The Turing Test was and continues to be a major source of inspiration in artificial intelligence research. It is highly relevant because it is simple to understand and apply. According to the Turing Test a computer is intelligent if it’s verbal ability is indistinguishable from a person’s verbal behaviour. Even since 1950, when Alan Turing published the article in which he described the test, the dispute regarding how adequate the test is to measure artificial intelligence has been unstoppable.The majority of academics and scholars are claiming that the Turing Test has multiple flaws and it is inappropriate as a sufficient condition for the acknowledgement of thinking in computers.
Alan Turing has developed the “Imitation Game” as a way to settle the question “Can computers think?” by providing a clear objective for the AI.
In this game a person converses with two parties (a man and a computer) that are in another room. The aim is that the person has to determine which party is the computer, while both the man and machine is trying to convince the interrogator that they are human beings. Turing believed that the purpose of AI is to create computers that can pass this test, more specifically, that they are indistinguishable from a human being regarding the verbal ability.
Turing did not argue the premise that the ability to convince an unspecified number of observers, with unspecified qualifications, for an unspecified amount of time would justify the conclusion that a computer thinks, he simply stated this. Turing speculated that by the year 2000 “an average interrogator will have 70% more chances to make the right decision”, i.e. in 30% of the cases the computer could be characterized as being intelligent. Currently the test lasts for 5 minutes, but Turing did not specify a time limit as being part of the test, and thus one might argue that if a computer thinks it should respond to as many questions as possible not only in term of 5 minutes. The year 2000 came and went by and the only improvements that were made were regarding chatbots.
In regard to the precision of his test to conclude that an artificially intelligent computer thinks, the opinions are strongly divided.
Those that are against it seem to be the most numerous. Some of them are:
– It doesn’t bring any notable contribution to the development and research of artificial intelligence because it doesn’t provide a clear purpose and it is too vague.
– It is highly subjective and biased. There is no evidence that the results can be reproduced and because it it called a “test” gives it a pseudo-scientific label that does not deserve.
– It relies to much on the use of language and does not consider other means of demonstrating intelligence or thought. It doesn’t give any indication whether the AI system is aware of what it is being discussed neither does it prove premeditation (to give a correct answer or to “fool” the interrogator).
– One of the most important goals of artificial intelligence (and the cognitive sciences in general) is to contribute to a complete theory of mind which would include an explanation of what intelligence is and how it is acquired. Turing Test is not needed for this purpose, and would be surpassed by a successful AI (which would be based on a theory of mind that incorporates a more comprehensive explanation on intelligence and thinking).
The pros point out the strengths of the test.
– The power of the test derives from its simplicity. Considering that neither philosophy nor psychology or neuroscience have been able to formulate definitions of intelligence or general thinking that are precise enough so that they can be applied to computers, Turing Test at least provides a framework of reference that can be measured. Thus it is a pragmatic solution to a complicated philosophical question.
– The test can be applied with the help of a lot of diverse topics. Its format allows the interrogator to use a variety of intellectual tasks. Turing wrote “question and answer method seems to be suitable for introducing almost every field of human activity that we want to include in the test.” In order to pass a Turing Test a well-designed AI computer must show reason, normal conversation skills and knowledge.
All in all, it is noteworthy that in the absence of a universally accepted alternative benchmark it is impossible to say what is and is not artificial intelligence. Any new software that comes from an institution that has AI in its title is usually called artificially intelligent when it it launched and who can contradict this assertion? Maybe it’s just another software.
Douglas Lenat, a former professor of computer science at Stanford and the founder and president of Cycorp said “Anthropomorphizing a computer program isn’t a useful goal.” He stated that unlike Turing, whose test is meant to fool an interrogator that a computer is actually human, he would want a computer to have common sense which is what Turing’s computer only claims to have.
The fact of the matter is that this common sense is a recurring theme when it comes to a unique, valuable, necessary but not sufficient element that an artificially intelligent computer should have in order to evolve.