Alan Turing's Perspective on The Artificial Intelligence

“Can machines think?” This is the question Alan Turing seeks to discuss in his paper "Computing Machinery and Intelligence.” Because defining “machine” and “think” would be an arduous, likely unproductive process, he designs a simple game which he calls the “imitation game” and then adapts this game into something, now commonly called the “Turing test,” which he believes can be used to gauge a machine’s ability to think. I will describe and evaluate this test in regards to its ability to answer, or more specifically replace, the question “can machines think,” the objections as to the validity of the test, as well as my own opinion on the Turing test.

Turing originally started with the idea of a simple imitation game in which there are three participants: the interrogator, a man, and a woman. It is the job of the interrogator, who may be of either sex, to guess which one is the woman, the woman’s job is to make the interrogator believe she is the woman, and finally the man’s job is to deceive the interrogator into believing he is the woman. Both the man and woman can lie and steps are taken to keep the interrogator from getting any audio or visual cues, such as the transferal of all messages through a computer or typewriter. Later, Turing evolves this game by replacing either of the participants with a machine and changing the objective of the interrogator to guessing which participant is the machine.

This evolution eventually becomes the “Turing test.” Turing states that if an interrogator cannot accurately pass this test, by successfully choosing which participant is human, then the machine has proved its ability to “think.” Turing believes this test is a much better way to discuss machines’ ability to think, as the original question “Can machines think?” would require defining “machine” and “think” in such a way that a vast majority of people would agree on how the words are “commonly used”. This would require a large “statistical survey” and would likely result in a question “too meaningless to deserve discussion,” thus the idea of using his test as a replacement.

If we agree on the necessity of the test as a replacement, then we must determine the validity of this test. One objection, which Turing refers to as “The Argument from Consciousness”, refers to a machines ability, or inability, to derive answers, phrases, and sayings from emotions. The argument, presented by “Professor Jefferson’s Lister Orientation for 1949” asserts that “’no mechanism could feel’ (…) ‘pleasure at its success, grief when its valves fuse, be warmed by flattery’” etc. This suggests that the machine should have an emotional motive, or conscious choice, behind what it says in order for it to be truly thinking about its answers. However, Turing argues that this would mean the only way to prove a machine thinks “is to be the machine and to feel oneself thinking”. Because this is, of course, impossible and applies to any possibly sentient thing, machine or otherwise, it is similarly impossible to prove a human, other than oneself, thinks. Therefore, this solipsistic point of view “makes communication of ideas difficult” and “instead of arguing continually over this point,” one should just assume that “everyone thinks,” or more specifically, everyone that appears to be thinking, is actually doing so.

Another objection, which Turing calls “Lady Lovelace’s Objection,” argues that, unlike humans, machines don’t have any “pretensions to originate anything”. In essence, this objection, originally directed at Babbage’s Analytical Engine, states that machines do not have the capacity to show or do anything new because they can only do what we have told, ordered, or programmed them to do. Turing simplifies this objection to the variant “a machine can ‘never take us by surprise.’. He argues against this objection by telling about the many times a machine has surprised him, whether due to his own mistakes when programming or to unexpected results from a correctly functioning program. Elaborating, Turing states that, though these surprises may have been the fault of the programmer, a machine is still doing something “original.” And if one argues that the machine didn’t choose originality consciously, then “this leads us back to the argument from consciousness,” which “we must consider closed”.

Moving past this response, in section 7, titled “Learning Machines,” Turing suggests that another way to refute Lovelace’s idea of always unoriginal machines is to create a “learning machine”. A learning machine would, naturally, be able to present something that it wasn’t explicitly told to do, but was rather something it learned to do. He then states that he believes a learning machine could be created, given enough time, resources, and “advances in engineering,” in such a way as to “be adequate for the requirements” of the test and, through correct programming, may even be able to perform a childlike method of learning.

Though it is far from perfect, the Turing test is simple, the question and answer manner of the test allows it to be used for multiple purposes, and it is designed in a way to help eliminate cheating and bias. These factors are important because they make the enactment of the test itself easy and practical. As to the ability for the test to prove a machine can think, it is hard to argue with successful results. I don’t believe it will be anytime soon that a machine can be created to pass the test a large majority of the time. However, if a machine is able to pass the test then I would believe that that machine is “thinking.” After all, if one of the things that defines humanity is its ability to think, then I don’t think it is too much of a stretch to believe that a machine who shows a strong enough degree of humanity, strong enough that another person cannot distinguish it from an actual human, is thinking. All of these things lead to my acceptance of the validity of the Turing test as a means of getting closer to, if not answering, the question “Can machines think?”