In this article, we discuss the Turing Test and how it is used to determine the intelligence level of a machine. If you haven’t checked out our other articles on artificial intelligence, we highly recommend gaining some understanding of AI first.
Artificial intelligence makes it possible for machines to learn from experience, adapt to new information and perform human-like tasks. The examples of AI that you hear about today – from chess-playing computers to autonomous vehicles – rely heavily on deep learning and natural language processing (NLP). Computers can be trained to accomplish specific tasks by processing huge amounts of data and determining patterns and trends in the data.
But how do we know when artificial intelligence surpasses human-like intelligence or achieves a level of intelligence where we are unable to differentiate it from a human?
Alan Turing
Alan Mathison Turing was a mathematician, cryptanalyst, philosopher and computer scientist who is most renowned for his influence in the development of theoretical computer science. He was also highly influential in the conceptual development of artificial intelligence, algorithms and computation. He is widely considered to be the founding father of computer science and artificial intelligence, having published a paper in 1951 where he posed the question: ‘can machines think?’
During the Second World War, Alan Turing was a leading participant in the breaking of German ciphers at Bletchley Park. Turing lead efforts to decode German message transmissions. The main focus of Turing’s work at Bletchley Park was in cracking the ‘Enigma’ code. The Enigma was a type of enciphering machine used by the Germans during WWII to send messages securely. Because the Germans knew the messages were vulnerable to decryption, they made it even more difficult for the allies to decipher the codes by changing the system daily.
In order to substantially decrease the time it took to decrypt German code, Turing played a key role in inventing a machine known as the Bombe. The machine helped decipher German codes and the intelligence gained helped British efforts in winning the war.
What is the Turing Test?
Alan Turing, in a 1951 paper, proposed a test called “The Imitation Game” that might finally settle the issue of machine intelligence. The first version of the game he explained involved no computer intelligence whatsoever. Imagine three rooms, each connected via computer screen and keyboard to the others. In one room sits a man, in the second a woman, and in the third sits a person, an interrogator. The interrogator’s job is to decide which of the two people talking to him through the computer is the man. The man will attempt to help the interrogator, offering whatever evidence he can to prove that he is human. The woman’s job is to trick the judge, so she will attempt to deceive him, and counteract her opponent’s claims, in hopes that the judge will erroneously identify her as the male.
What does any of this have to do with machine intelligence?
Turing proposed an altered version of the test, in which instead of a man and a woman as participants, there was a human, of either gender, and a computer at the other terminal. The interrogator job is to determine which of the participants is human, and which the machine. Turing proposed that if an interrogator were less than 50% accurate, given that they are as likely to pick either option, then the machine should be considered intelligent.
Figure 1: The standard interpretation of the Turing Test, in which player C, the interrogator, is given the task of trying to determine which player – A or B – is a computer and which is a human. The interrogator is limited to using the responses to written questions to make the determination.
In 1948, Alan Turing, along with his former undergraduate colleague, D.G. Champernowne, began writing a chess program for a computer that did not yet exist. The program was completed in 1950 and was dubbed the Turochamp. Two years later Turing tried to implement it on a Ferranti Mark 1. However, the computer lacked enough power and was unable to execute the program. Instead, Turing ran the program by flipping through the pages of the algorithm and carrying out its instructions on a chessboard. This method took half an hour per mode and the game was recorded. Turing’s chess program played a recognisable game of chess but it lost to Turing’s colleague Alick Glennie. However, it is said to have won a game against Champernowne’s wife, Isabel. This program was a significant and lasting contribution to the debate regarding artificial intelligence.
Figure 2: One of the prevailing problems of the Turing test is that it does not directly test whether a computer is behaving intelligently.
The Turing Test only tests whether the computer behaves like a human being, for which there could be countless parameters to examine. Human behaviour and intelligent behaviour aren’t precisely the same, which is why the test can fail to accurately measure if a computer is intelligent.
This is for the following two reasons:
Some human behaviour is unintelligent
One of the requirements is that the computer is able to perform all human behaviours. This includes both unintelligent and intelligent behaviours. It tests for behaviours that aren’t necessarily considered as intelligent such as the ability to insult, make a joke, stutter, take moments to think of an answer, be unable to give an opinion, etc. It also tests for temptations to lie, or make typing mistakes. These are all the unintelligent things that the Turing Test is testing for.
Some intelligent behaviour is inhuman
The other issue is that the Turing test doesn’t test for highly intelligent behaviours, such as the ability to solve difficult problems or come up with original suggestions. It specifically requires deception on the computer’s part – if the machine is more intelligent than a human it must purposely avoid appearing too intelligent. If it were to solve a computational problem that is essentially impossible for a human to solve, then the interrogator would know the program is not human, and as a result the computer would fail the test.
Because of these underlying weaknesses of the Turing test, several test alternatives that would be able to evaluate super-intelligent systems have been suggested.