thought, have as much diversity of behavior as a man, do something really new.”
Turing argues that people, in general, falsely get this impression because of the usual induction used by individuals in their everyday lives. Judging by the quality of existing machines, which are poorly designed and of very limited storage capacity, a person is inclined to foresee limitations in what can be achieved by machines in the future. With the increase in storage capacity, as explained in the paper, will come the diversity of actions and possibilities related to the behavior and answers a machine can give. There is therefore no inherent limitation Alan Turing sees.
The next argument is related to the “learning machine”. This, I find, is the most interesting part the paper, since it summarizes neatly the foundations behind the field of Artificial Intelligence, which has been gaining increasing popularity during the last decade. The argument, that can be advanced by some, states that a machine can “never do anything really new”. To this argument, Turing argues that “surprises” can be perceived quite subjectively, and that “original work” can often be seen as the result of planted seeds that grow with time as a result of many teachings and experiences. Thus, no work is really “original” work. He also explains that: “In the process of trying to imitate an adult human mind … We may notice three components.
(a) The initial state of the mind, say at birth,
(b) The education to which it has been subjected,
(c) Other experience, not to be described as education, to which it has been subjected.”
He continues “Instead of trying to produce a program to simulate the adult mind, why not rather try to produce one which simulates the child’s? If this were then subjected to an appropriate course of education one would obtain the adult brain. Presumably the child brain is something like a notebook …”. And explains “We have thus divided our problem into two parts. The child program and the education process. These two remain very closely connected. We cannot expect to find a good child machine at the first attempt. One must experiment with teaching one such machine and see how well it learns. One can then try another and see if it is better or worse. There is an obvious connection between this process and evolution”. These sentences represent a very early explanation of what the field of Machine Learning, and in connection Artificial Intelligence, are about. Turing explains how establishing this can in part be achieved by the principle of punishment and reward in education, in addition to sequential imperatives in behavior and actions. He then argues that adding a random element to a learning machine is a wise choice. In the same way that solutions to some problem can be reached faster when drawn randomly from a larger set, Turing argues that the choice made by the machine of an appropriate behavior can benefit the learning process and thus satisfy the teacher. He also adds about the random process: “It should be noticed that it is used in the analogous process of evolution”, in a reference to Darwin’s theory of evolution.
The argument from the continuity of the nervous system, compared to the digital computer which is a discrete-state machine, is explained to be of little impact on the imitation game.
Next comes the argument from informality of behavior. Turing accepts that it would be almost impossible to include every possible rule of conduct in the behavior of a machine. With clarity he mentions: “To attempt to provide rules of conduct to cover every eventuality, even those arising from traffic lights, appears to be impossible. With all this I agree”.
The argument from Extrasensory Perception, or Telepathy, is explained be of little relevance to the imitation game. I do not find this argument interesting, nor do I understand why it was included in this paper since, as I am sure, not many people are worried about telepathy in machines.
To finish his paper, Turing concludes:
“We may hope that machines will eventually compete with men in all purely intellectual fields. But which are the best ones to start with? Even this is a difficult decision. Many people think that a very abstract activity, like the playing of chess, would be best. It can also be maintained that it is best to provide the machine with the best sense organs that money can buy, and then teach it to understand and speak English. This process could follow the normal teaching of a child. Things would be pointed out and named, etc. Again I do not know what the right answer is, but I think both approaches should be tried.”
At the very end, he finishes with this very beautiful statement, which I believe is valid for all forms of science and human thought in general:
“We can only see a short distance ahead, but we can see plenty there that needs to be done.”