As I understand it, Turning postulated that a (sight unseen) machine could be designed to respond intelligently enough to a random person, through writing, that it could fool that person into thinking it was an actual person. And that would be his test of the level of artificial intelligence programmed into that computer. This seems so pedestrian a measure. What is so remarkable about this test that it is referred to so often? And what was his initial point in proposing this test? I mean, isn’t the real test of artificial intelligence exactly that - that the machine appears to have human thought, feelings, sensibilities? The test seems so self-evident.
I think probably one of the most sticky aspects of it is that Turing made the statement quite a long time before the technology existed to even attempt the test. Thus it remained an out-of-reach (and therefore tempting) goal for quite a while, ingraining it on the collective conscious.
The test was not by any means Turing’s most significant work.
I suspect this is the core of your misunderstanding. Exhibiting human cognitive abilities, perceptions and emotions is merely one small aspect of AI, and not even necessarily the primary one.
For “machine”, are you envisioning a robot or a bare-bones computer with a face and arms?
If it’s just a bare-bones computer with the only input/output being a keyboard and a screen, then the pedestrian Turing test is the one that makes sense. Turing wasn’t talking about robots.
First of all, try reading the original article. It’s very short. I’ve always regarded Turing’s “imitation game” as sort of a “thus I slay the Gordian knot” approach. You can make a huge number of arguments against machine intelligence (many are summarized in Turing’s essay) , but once a machine is developed that performs in such a way that it can’t be easily distinguished from an actual person, then most of those arguments become fairly irrelevant.
By machine, I meant a computer with keyboard/screen input/output. Maybe I need to try to read the article to get a sense of what was so remarkable about his proposition. In naivite, it seems to me that what he proposed would be what anyone would propose in regard to the notion of artificial intelligence - “thinking machines.”
Well computers can now play chess better than humans, which certainly counts as artificial intelligence.
However it’s still possible for an experienced player to tell a computer progam apart from a human chessplayer, so they fail the Turing test.
In the sense that Turing is exploring, it is the primary one. The * discipline* of AI includes a lot of subfields such as machine learning, search, natural language-based information retrieval, robotics, vision, and multi-agent systems, but most people would agree that when the discussion of machine intelligence gets hot and heavy, what they’re arguing about is precisely the issues of emulating human cognitive abilities, perceptions and emotions.
In retrospect, it’s completely obvious. At the time, it wasn’t.
I think the point was that Turning expected machines to mimic a taking person to appear really soon. He wasn’t imagining a “superhuman AI” and setting a high bar, but a robot that would throw out words and wing it. If we’ll soon have machines that do this and fool us, he thought, it would have a huge consequence of how we felt about machines even if true AI did not exist.
Of course, he was wrong. It’s damn hard for a computer to talk. But maybe he had a point. We’ll have to deal with machines that seem conscious long before we deal with those that actually might be.
Though a lot of chess playing theory came from AI, I don’t think many AI people today would consider chess playing to be AI. Chess playing is done with a lot of excellent heuristics for search strategies, but nothing resembling intelligence. Perhaps a chess program could be crippled to look human, but no one has ever tried it.
The reason the Turing test is interesting is that he came up with it at roughly the same time Shannon was proposing ways of letting computers play chess. What Turing was doing was laying down when AI could be said to have succeeded even before AI really existed.
When I took AI in 1971, there was still a lot of hope of developing real AI. This was just before it was recognized as dealing mostly with vision, robotics, and heuristics. Almost all of the goals I learned (solving equations, giving directions) have been met, but I’d say we’re no closer to AI than we were then. Intelligence seems a lot more fundamental than the sum of the intellectual tasks we accomplish.
There’s a host of objections to the Turing test, probably the most famous one being Searle’s “Chinese Room” argument. There’s also similar objections to the OP’s, namely that the Turing Test focuses way too much on conversational ability. In light of this, several “extended Turing tests” were proposed.
