This paper critically discusses the possibility of the Turing test to determine whether the artificially created machine could think that is to be intelligent. Turing test is a tool with benchmark standards that allow declaring that the computer has intelligence based on its capacity to give some range of sensible verbal responses.
According to Shieber (2), Turing test cannot be a good test for stating that an agent who passed it can be called intelligent. He supports this conclusion by the statement that “the arguments against the sufficiency of the Turing Test for determining intelligence rely on showing that some extra conditions are logically necessary for intelligence beyond the behavioural properties exhibited by a subject under a Turing Test” (Shieber 1). The author underlines that the core idea of the test is checking the agent’s verbal behaviour adequacy. Knowledge inherent in intelligent systems is only a fragment of a particular field of knowledge relating to the fixed class of problems solved by the system.
Many opponents of the Turing’s theory such as substance dualists, the ‘Heads in the Sand’ and others stated that Turing test does not efficient tool for intelligence measuring while it rejects the supremacy of human beings over all creations and does not take into consideration that human behaviour and emotions cannot be programmed. Substance dualists argued that thinking process is performed by separately existing and non-material substance that create a person by combining with his material body in some specific way. Other representatives that appeal against the artificial intelligence stands for the evidence that machine does not have an ability to be emotional and transfer it in the works of art as a product of the flight of thoughts. Moreover, it is impossible to set the rules what decisions will be made by humans in different situations including the ability to change the direction based on the learning from the previous mistakes.
Turing decided that it made no sense to develop a comprehensive list of the requirements needed to create artificial intelligence, which also could be controversial, and proposed a test based on the fact that the behaviour of an object could be intelligent. Some day or other a machine cannot be distinguished from the behaviour of human beings judged upon the given responses to set stimuli. Agent will successfully pass the Turing test, if one experimenter, who asks him certain questions, will not be able to determine whether the written responses are received from another person or from a device (Hodges n.p.).
To be prepared to this intelligent test the machine should have means for processing natural language (Natural Language Processing – NLP) to successfully communicate with an interrogator; means of knowledge representation, which will store in memory of the computer everything it experiences or gets to know; tools automatically generating logical conclusions, providing the ability to use stored information to find answers to questions and bring new evidence; machine learning tools that enable the machine to reveal extrapolate signs of standard situations and adapt to new changed circumstances. In the Turing test the direct physical interaction between the experimenter and the computer is deliberately excluded, because the artificial intelligence does not require a physical imitation of man. But the so-called Turing Test provides full use of the video signal to the experimenter could test the ability of the test object to the perception, and also had the opportunity to present physical objects. Turing predicted that computers will eventually pass his test. He believed that by 2000 computer memory of 1 billion bits (about 119 MB) during the 5-minute test can fool the judges in 30% of cases. This prediction has not come true (Hodges n.p.).
Passing the Turing test by the programs
So far, none of the programs came close to passing the Turing test. Programs such as Elisa (ELIZA), sometimes makes people believe what they say to a man, such as the informal experiment called AOLiza. But such “successes” did not pass the Turing test. Firstly, people in these conversations did not have any reason to believe that they speak with the program, while in the present Turing Test person actively tries to determine with whom he converses. Secondly, the documented cases usually refer to those chat rooms, where many conversations are fragmentary and meaningless. Moreover, many users chat using English as a second or third language, and meaningless response programs are likely to write off of the language barrier. At last, many people do not know about Eliza and its programs like this and cannot recognize completely inhumane mistakes that these programs allow. Produced annually conversing competition between programs, which are the most human-like, in the opinion of the judges are awarded by the Loebner prize. There is also an additional prize for the program, which, in judges’ view, will pass the Turing test. The best result was shown by the program ALICE that won the Loebner Prize three times (Matthews n.p.).
Turing test results depend heavily on the interrogator’s judgment. The reliability of such a judgement is highly questionable. Moreover, “with a suitable dictionary such a machine would surely satisfy Turing’s definition but does not reflect our usual intuitive concept of thinking” (Shieber, 2). Consequently, it is a very controversial issue whether Turing’s idea of Imitation Game could really determine the level of participant’s intelligence. Taking into account the given above facts and thoughts the answer is rather negative. Too many “if” and “when” exist that will allow creating machine to pass this test.