3 ms·

Ask HN: Is the Turing Test still the gold standard of AI?

2 points by TekMol 7y ago

With all the stuff that is going on in AI land recently, I hear little about the turing test and advances in conquering it. Is it still the gold standard to measure if we "have made it"?

ktpsns 7y ago
The Turing test is quite abstract. Talking with Siri can be seen as a Turing test (would you think the machine passes it? I would not). Captchas are obviously an annoying one (which are in some crisis because they need to get harder and harder). Automatic driving can also be seen as a Turing test from the perspective of the other drivers. All those problems are bleeding edge of research.
mindcrime 7y ago
I would argue that the Turing Test was never really the "gold standard." It has, and has always had, a lot of issues. First and foremost to me, is that "teaching" a computer to "pass the Turing Test" amounts to teaching it to lie very well.
gus_massa 7y ago
Nobody have a good definition of intelligence or general intelligence. How are you sure that your friends are really intelligent and they are not faking it? A few year ago, playing chess was a task difficult enough to require intelligence, until someone solved it and it became only a brute force trick. Later, playing go was a task difficult enough to require intelligence, until someone solved it and it became only a brute force trick. You can add image classification, translation, whatever task or test you want to add. When it is unsolved it requires natural or artificial intelligence. Once it is solved is only a trick. https://en.wikipedia.org/wiki/AI_effect https://en.wikipedia.org/wiki/AI_effect We only agree that humans are intelligent, so the only test that we can make to ensure that a computer is intelligent is that the computer can fake a human. Anything else will be dismissed as a trick. And when we have a computer that can pass the Turing test it will dismissed as a trick anyway. The problem is that making a computer that can fake a human is more complicated than we expected (or perhaps less complicated than we expected, it depends on whom you ask). So the current advancement are in small subtask. When we have more computing power, more subtask solved, more clever ideas to solve subproblems and more time to make attempts, we will be able to make a computer that pass the Turing test. (Or not, nobody is sure, I guess we can.) How about a nice game of chess?

2 points by TekMol 7y ago

With all the stuff that is going on in AI land recently, I hear little about the turing test and advances in conquering it. Is it still the gold standard to measure if we "have made it"?

ktpsns 7y ago

The Turing test is quite abstract. Talking with Siri can be seen as a Turing test (would you think the machine passes it? I would not). Captchas are obviously an annoying one (which are in some crisis because they need to get harder and harder). Automatic driving can also be seen as a Turing test from the perspective of the other drivers. All those problems are bleeding edge of research.

mindcrime 7y ago

I would argue that the Turing Test was never really the "gold standard." It has, and has always had, a lot of issues. First and foremost to me, is that "teaching" a computer to "pass the Turing Test" amounts to teaching it to lie very well.

gus_massa 7y ago

Nobody have a good definition of intelligence or general intelligence. How are you sure that your friends are really intelligent and they are not faking it? A few year ago, playing chess was a task difficult enough to require intelligence, until someone solved it and it became only a brute force trick. Later, playing go was a task difficult enough to require intelligence, until someone solved it and it became only a brute force trick. You can add image classification, translation, whatever task or test you want to add. When it is unsolved it requires natural or artificial intelligence. Once it is solved is only a trick. https://en.wikipedia.org/wiki/AI_effect https://en.wikipedia.org/wiki/AI_effect We only agree that humans are intelligent, so the only test that we can make to ensure that a computer is intelligent is that the computer can fake a human. Anything else will be dismissed as a trick. And when we have a computer that can pass the Turing test it will dismissed as a trick anyway. The problem is that making a computer that can fake a human is more complicated than we expected (or perhaps less complicated than we expected, it depends on whom you ask). So the current advancement are in small subtask. When we have more computing power, more subtask solved, more clever ideas to solve subproblems and more time to make attempts, we will be able to make a computer that pass the Turing test. (Or not, nobody is sure, I guess we can.) How about a nice game of chess?