As to why the test is so infamous, I’d guess it has nothing to do with any technical or philosophical reason. Rather, its name is nicely alliterate and is peripheral to a subject that the general public has a deep fascination with. The fact that it can be easily (mis)understood, and is named after a famous polymath and eccentric, is also a bonus.
I think the idea is that it is less of a test and more of a definition, or even a goal. AI research demands somewhat rigorous meanings behind words like “think” and “intelligence”, and ways to tell when a machine is thinking and/or intelligent. The Turing test provides a goal for those in the field to work towards, and a baseline for determining when a machine can be said to think.
ETA: Oh yeah, IANAComputerScientist.
From the paper: "In order that tones of voice may not help the interrogator the answers should be written, or better still, typewritten. The ideal arrangement is to have a teleprinter communicating between the two rooms. Alternatively the question and answers can be repeated by an intermediary. "
AFAIK, Turing never claimed a thinking machine was necessarily possible (and his work on the Stopping problem seems to indicate that there may be classes of problems not solvable by algorithmic computers). Rather he was simply trying to dismiss objections of the type “but even if it was a very clever mechanical parrot, it still wouldn’t really be intelligent”. Turing proposed that given unbounded opportunity to query a conversant, if it’s responses were indistnguishable from a human’s, it would be pointless to say it wasn’t “really” thinking. As pointed out upthread, some people have poked holes in that logic.
BTW: I thought of what might be the best possible question you could ever ask during the Turing Test: “If you were administering the Turing Test to someone else, to see if they were sentient, what questions might you ask them?”
The best possible reply would be “beside that one?”. 
Again, from the paper: “It will simplify matters for the reader if I explain first my own beliefs in the matter. Consider first the more accurate form of the question. I believe that in about fifty years’ time it will be possible, to programme computers, with a storage capacity of about 109, to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning. The original question, “Can machines think?” I believe to be too meaningless to deserve discussion. Nevertheless I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted”
Turing also addressed questions of the Stopping problem in the paper. His argument was approximately: “Suppose that a given machine can’t solve a given problem or would give the wrong answer. So what? Humans make mistakes too, and have problems they can’t solve.”
I welcome correction, but I think what is indicated is just that for any particular computer, there are problems that it can not solve. I think the stopping problem stuff is not taken to indicate that there are problems that no computer could solve.
-FrL-
From the OP:
Funny thing is, the test only seems self-evident to many of us today because Turing wrote that paper. At the time, it was not self-evident at all to most people.
You can imagine how this could have been. When I say that you are a thinking being, do I really mean just that you appear to have thoughts, feelings, sensibilities, etc.? No–I mean that you have the things that cause those appearances, namely, thoughts, feelings, and so on. I know you have them because I see the appearances, but by saying you have them, I don’t just mean you appear to have them. I mean you have them.
But Turing was arguing that we should re-interpret the question “Can computers think?” to mean “Can computers appear to think?” And he argued that since the evidence for affirmative answers to either question would be identical, there is no reason to make the distinction between the questions. This was–and to my mind, still is–not completely obvious.
-FrL-
I guess that in the intervening time, and in the absence of any familiarity on my part of AI, that such a test does seem self-evident, but I understand that at one time it did not. I gather that at the time Turing came up with his “brainstorm” that artificial intelligence was taken by many to mean a sort of logic machine, and he wanted to move the notion beyond that, closer to what we think we mean by intelligence - the human version. At the same time, I wonder why you needed X AND Y in addition to C to run the test. Isn’t the point to determine if X is a computer? Y is only present to provide some sort of triangulation, but there’s no guarantee that it is honest, so why do you need it? If C ultimately cannot determine if X is a computer or not, then the test has been completed, hasn’t it?
No one has put a real challenge to the original proposal, partly because intelligence is a matter of definition. We think we know what intelligence and self conciousness is, but one of Turings contributions was to make us doubt our assumptions. Searle’s objections are silly on several levels.