Turing Test & Imitation Game
From the following text you will learn about:
- Enigma & thinking computers
- Are computers capable of thinking?
- What is Imitation Game?
- Turing Test
- Turing Test in movies
- Will the machine ever pass the Turing Test?
Do you remember the 2014 film “Imitation Game” directed by Morten Tyldum and starring Benedict Cumberbatch? The film tells the tragic story of a brilliant British mathematician Alan Turing – one of the main members of the team that broke the German Enigma cipher machine [1].
Enigma – encryption machine
What does the title “Imitation Game” have to do with deciphering Enigma and Alan Turing’s role in the process? The work on the Enigma code and decryption machines was covered by the highest level of security and absolute military secrecy. The process of declassifying the documents and the role Alan Turing had in this work did not begin until 1982. However, Alan Turing in the world of mathematics and computer science is almost a cult figure – known for over 70 years – thanks to him we have a modern computer, and he owes his fame to the “Imitation Game” – but one thing a time.
Computers and the question of whether they can think
Turing as a mathematician was already well known by the 1930s. In 1936, when he was only 24 years old, he wrote his most important work in the field of mathematics “On Computable Numbers with an Application to the Entscheidungsproblem”, in which he proposed an abstract concept of a computing machine equipped with a program – a sequence of instructions and memory – the role of the memory was to be played by a tape with the stored instructions and parameters of the program. Today we know this concept as the so-called “universal Turing machine” and all modern computers are built according to it (with the exception of quantum computers). In times when Turing wrote his article there was no technical possibility to build such a computing machine. However, the outbreak of World War II accelerated the work on creating machines that in a certain sense automated calculations and the so-called symbol transformations (necessary for breaking Enigma codes). These devices were based on electromagnetic relays. The use of vacuum tubes was also tested. These works eventually led to the creation of the first computers based on the assumptions indicated by Turing in the mid 1940s.
The rise of computers-machines that have far surpassed human abilities in the area of Arithmetic has led to the increasingly asked question – “Can a machine think?” Will a computer one day think like a human, solve problems, play chess, or become creative?
Imitation Game
In October 1950, Turing published perhaps his most famous article, “Computing Machinery and Intelligence,” in the Mind journal. It is because of it that he is so well known today. Turing decided to “solve” the problem of thinking machines once and for all.
At the beginning of the article Turing makes it clear that the question posed is senseless – computers will never, according to the author [2], think in a way known to man because the operation of the computer is based on other principles. But will they think at all? How to find out? How to check if a machine thinks? The mathematician concluded that the only way would be to observe the “behavior” of such a machine – if we come to the conclusion that we will not be able to distinguish its behavior from human behavior, then such a machine can be considered as thinking [3]. Here we come closer to explaining the title of the mentioned movie – Turing came up with an idea that is brilliant in its simplicity. Before World War II, in Great Britain, a very popular social game was played – “Imitation game (https://en.wikipedia.org/wiki/Computing_Machinery_and_Intelligence ). The game involved three people: a woman, a man, and a “judge” who could be any other person. The woman and the man were locked in separate rooms in such a way that the judge did not know in which room was the woman and in which was the man. The game assumed that the judge could ask each person any questions he wanted, which he passed on in written form on cards passed around the door. The man and woman answered these questions, also in written form. The “judge’s” job was to guess who was a woman and who was a man.
Fig. The original concept of the imitation game source: https://en.wikipedia.org/wiki/Turing_test
Turing came up with the idea of replacing one person with a computer. The judge’s job would be to identify who was human and what was machine.
Fig. An imitation game with a computer as one of the people source: https://en.wikipedia.org/wiki/Turing_test
Turing proposed that communication be done via a so-called teletype instead of hand-written questions and answers (there was no Internet back then!). He even suggested simplifying the “test” by eliminating the human as the second player. In that case, the judge’s only task would be to identify whether the computer on the other side was so convincing that its answers (behaviors/reactions) were indistinguishable from the human’s. If a computer became indistinguishable from a human in judging another human, then we could conclude that such a computer in its own way thinks and can imitate behaviors indistinguishable from those of a human.
The Turing Test in a movie
While writing about the Turing Test it is impossible not to refer to its influence on culture – especially literature and cinema. The mentioned test is often shown on cinema screens.
