Lars Marius Garshol (larsga@ifi.uio.no)
Thu, 9 Oct 1997 23:03:04 +0100
(I changed the subject of this thread so that people who don't want to
read it can kill it, as we're slowly moving off-topic here. I'm still
waiting for Bos AI article, which will hopefully give some further
insights into how the essential AI problem can be formulated in terms of
the MoQ. IMHO, that's what we really need. And, Bo (or anyone else), please
don't post that under this subject in case people really kill this thread.)
* Lars Marius Garshol wrote:
ö
ö I don't agree with this. If this is to be a criticism of the Turing test
ö you'll have to specify what kind of information is "stripped away" that
ö makes it more difficult to decide whether the entity on the other side
ö of the screen/keyboard is intelligent or not.
* Magnus Berg
ö
ö The human at the other end of the
ö keyboard has to intellectualize all answers before typing them.
This is excactly what I disagreed with. Verbalizing is distinct from
intellectualizing, and even intellect has DQ. Think of poetry! There's
nothing in the test that says the judges can't ask the participants
to write a poem. In fact, it encourages it.
IMHO, this is where it can very easily fail. Imagine this:
JUDGE: Write some haiku for me, please.
PARTY1: Sorry, I don't speak Japanese.
PARTY2: What is haiku?
Now, which one is the human? Both are perfectly reasonable answers,
and easy to automate.
ö Ok,
ö this is true for most types of communication but I think the written
ö language is more static than, say the spoken one. You have to think
ö one step further to write a thought than to speak it.
Sure. This sounds like a possible starting point for a rejection of the
Turing test. You still have to specify excactly _how_ this invalidates
the Turing test. Why does it make it possible for a non-intelligent
being to pass the test?
ö You still talk like intelligence is something you can define in
ö the same way as you define the operation plus or something. Imagine
ö that a definition *is* possible, what would it say?
We're testing a definition right now. "An entity is intelligent if it
can pass the Turing Test."
ö List all the
ö right answers to the questions? Then it would be quite easy to fake
ö such an intelligence, just as all computer manufacturers tries to
ö optimize their computers to the current Benchmark tests. This is
ö very similar to Doug's rational criticism of the test.
Mhhm. Imagine this:
JUDGE: What does this make you think:
Snowfall,
unspeakable, infinite
loneliness.
How do you list an answer to that?
ö I think the minimum requirement for true AI is that it has to
ö be able to surprise its creators. The answers must not be rationally
ö deducible from the input.
This happens even with something as simple as a web server. Optimizing
one for full speed can be very difficult. By this criterion alone,
innumerable programs are already intelligent.
In fact, I once made a Scheme interpreter that sometimes surprised me
by rebooting my computer in the middle of a computation, and I never
could figure out what did it. :-)
If you interpret the last sentence strongly enough you have effectively
excluded all kinds of computing machinery, with the possible exception
of chaotic and quantum ones. (I have a hard time seeing how the quantum
one can have any sort of useful interface. The chaotic one _might_ be
theoretically possible.) There is nothing a computer can do that a Turing
machine cannot do (given a little time :).
(For those of you who don't know what a Turing machine is: it's a simple
computer that you can construct on a piece of paper and which follows a
set of very simple rules. You can "run" it on paper as well. If there's
any interest/need I can give a complete description, which shouldn't
need to be very long. One page should do.)
-- ________________________________________________________________________Lars Marius Garshol
"Make it idiot proof and someone will make a better idiot", Bill Arnett http://www.ifi.uio.no/ülarsga/ http://birk105.studby.uio.no/
-- post message - mailto:skwokÉspark.net.hk unsubscribe/queries - mailto:dianaÉasiantravel.com homepage - http://www.geocities.com/Athens/Forum/4670
This archive was generated by hypermail 2.0b3 on Thu May 13 1999 - 16:42:05 CEST