Artificial intelligence experts Kevin Warwick and Huma Shah have found a serious problem in the standard Turing test. The test determines whether artificial intelligence is capable of thinking like a human. A press release on the work of British scientists is available at TechXplore. Scientific article published in the Journal of Experimental & Theoretical Artificial Intelligence.
Cybernetics have found that if the machine resorts to a kind of analogue of the Fifth Amendment (to the US Constitution), that is, it remains silent during the entire test, then it will pass the test. In this case, artificial intelligence, which simply refuses to answer the judge's questions, will be recognized as a thinking creature. However, any primitive machine that does not possess intelligence can be silent, so the Turing test stops working under these conditions.
The test to test the thinking abilities of artificial intelligence was proposed in 1950 by the English mathematician and cryptographer Alan Turing. According to the scientist, it is the only way to find out if a machine is capable of human thinking. Also known as the Imitation Game, the test asks a human judge to blindly ask written questions to two subjects: another human and an artificial intelligence. Based on the answers received, you need to determine which of them is who. If this fails, the machine is recognized as thinking.
Warwick and Shah examined the actual test tapes from the Lebner competition, in which the artificial intelligence refused to "talk" to the judge. In each case, the judge was unable to determine whether he was dealing with a machine or a person. The subject could be a person who chose not to answer questions, artificial intelligence, or a machine simply incapable of answering. In the tests considered, it was the third option that took place.
Scientists have called this problem the "Fifth Amendment" by analogy with the amendment to the US Constitution, which states that a person should not be forced to testify against himself.
The Lebner Competition is the main platform for conducting Turing tests. It involves the developers of chat bots and virtual interlocutors. Competition critics noted that the winning programs were designed to deliberately mislead judges by deliberately making spelling mistakes and other tricks.