DIFFERENT WAYS OF CONSTRUING THE TURING TEST

In spite of the fact that Turing's original article (Turing, 1950) is written in very clear and direct prose, there are a number of different ways to interpret the claims made in it. I am not, in this article, going to discuss what I think Turing's actual intentions were, but instead I will focus on three different ways of construing the results of the Turing Test that have been prominent in its application. I will assume for the sake of this article that the test itself is unambiguous. My discussion concerns the question: How do we interpret a positive result? On one natural construal, the test gives us a way of telling whether or not we have successfully simulated some human cognitive capacity, some human form of intelligent behavior that manifests thinking. If the machine can perform in such a way that an expert cannot distinguish the performance of the machine from the performance of a competent human, then the machine has successfully simulated the intelligent behavior of the human. Indeed, if our aim in Artificial Intelligence is to produce machines that can successfully simulate human intelligence then the Turing Test gives us a criterion for judging our own success and failure. I do not see how one could object to such a test. If the question is whether we have actually simulated, i.e., imitated, human behavior then, so construed, the Turing Test seems trivially right: If you can’t tell the difference between the original and the imitation, then the imitation is a successful imitation.