In the 1982 classic “Blade Runner” directed by Ridley Scott and starring Harrison Ford, there is a famous scene when android hunter police officer Rick Deckard (played by Ford) – interrogates a suspect and tries to determine whether he is human or a replicant:
In “WestWorld” the androids are so perfect and indistinguishable from humans that it is only mentioned perfunctorily that they have already passed the Turing Test in the distant past:
Perhaps the title that touches the subject the most is 2015’s Ex Machina, almost entirely devoted to the test of a thinking machine:
Will the machine ever pass the Turing Test?
Turing believed that “in about fifty years’ time it will be possible, to program computers, with a storage capacity of about 109, to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning” [4].
Sometime later, in a BBC interview, Turing added that a computer would also be able to win against a human at grandmaster level chess before the end of the 20th century. Viewed from Turing’s perspective – playing chess with a computer and playing imitation may have been synonymous with “thinking”. Although Turing’s prediction of a computer winning at chess proved accurate (as we know in 1997 IBM’s Deep Blue computer defeated the then world champion Garry Kasparov in a match), no computer to date has ever passed a reliable Turing Test. It must also be admitted that the operation of Deep Blue is largely based on the principle of so-called “brute force” based on the introduction of a sufficiently large number of so-called “openings” and “ends” of a chess game into the computer’s memory, rather than on strict thinking. Although the complexity of the game of chess reaches 10120 and exceeds the number of atoms in the Universe, the number of openings and ends of the game is much smaller, not to mention the number of optimal moves. Hence the relative ease of building a computer program that would be able to beat a chess master. The problem was largely based on computing power and memory (today smartphones have similar computing power to Deep Blue…).
However, much more interesting was the computer’s victory in another puzzle game, GO, the complexity of which is estimated at 10720 combinations (which is 600 orders of magnitude more than for chess: 10 with six hundred zeros!). In this case, computing power and memory do not seem to solve the problem – but with the solution came artificial neural networks and techniques of so-called machine learning. In 2016, DeepMind built the AlphaGo program, which from March 9 to 15, 2016, crushed the then world champion Lee Sedol in five duels. The artificial neural network works on the same principle as the neurons in the brain and hence it can be said that it brings us closer to building a machine that can think (https://en.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol).
And here we return to the issue of a computer passing the Turing Test – although since 2014 every now and then we learn that some program “passed” the Turing Test, it soon turns out that there is a “catch” behind this information. As of today, we can say with certainty that no program has passed the Turing Test yet. Although it is estimated that modern supercomputers have reached the computational power of the brain (and even more) that is about 2x 1015 operations per second, and in the third decade of the XXI century personal computers will reach such a power, the problem lies in the software. The complexity of language (unlike chess or GO) is infinite. To pass the Turing Test a program will not only have to understand natural language, but also have knowledge of the world and its environment, common sense, their own experiences, and understand the laws that govern their behavior. When will this happen? I think soon (before the end of the 2020s), but even Turing was wrong… Or maybe you will try to do a Turing Test for our chatbot (https://glivia.com/chatbot) ?[MS1]
[1] The Enigma code was actually broken by Polish mathematicians: Marian Rejewski, Henryk Zygalski and Jerzy Różycki already in 1932 (!). They also worked out a method and a machine thanks to which it would be possible to break the cipher – the so-called “bomb”. Unfortunately, the budget of the Cipher Bureau of the Polish Army did not allow for the development of the project and finally the plans of the “bomb” and the cipher machine itself were handed over to the British Intelligence. On 25-26 July 1939 a conference of Polish, British and French cryptologists was held in Pyry near Warsaw during which Polish mathematicians explained the operation of the Enigma and the decryption machines and presented their cryptographic skills. This information was used to create a decryption machine called the “Turing bomb” – it is exactly this machine which was presented in the above-mentioned film.
[2] When Turing wrote the article, there were no artificial neural networks, which resemble the operation of the human brain. The first artificial neural network was the Perceptron developed by Frank Rosenblatt in 1957. Today, some cognitive tasks performed by computers such as recognition of speech, images or behavioral patterns are based on simulation of neurons and the so-called machine learning. So, it seems that the computer which will pass the Turing Test will be built on the basis of simulation of neurons and artificial neural networks.
[3] The 1950s were the time of the triumphs of behavioral psychology, which primarily studied the external factors of mental processes – that is, behavior. This trend seems to have strongly influenced Turing’s views and ultimate idea of how to verify whether a machine can think.
[4] https://www.csee.umbc.edu/courses/471/papers/turing.pdf
[MS1]Link kanoniczny?