Turing:
​Computing Machinery and Intelligence
Alan Turing’s Computing Machinery and Intelligence paper from 1950 [116] was the very first scientific attempt at giving a formal definition of intelligence based on a test that will be later known as the Turing test (refer to here for more details). Although dating back from the 50s, some variants of the Turing test are still widely used nowadays.For instance, the Internet would not be as secure as it is today without the use of CAPTCHA’s in order to prevent malicious users to use bots all over the Internet. The CAPTCHA test (illustrated in figure 12) illustrates one of the impact of Turing’s paper in our daily lives and will be discussed here in more detail.
In the next section, an overview of the man behind the paper (Alan Turing) will first be given. This will help with the understanding of the background and the context in which the paper was written. We will then dive into the content of the
paper itself by properly understanding which question Turing was trying to answer, what was his preferred solution and then some of the different objections that he faced.
Fig.12 Example of CAPTHCHA test on the Internet
1. A word about Turing
​
Alan Turing was born in 1912 in London. Even at young age he showed interest in maths and sciences. It was only natural that he studied maths at Cambridge University and became a brilliant mathematician. For his dissertation and as his very first official paper, Mr Turing produced a paper that proved the central limit theorem (1) . He then continued delivering several papers as a research fellow like for instance On Computable Numbers, with an Application to the Entscheidungs problem (2) [115] which introduced the concept of the Turing machine and which is recognised today as the central concept of algorithmic computation.
Alan Turing then proceeded to study cryptography at Princeton University at which he received his Ph.D. He then came back to England as part of a British code-breaking organisation during the war. His team helped building a machine that would break Nazi’s cipher texts (refer here for more details about the scientific advancements made during WW2).
In 1952, he admitted to the police that he was homosexual, which was illegal back in the days. He then was forced to choose between chemical castration and imprisonment. Turing died in 1954 due to cyanide poisoning which was followed by a postmortem exam and with which the authorities ruled for a suicide. In 2009, former British Prime Minister Gordon Brown officially apologized on behalf of the British government for the prosecution of Alan Turing.
​
2. Moving away from the definition of intelligence
​
​
The question
​
In the 1950s, it was the very beginning of the development of computers and some of them had even been built at leading universities. Even though the number of operations per second that could be carried out by such computers was negligible
compared to nowadays computers, it was very impressive at that time. It was only natural that questions like “How will computers evolve in the future?” and “Will they ever outperform humans in certain domains?” would arise among the scientific community.
​
Back at that time, Alan Turing was no exception. Indeed, the very question he tried to answer in his paper was “Can machines think?” . While being very deep and interesting, this question has a major issue: it is not well defined. Indeed, in order to answer such a question, one must first define the terms machine and thinking, which is not an easy matter. Turing said that the definition from the most current uses of the words could be used (or in other words, the definitions from the dictionary) but that would be a dangerous attitude as it would be relatively equivalent to making an opinion poll in order to answer the question and thus this solution is described as “absurd” by Alan Turing in his paper.
Instead of trying to answer such a delicate question, Turing decided to change it to another one that is said to be closely related and unambiguous: the imitation game (or as it will be called later: the Turing test). The question is therefore no longer “Can machines think?” but rather “Can machines pass the Turing test?”.
​
​
The answer
​
According to Alan Turing, the way to answer the question defined in the previous section is to design a game (the imitation game) such that if passed, the answer to “Can the machine think?” would be “yes” and “no” otherwise.
The original game as it was described by Turing is actually quite simple. It involves three players: A, B and C. Player A is a man, player B is a women while the gender of player C does not matter. Player C is called the interrogator as his job will be to interrogate both other players in order to find out their gender. Player C can not see the other player and can only communicate with them via written notes. He can not infer any piece of information from the context (i.e. he can only rely on the
content of the notes to make up his mind about the genders of players A and B). The goal of player B would be to help the interrogator to make the right decision while player A would do the opposite as he wants him to make the wrong decision in the end. This is illustrated in Figure 13.
Fig. 13 The original imitation game.
Alan Turing then imagines the exact same game but in which player A would be replaced by a machine (this is illustrated in Figure 14). He then states that “if the interro-gator decide[s] wrongly as often when the game is played [with the computer] as he does when the game is played between a man and a woman, it may be argued that the computer is intelligent ” [116].
​
Now that the question is clearly defined, there is still one issue that is yet to be solved: which machines are concerned by the game? The first natural answer would be that we allow any engineering technique imaginable but only exclude men born in the usual manner from the game. Turing then argues that is it hard to have a definition for such machines that satisfy the condition. He therefore suggests to restrict the machines concerned by the game to digital computers as they are the ones that triggered the original question to be asked in the first place (3) . While this restriction might seem as a very drastic one, he proceeds to show that digital computers are nothing but “discrete state machines” that are universal machines. This means that for different purposes, there is no need to build different digital computers. It is sufficient to correctly program one digital computer so that it solves the problem.
The objections
​
When writing his paper, Alan Turing already knew that there would be objections. He therefore pulled the rug from under the people behind these objections by addressing each of them separately.We will here present a non-exhaustive list of some objections and their corresponding answers from Turing.
​
-
The theological and the “head in the sand” objections. The first one states that thinking is a function of the man’s soul and any machine lacking one therefore can not think. The second one states that the consequences of machines becoming intelligent would be too dreadful and therefore we should not try and render them intelligent. The first one is quickly rejected by Alan Turing simply by pointing out that any theological statement is based on the bible and that a lot of these arguments have often been unsatisfactory in the past (e.g. the bible clearly states that earth is flat while we know for a fact that it is not). The second argument is clearly connected with the first one as it comes from Man’s fear of losing his dominating position. Turing does not even refute this argument as he thinks it is not substantial enough to even be a valid objection.
​​
-
The argument for consciousness. This objection is based on the fact that machines can not feel. Indeed it states that not until, for instance, a machine can feel the pleasure as it succeeds, be warmed by flattery, be made miserable by mistake or be angry when it can not get what it wants could we say that a machine can think. Turing answers that, according to this point of view, the only way for a human to detect feelings in another human is for the second one to describe them or to have a physical reaction that suggests it (for instance a smile). Although generally true, observing a smile on a person is no proof that this person is happy (he could be mimicking it). Therefore the only way to be sure that a human feels would be to be that very human. Turing considers this point of view quite extreme. While he admits that there are mysteries about the consciousness, he also believes that solving these mysteries is not required to answer the question that the paper was interested in.
​
-
The argument from the continuity of the nervous system . It states that given the fact that the nervous system (and therefore the brain of player B) is continuous, it can not be simulated by a digital computer (which is nothing be a discrete state machine). Turing agrees on the premises but not on the conclusion. Indeed, the nervous system is continuous and digital computers are certainly not. However the difference it makes has not impact whatsoever on the output of the imitation game. In other words, the interrogator will not be able to exploit the analog/digital difference of player A and B in order to make up his mind. It is worth noting that the analog/digital debate is a very fundamental debate that will be discussed in more details on the page related to Shannon.
Alan Turing revolutionized the way we think about intelligence and machines by not only redefining the term but also by giving a way to assess such intelligence on machines that were arising at that time. However widely used all over the Internet, the different implementations of the Turing test (i.e. the variants of the CAPTCHAs) might not be enough to face nowadays new challenges. This will be discussed in more details on the page related to the imitation game here.
​
-
It is only after submitting his dissertation that Turing learnt that it had already been proven. It therefore was never published.
-
This problem was already explained in Section 1.1.2
-
As already said at the beginning of this section, there were a few digital computers built at leading universities in 1950.