In 1950, Alan Turing invented a test that is crucial in the development of Artificial Intelligence, even 70 years after its creation. Turing’s test was first presented in his paper Computing Machinery and Intelligence, which questioned the ability of machines to think for themselves and where he proposed that a sufficiently trained computer could display near-human intelligence.
The Turing test studies machine intelligence, which can be almost indistinguishable from that of human beings, based on language. A human being acting as an evaluator has a casual conversation with a computer and a human being, without knowing which is which. If the evaluator cannot distinguish between the two, then the machine passed the test.
The Imitation Game
Alan Turing started with something called “the imitation game,” where a person must guess which of his interlocutors is a man and which is a woman. It requires three players: Player A, Player B, and the interrogator, Player C.
While player A tries to confuse his interrogator with misleading answers, player B responds intending to get the player to come up with the correct answer. A tries to confuse them and B tries to help. It was based on that idea that Turing formulated two key questions: what if the interlocutor trying to mislead the interrogator was a machine? And, what would happen if both, the machine and the human, were trying to fool the interrogator?
The Turing test
In the Turing version, the objective is to determine which of the two interlocutors is a computer. On one side of the screen is a human being who acts as a judge. His job is to converse with several interlocutors, unknown to him, on the other side of the chat. Most of them are human beings, but one will be a robot that tries to pretend to be a human as well. To pass the test, the machine must manage to be mistaken for a human more than 30% of the time.
The Turing test is not foolproof. For one thing, it assumes that the intelligence of humans and machines is equal and can encourage a kind of “artificial stupidity:” to pass the test, many machines imitate typos and misspellings like those made by humans. Thus they use imitation to pass the test, and not really intelligence. On the other hand, not all computers need to demonstrate that they are intelligent through the use of language, as Artificial Intelligence does not necessarily rely on these skills to function.
Story originally published in Spanish in Cultura Colectiva