Animated fight scenes and the Turing Test


I recently gave some analogies to show the emptiness of the Turing Test in terms of deciding whether a computer is intelligent or not. Commenters managed to find some completely irrelevant ways in which my analogies were not exactly like the Turing Test. One of them went so far as to claim that the Turing Test wasn't about deciding whether computers are intelligent. In that case, fine, I can stop writing about it. But it certainly is used that way, again and again, by people who want to be able to claim, "Well, a computer passed the Turing Test, therefore, it is intelligent!"

In any case, I thought I would offer and even closer analogy, to make it harder for AI devotees to evade the main point of these examples. (That they will try to evade it, I have no doubt, for the will to believe is strong!)

So let us consider the equivalent of a Turing Test for simulated battles, such as those in The Lord of the Rings. In this case, the analogous test will involve having the judges sit in front of two monitors, watching a battle scene on each one. One of them will be a battle of real armies, while the other one will be computer-generated. Now, if the judge cannot tell which is the flesh-and-blood battle, that is an excellent sign that the people creating the simulation have done a very good job of simulating a battle.

But Turing Test advocates want to claim much more than that: the analogy to their claim, for the case of the two battles, is that if we can't tell the two battles apart, then the computer simulation is a real battle. When someone gets their head cut off in the computer simulation, if the viewer cannot tell if it was a real head and a real severing or not, then it was a real severing!

And of course, this is nonsense. The fact that one can simulate something so well that it can't be distinguished from the real thing does not make it the real thing!

5 comments:

  1. So the mind isn't a Turing machine?

    ReplyDelete
  2. If one of the definitions of "real battle" is that is has to have taken place in real time and space, then the simulated battle will not qualify even if passes the Turing test for "real battle". Such a test would indeed be vacuous.

    If there is some definitional aspect of "being intelligent" that the Turing test for intelligence misses then it would likewise be spurious. But as a reasonable definition of being intelligent might be "consistently passes objectively-run tests to determine intelligence" then the Turing test will (if run correctly) never produce false positives in the way that the other Turin test-like analogies for things other than intelligence you are describing might do.




    ReplyDelete
    Replies
    1. Rob, that is far from a reasonable (or useful) definition of intelligence.

      Delete
    2. Mike, yes probably not very well stated. What I meant was that for most other things there is an objective standard of some sort that allows you you verify if a Turing-type test has given an accurate result or not. When it comes to intelligence I can think of no such objective standard and therefore a Turing-type test (Perhaps the actual specifications provided by Turing may need to be updated now) may be the best pragmatic approximation to such a standard that we can come up with.

      Delete
  3. Yes! Good post Gene. I think this shot will get into the reactor of the Death Star, whereas in your last sortie you got shot down by TIE fighters.

    In case you didn't bother reading because you were sick of my insolence, I think in the comments here I spelled out why your critics were being unfair to you, but why I thought your post was lacking.

    Here's another example that maybe you will like: Suppose a man goes behind a screen and 1000 people can't tell the difference between his answers and those of a 25 year old woman. Would we conclude that the guy *is* a woman, and insistence otherwise is just superstitious belief in body parts?

    Now of course, the defender of Turing could come back and say, "Well, that's because we already have a good notion of what male/female means, and it's not irrelevant to focus on the constituents of the body. But when it comes to intelligence, it's pure prejudice to rule out human-built machines a priori..."

    OK, which sounds like saying, "Ruling out things a priori is fine when I do it, but not when Turing's critics do it. We all know that it must be possible to build a thinking, sentient computer, and so let's not rule that out."

    ReplyDelete