The Turing test is not simply some hypothetical test, it's laidnout pretty specifically by Turing. From Wikipedia:
> Turing proposed that a human evaluator would judge natural language conversations between a human and a machine designed to generate human-like responses. The evaluator would be aware that one of the two partners in conversation was a machine, and all participants would be separated from one another. The conversation would be limited to a text-only channel, such as a computer keyboard and screen, so the result would not depend on the machine's ability to render words as speech. If the evaluator could not reliably tell the machine from the human, the machine would be said to have passed the test. The test results would not depend on the machine's ability to give correct answers to questions, only on how closely its answers resembled those a human would give.
I think the Turing test is fundamentally flawed. Humans inherently anthropomorphise everything. Hell, we regularly attribute complex internal thought processes that almost certainly are not there to our damn pets, or at least I do.
I'm not sure what the alternative is, but I feel pretty confident that the Turing test is insufficient.
I'd recommend reading the original Computing Machinery And Intelligence paper[1], rather than the Wikipedia article. It's extremely easy to follow, being that it's more of the nature of a thought experiment than of being hard computer science.
You're quite right that he lays out a fairly clear framework for how to play 'the imitation game', but it's still not at the level of 'experimental method'. The result will ultimately depends a lot on how good the human players are at catching out the computer in its lie. And it's only statistical, at best - there's no pass/fail that it can possibly produce, only a p value.
In general, I get the impression from Turing's general tone, though that he is trying to lay out the imitation game as an example, because he wants you to grapple with the fact that there may not be any better way to tell. That's kind of his point.
Turing was, by this time, I think, utterly convinced that universal computation is literally all that anything can do with information, so in his mind, the human player of the imitation game is just a Turing machine on the end of a wire; the computer player is just another Turing machine on the end of another wire; and if there are two Turing machines which produce statistically indistinguishable outputs, he's trying to suggest that they're basically equivalent.
Most of his paper is trying to overcome objections not to the idea that the imitation game isn't a valid test, but to the idea that a human mind is just a Turing machine.
So I actually think the point of the imitation game in his paper is not about the idea that this is a "good test". It's that it's a thought experiment that gets you to consider that when you reduce the human player to a box into which you feed input, and which produces output, that any information processing going on inside that box can be no more than Turing complete universal computation.
This was a helpful clarification, thanks. This paper is definitely on my reading list. Ahead of it for now is that book version of the Church-Turing hypothesis paper. I believe it's called The Annotated Turing. Supposed to be a great for self-learners like me who don't necessarily read a lot of science papers directly. Might pick it up tomorrow actually.
Turing was a remarkable person. So ahead of his time, sometimes I half-seriously wonder if he was a time traveller doing what he could with 40s and 50s technology...
> Turing proposed that a human evaluator would judge natural language conversations between a human and a machine designed to generate human-like responses. The evaluator would be aware that one of the two partners in conversation was a machine, and all participants would be separated from one another. The conversation would be limited to a text-only channel, such as a computer keyboard and screen, so the result would not depend on the machine's ability to render words as speech. If the evaluator could not reliably tell the machine from the human, the machine would be said to have passed the test. The test results would not depend on the machine's ability to give correct answers to questions, only on how closely its answers resembled those a human would give.
I think the Turing test is fundamentally flawed. Humans inherently anthropomorphise everything. Hell, we regularly attribute complex internal thought processes that almost certainly are not there to our damn pets, or at least I do.
I'm not sure what the alternative is, but I feel pretty confident that the Turing test is insufficient.