What is the Turing Test, and Why is it so Difficult to Pass - Report Example

What is the Turing Test, and why is it so difficult to pass? by What is the Turing Test, and Why is it so Difficult to Pass? One of the most difficult challenges in developing Artificial Intelligence (AI) is to create a machine that “thinks” as intelligently as humans do. However, devising a definition for the word “think” itself is quite a task. This is because it is yet unclear as to what comprises a human being’s thoughts, and what is the driving force behind his/her intelligence. Is it a manifestation of the immortal soul or is it just a complex network of nerves comprising the nervous system? To create an intelligent machine or a computer, it is necessary to grant it with thinking capabilities that are at par with humans. If such an intelligent machine is ever created, how can we test whether it can think on its own? How can it be certified as Artificial Intelligence? Alan Mathison Turing, a computer analyst, mathematician and cryptoanalyst, provided a simple solution to this problem. In a paper published in the Journal Mind, in 1950, Turing suggests that rather than creating complications by using the word “think”, defining it, or asking whether machines can “think”, it is easier to develop a task that requires thinking, and testing whether a machine can succeed in that task. In Turing’s own words, “Instead of attempting such a definition I shall replace the question by another, which is closely related to it and is expressed in relatively unambiguous words” (Turing, 1950, p. 433). These “unambiguous words” were in fact the “imitation game”, now known as “Turing’s Test”. This test suggested by Turing has been used ever since to test artificial intelligence. In spite of the technological advancements since the Turing test was first published, no machine has yet passed the test. Turing’s paper has been a frontrunner in all publications and research material on Artificial Intelligence, and has been cited in innumerable publications since the moment of its inception. What is the Turing Test? The Turing Test was suggested by Alan M. Turing while he was employed at the Computing Laboratory in Manchester University (“The Alan Turing Internet Scrapbook”). Turing put forth the idea that machines could be devised to think and be capable of “rivaling human intelligence”. He writes: The original question, "Can machines think?" I believe to be too meaningless to deserve discussion. Nevertheless I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted (p. 440). This idea was revolutionary and contrary to the popular notions and beliefs of his time (“The Alan Turing Internet Scrapbook”). Turing suggested an imitation game that could be used to test whether a computer is as intelligent as a human being. The directions of the game are as follows: A human and a machine, labeled ‘X’ and ‘Y’ are placed in separate rooms. Another human, who is the judge/interrogator/observer, is placed in another room and is unaware of who is in which room. He only knows the human and the machine by the labels, ‘X’ and ‘Y’. The judge/interrogator will have to find out who is the human and who is the machine based on the responses he gets from both, the human and the machine. The machine will try to make the judge conclude that the other human is the machine. The human will try to help the judge in identifying the machine (Turing, 1950). In simple words, a computer and a human should be placed in two separate rooms and should communicate through textual messages with a third human, who is the judge or the observer. The judge would interrogate both, the human and the machine. The judge will then decide who is in which room. If the judge fails to make the right judgment and fails to distinguish between the human and the machine, the machine is said to be successful and will have passed the Turing Test. According to Turing, if the computer succeeds in convincing the judge that it is human, it is said to be intelligent. Thus, the machine would be certified to have “artificial intelligence”. Although Turing did not use the phrase “Artificial intelligence”, his conclusions imply the same. Significance of the Turing Test In his paper, Turing asserts that in fifty years, computers with high storage capacities, up to 109, would be created and programmed in such a way that they could play the imitation game with ease. It will be difficult for the judge to have more than a 70% chance of identifying the computer correctly after the first five minutes of interrogation (Turing, 1950, p. 440). The Loebner Prize Competition, held annually for computer programs tests this assertion. Computer programs are analyzed and subjected to the Turing Test. In spite of innumerable entries and cutting edge technologies, they “come nowhere near the standard that Turing envisaged” (Oppe and Dowe, 2011). The significance of Turing Test lies in the fact that it analyzes the ability of a computer to think logically and continue a conversation with a human being without making him realize that he is conversing with a computer. The logic behind this test is that if a machine possessing artificial intelligence can converse with us like any other human does, it possesses intelligence that is equal to that of humans. Shieber (2004) describes these conclusions as follows: Humans are intelligent. The conversational verbal behavior of humans reveals that intelligence. If an agent has behavior of a type that can reveal intelligence and that is indistinguishable from that of an intelligent agent, the former agent is itself intelligent. Any agent that passes the Turing Test has conversational verbal behavior indistinguishable from that of humans. Therefore, any agent that passes the Turing Test is intelligent (p. 409). The Turing Test is, therefore, the ultimate test for artificial intelligence. It provides a “clean and novel test for intelligence [and] neatly sidesteps the vast philosophical quagmire of the mind-body problem” (French, 1990, p. 53). Once a machine passes the Turing Test, it implies that it possesses intelligence equivalent to that of humans. It would be the biggest breakthrough of computer science and programming. Why is the test difficult to pass? No machine has really passed the Turing Test until date. In fact, the Loebner Prize Competition that is held to encourage programmers to create AI that can pass the test, is an “embarrassment precisely because we are so far from having a computer programme that could carry out a decent conversation for a period of five minutes” (Oppe and Dowe, 2011). The Turing Test is considered to be extremely difficult for machines for a variety of reasons. One of the primary reasons is that the Turing Test is not general and “defines intelligence purely in terms of behavior” (Copeland, 2000, p. 519). Furthermore, to pass the test, the computer will have to experience the world like humans do, to acquire