Turing Test and Machine Intelligence

Author: Peter Bradley

In 1950, Turing published "Computing Machinery and Intelligence" in the Philosophic Journal Mind. In it, he proposed a method by which we might answer the question 'Can Machines Think?'. His idea — which he calls the 'imitation game' — is really quite simple.

The Imitation Game is played by three participants: the interrogator, a human subject, and an artificially intelligent machine. The three are in separate rooms, and can only communicate via teletype. The goal of the interrogator is to determine which participant is the machine. They are allowed to ask questions of any sort. Turing hypothesizes that by the end of the century (the 20th, that is), computers will exist that can play the machine so well that "the average interrogator will not have more than a 70% chance of making the right identification after five minutes of questioning." (442) He does not suggest that this test definitively settles the matter, but he does claim that if these conditions obtain, one could speak of machines thinking "without expecting to be contradicted".

He then lays out nine possible objections to the view that machines may be intelligent.

Turing's Objection's and Replies:
Theology The objection claims that thinking is a function of the humans' immortal soul. Machines have no souls; therefore, they cannot think. Turing responds in two ways: first to note that if God is omnipotent, he could give souls to machines. Second, and more importantly, Turing's notes that a very similar argument was made against the Copernican theory of the movement of the planets. We rejected that argument in light of empirical evidence, and we should do the same in this case.
Heads in the Sand This objection claims that the consequences of a machine thinking is too dreadful to contemplate; therefore, we should hope and believe that they can never do so. Turing's reply is simply to note that this objection, like the objection from Theology, is based on the idea that Humans are necessarily superior to everything else. And that is an empirical thesis, subject to test.
Mathematics This objection turns on results, due to Gödel and Turing, that no formal system or (fixed, i.e. unchanging) machine can generate all arithmetical truths. Take, for example, the statement 'This statement cannot be generated by a fixed machine'. If the statement can be generated, the statement is false, and the Turing machine has proven a false statement. If the statement cannot be generated, the statement is true, but therefore escapes the scope of that Turing machine. The objection continues: human intelligence is not so limited; therefore, logical systems and machines cannot be intelligent. In response, Turing makes two points: the first is to question the premise that human intelligence is not limited in this way. According to Turing, we do not know if human intelligence is limited or not, and one way to determine if the machine is limited in precisely the same way as human intelligence is the already proposed Turing Test. The second is to note that a machine that is capable of inventing its own method of proof, or its own rules of syntactic processing, would be capable of generating all the arithmetic truths. Of course, it would also be capable of making mistakes. But, of course, so are we.
Consciousness The argument from consciousness proceeds as follows: No machine can have emotions, feel pleasure, grief, depression, etc. Intelligent humans clearly do. Therefore, machines cannot be intelligent. Turing's response is to claim that this argument is a denial of the validity of the Turing test itself. If a machine can talk intelligently about, say, a sonnet that it has composed, that that machine would be intelligent. After all, discussion about such feelings with other humans is the only evidence we have that they feel in the same way we do. If we are unwilling to attribute consciousness to a machine in such a scenario, we must also be unwilling to attribute consciousness to other humans.
Turing's response here is interesting, but somewhat unsatisfying. The argument from consciousness has become one of the central problems in the philosophy of mind in recent years, and philosophers like David Chalmers are perfectly willing to allow that there may be people who look and act just as we do, but have no conscious states. These 'zombies', as Chalmers calls them, might be equivalent to any machine that can pass the Turing test.
Various Disabilities This argument takes the form 'Not until a machine can do X, can it be said to be intelligent', where X is one of 'be kind, have a sense of humor, fall in love, enjoy strawberries, make mistakes, etc. Turing responds that none of these argument have any support, and all seem to be grounded in mere induction. In certain cases (like 'enjoying strawberries'), the objection is really a version of the argument from consciousness. The 'make mistakes' disability is of particular interest. Why should a machine be punished because it might be better at Math than a human?
Lady Lovelace Lady Lovelace, who, with Charles Babbage, invented the first mechanistic calculator made the following claim: "The Analytic Engine has no pretensions to originate anything. It can do whatever we know how to order it to perform". Intelligent humans perform original actions, and create original objects. Therefore, machines cannot be intelligent. Turing's response is that machines often do take their creators by surprise, and this surprising result can be analogous to original thinking.
Later thinkers have been unconvinced by Turing's response. Many more recent AI theorist have argued that an intelligent machine's machines tables should include some sort of probabilistic calculation, thereby allowing for original, unexpected results.
Continuity of the Nervous System This argument claims that there is a disanalogy between Turing machines and the Nervous system, and that this disanalogy is so great, that no intelligent machine would ever be possible. Turing's response is simple: yes, that's true, but if the machine passed the Turing test, would it matter? Turing is arguing here for a kind of primitive thesis of multiple-realizability. According to the thesis of multiple-realizability, or, at least a functionalist version of the multiple realizability thesis, what matters for intelligence is not the physical make up of the intelligent thing, but rather the fact that the thing functions in much the same way we do: including acting a certain way given a certain condition, having internal states in response to a certain input condition, etc. Turing's point is simply to press on the fact that the way in which the intelligent machine is physically put together should not change our intuitions regarding machines that can pass the test.
Informality of Behavior This argument claims that we do not know of a set of rules that will govern all human behavior, but in order to create an intelligent machine, you must create just such a set of rules. Turing's response is to note the difference between knowing about such a set of rules and there actually being such rules. There may be such rules, even if they are outside of the realm of current understanding.
Again, it is worth noting that this objection is part of what inspired the neural network modelers to seek an alternate way of modeling intelligence.
ESP This argument is simple: intelligent humans possess ESP, machines do not. Therefore, machines can't be intelligent. Surprisingly, Turing takes this argument quite seriously. But, he notes, the only way to prove ESP would be by empirical test, and any such test could be given to an artificial intelligence as well.

Which of the objections do you find the strongest? The weakest? Are Turing's responses satisfactory?

Alan Turing is easily one of the most fascinating people in the history of Computer Science. In addition to developing the Turing machine and proposing the Turing test, he was largely responsible for breaking the Nazi's code during WWII. In 1954, he committed suicide after being convicted of committing homosexual acts and sentenced to not only lose his military clearance, but also to undergo a course of estrogen injections. There is some controversy over his death - Turing's mother always contended that his death was not suicide, but rather an accidental overdose due to poor laboratory techniques. And there is some evidence that she might have been right. Click here to learn more about Alan Turing.

There have been a number of different programs that attempt to pass the Turing test. Take a moment to try some of these out, and see if you can determine if the respondent is a human or a computer.

Copyright: 2002

You've reached the end of this component.
[Explore Complete Module]