tim (11/28/82)
I can't help feeling that people are missing the point of the Turing test. It doesn't try to make any great meta- physical point about the existence or nonexistence of some ephemeral quality called "intelligence". It attempts to avoid such silly and interminable debates by defining when we can |say| that a system is intelligent. How is that we can say that other humans are intelligent? We observe their behavior and make our own value judgements. How can we say that a computer system is intelligent? We observe its verbal behavior (since physical behavior is likely to be rather alien) and make our own value judgments. However, there is a great possibility of prejudice here, since some people will never say that it's intelligent if it is a computer, and some people are easily amazed. To remove this prejudice, we add a level of indirection. We say that if a computer can successfully imitate a system we've all agreed is intelligent, the computer is intelligent. This is not an arguable definition: if we can't tell the computer from a person, then either both are intelligent or neither are. (However, if you start off with the premise that "People just |are| intelligent; this is not something that needs to be observed", the whole thing falls to the ground. This premise is not at all scientific, though, so it can be ignored.) Note that the test is conclusive but not inclusive; any system that passes it can be said to be intelligent, but failure does not neccessarily indicate lack of intelligence (unless you decide that human intelligence is the only real kind.) To sum up: The Turing test merely means applying to computers the same criteria that you used to decide humans were intelligent. Nothing more, nothing less. Tim Maroney unc!tim PS. Please put any further discussion on net.misc.
pcmcgeer (11/29/82)
Whillikers. Think about this: isn't the smooth, powerful performance of Berkeley Unix running on a VAX 11/780 a much better indication of intelligence than the conversational skills of most people you know? Searle's right, but for the wrong reasons. Computers are already more intelligent than most people will ever be. Rick.
ark (12/02/82)
Mark reproduced the following dialogue from Turing's paper: Q: Please write me a sonnet on the subject of the Forth Bridge. A: Count me out on this one. I never could write poetry. Q: Add 34957 to 70764. A: (Pause about 30 seconds and then give as answer) 105621. Q: Do you play chess? A: Yes. Q: I have K at my K1, and no other pieces. You have only K at K6 and R at R1. It is your move. What do you play? A: (After a pause of 15 seconds) R-R8 mate. A wonderful subtlety, often missed, in this dialogue is that 34957 + 70764 is 105721, not 105621.
steveb (12/02/82)
Tim Maroney (unc!tim) : "Note that the test is conclusive but not inclusive; any system that passes it can be said to be intelligent, but failure does not necessarily indicate lack of intelligence" I would maintain exactly the opposite. If I *can* tell that a system is just that, following simple `programmed' rules, then surely that system is not intelligent. If, on the other hand, the system is complex enough that I cannot determine what its underlying motivations are, is it then necessarily intelligent? No, it's only shown one piece of (rather weak) evidence that it is. Even if this evidence stands the test of time and repeated scrutiny from experts, the best one can say on the basis of it is that the respondant in question is probably intelligent. Accepting your interpretation leads one to the ridiculous assertion that DOCTOR is intelligent, because (as is well known) with many naive people it passed with flying colors (i.e., even after being told that they were talking to a computer, refused to believe it).
jerry (12/04/82)
The rumor that various programs (Doctor and Parry) have "almost" passed the Turing test has been around for quite a few years. It is definitely false. Two essential elements of the test are the "control" human to which the artificial intelligence is compared and a full understanding of the purpose of the test by the two humans. Doctor is a program that simulates a "Rogerian psychiatrist". It is "common knowledge" that naive "patients" will believe it really is a psychiatrist. However the program does not stand up to even the mildest test of it's intelligence. (That assertion is based on personal experience.) For example, if you say "Are you a person?" it is likely to respond with "Do you think I am a person?") Parry is a much more sophisticated program that simulates a "paranoid" individual. The experiment using it (reported, I think, quite a few years ago the AI Journal) had psychiatrists "interview" it and sent transcripts of these interviews together with transcripts of real patients to other psychiatrists who were asked to judge which was the real patient. While the experiment contained controls, the human participants did not have the intention required by the test. I don't have any first hand knowledge and Parry is more sophisticated than Doctor, but I doubt that it could stand up to anything but the mildest of attempts to test its intelligence. (A "cute" aspect of the program is that it would become silent if it thought the interviewer was being hostile. This being a natural reaction of a paranoid person.) Jerry Schwarz eagle!jerry
laman (12/04/82)
The only way not to be misunderstood is not to submit anything. And then again not submitting anything could be misunderstood, just as badly as this note can be.