intelligence equivalent to an adult human (Frech, 1990, p. 53). Without these experiences, it is impossible for a computer to acquire the kind of intelligence we expect. In order to achieve this, the computer will have to be fed with a program that could replace the years of experience and learning that imparts intelligence to human beings. According to Costa (1998), to pass the Turing Test: First of all, the computer would have to be able to learn about its environment. It would be impossible to program a computer about every fact in the world, every cultural bias and trend, etc. The computer would, in a sense, have to have an ability to deduce information and apply it to relevant situations. In other words, to pass the spirit of the Turing Test, the computer would have to experience life like a human would (para. 7). As is evident, a computer cannot pass the Turing Test until it sees the world from the “eyes of a human”, and lives it like a human would, which is impossible. Human beings are self conscious and possess cultural understanding and survival instincts. These characteristics impart humans with the intelligence they currently possess. Unless a computer “grows up” like a human does, it is impossible to develop the kind of intelligence, understanding and consciousness that a human possesses. French (2000) writes, “One important question is: to what extent is the level specified by Turing sufficient to adequately probe the deeper sub cognitive and even physical levels of the candidates?” (p. 121). According to him, if “carefully worded questions” are asked in the test, even the minutest physical differences can be caught from the responses of the human and the machine. For instance, asking a question like “Rate on a scale of 1 to 10 how much keeping a gulp of Coca-Cola in your mouth feels like having pins-and-needles in your feet” can easily test the physical differences between a human and a machine. Even if the machine makes a few correct guesses, it is impossible for it to make such guesses again and again for a broad range of similar questions. In order to answer all such questions as convincingly as the human, the machine will have to have experienced the life of a human. It is impossible to feed every bit of such data via a computer program. Therefore, passing the Turing Test is an extremely difficult and an almost impossible task. Many present day innovators and programmers are focusing on the behavioral training of computers rather than just programming them with rules and instructions. The computers are given the basic behavioral training and conditioning to encourage self-learning. The basis of such an approach is that by providing computers with “basic behavior” and allowing them to “build on that behavior”, it is possible to enable them to learn the way humans do, and bring them closer to passing the Turing’s Test (Costa, 1998). How successful such strategies will be, only time can tell. As of now, only two computer programs have come closer to passing the test, the ELIZA and PARRY. These programs simulate human behavior. Weizenbaum’s ELIZA is a program that simulates the discussion between a psychiatrist with his patients, and Colby’s PARRY is a program that simulates a “paranoid schizophrenic by means of a large number of canned routines” (French, 2000, p. 122). Though these programs came very close to passing a resricted Turing’s Test in the Loebner Prize Competition, the test still remains invincible. Many controversies abound around the Turing Test, questioning its validity as a test of intelligence. According to French (1990), the sub cognitive and cognitive levels of thinking are intertwined in the Turing Test. He writes, “It is this essential inseparability of the sub cognitive and cognitive levels – and, for that matter, even the physical and cognitive levels – that makes the Turing Test a test for human intelligence, not intelligence in general”. He suggests using a “Theory” of intelligence rather than a “Test” for intelligence. In spite of the many controversies and limitations propounded by many, the Turing Test continues to be popular worldwide. This is because the test comes very close to assessing artificial intelligence accurately, in the context of the human world. Some computer scientist and logicians maintain that it is possible to pass the Turing test with the right approach to programming and algorithm development. In a study, Stoica (2008) points out, “Under general assumptions, the Turing test can be easily passed by an appropriate algorithm” (p. 1). For any test, if several general conditions are satisfied, an algorithm can be written to pass the test. Stoica, therefore, suggests a test that is complementary to the Turing Test that can measure our understanding of the human mind (p.1). This will be helpful in devising a program that can pass the Turing Test. Therefore, to say that passing the Turing Test is impossible, would be inaccurate. With technological advancements in full swing and the unyielding quest for developing artificial intelligence, it may not be long before a computer program passes the Turing Test. Till then, new ways of defining and studying human behavior, reasoning, logic and thinking, will have to be explored. It is only through a proper understanding of human logic that artificial intelligence can be brought at par with human intelligence. The “thinking machine” as envisaged by Turing may not remain a distant possibility. References Copeland, B.J., 2000. The Turing Test. Minds and Machines, 10(4), p.519-539. Available at: http://www.mendeley.com/research/the-turing-test-2/#page-1 Costa, D., 1998. What if a Robot passed the Rigorous Turing Test? indyflicks.com. Available at: http://www.indyflicks.com/danielle/papers/paper05.htm French, R.M., 1990. Subcognition and the Limits of the Turing Test. Mind, 99(393), p.53-65. Available at: http://www.mendeley.com/research/subcognition-and-the-limits-of-the-turing-test/#page-1 French, R.M., 2000. The Turing Test: The First Fifty Years. Trends in Cognitive Sciences, 4(3), p.115-121. Available at: http://leadserv.u-bourgogne.fr/files/publications/000279-the-turing-test-the-first-50-years.pdf Oppy, G. and Dowe, D., 2011. The Turing Test.  The Stanford Encyclopedia of Philosophy (Spring 2011 Edition). Available at: http://plato.stanford.edu/archives/spr2011/entries/turing-test/ Shieber, S., 2004. The Turing Test: Verbal Behavior as the Hallmark of Intelligence, The MIT Press. Available at: http://acl.ldc.upenn.edu/J/J05/J05-3006.pdf Stoica, Cristi (2008) Turing test, easy to pass; human mind, hard to understand. [Preprint] http://philsci-archive.pitt.edu/4345/1/TuringTest.pdf “The Alan Turing Internet Scrapbook”, n.d. turing.org.uk. Available at: http://www.turing.org.uk/turing/scrapbook/test.html Turing, A.M., 1950. Computing Machinery and Intelligence. Mind, 59(236), p.433-460. Available at: http://www.cs.umbc.edu/courses/471/papers/turing.pdf Read More