Rating boundaries
Kevin Stokker
What would be the ratings of the following persons A, B, and C, where:
A is perfection. This player is capable of always making the best move in
all circumstances.
B is a random move generator. He is essentially the 'worst' player one
could imagine, because he has absolutely no sense of how to win. He may or may
not checkmate if this option is available. He is not Worst, though, because
he is not actually trying to lose. He wants to win, but doesn't know anything
more than just the rules of chess. As a matter of fact, he is quite dumb
indeed, because although he knows of 'capturing', he does not realize that this
will help him attain his goal, and although he is aware of promotion, he
similarly does not realize that this will aid his cause. He can see zero
moves ahead.
C is The Worst player. That is, he consciously, consistenly, and in the
smartest way possible attempts to helpmate himself. Note that this person
might almost be as intelligent and knowledgeable as person A (maybe even
'smarter' -- he can calculate statistical probabilities into the indefinate
future and would stand a much better chance of helpmating himself against,
say, person B, than would a master, or even person A).
I would put A (perfect chess) at 3100.
Perhaps B is 0?
What would C be? Perhaps -3100.
Perhaps person B should checkmate if it is immediately available? Just
because he can see zero moves ahead doesn't mean he will refuse a checkmate.
This might make more sense. I will still give him a rating of zero, although
perhaps this 'raises' C's rating, i.e., demands that he not be so 'smart'.
Person A and C probably need to be further defined in that they can perform
these operations within any given time control. Given this, that they can
come up with their respective 'perfect' moves, in say 0.1 seconds, this
would substantially raise and 'lower' their ratings. Hmmm.
Getting a little goofy here,
kevin
J Michael Hammond
(Kevin Stokker) writes:
...
>B is a random move generator. He is essentially the 'worst' player one
>could imagine, because he has absolutely no sense of how to win. He may or may
>not checkmate if this option is available. He is not Worst, though, because
>he is not actually trying to lose. He wants to win, but doesn't know anything
>more than just the rules of chess. As a matter of fact, he is quite dumb
>indeed, because although he knows of 'capturing', he does not realize that this
>will help him attain his goal, and although he is aware of promotion, he
>similarly does not realize that this will aid his cause. He can see zero
>moves ahead.
>Perhaps B is 0?
To rate A (God) requires evaluating Fischer's claim that he could hold
God to a draw given white (I forget the precise quote but it involved
playing e4 and the Ruy Lopez if allowed) and to flame about how well
current top grandmasters play. I think the Go players have enough
basis to evaluate their top players these days; it seems that the
chess world doesn't yet have the same certainty.
I guess C (the perfect helpmater) is going to draw some games, to his
great displeasure!, against B and worse players. (For B to draw
against the perfect helpmater, he has to accidentally miss a mate for
long enough to have the opportunity to claim by threefold or
fifty-move and then must randomly select the draw claim from his
options; worse players than B might take the draw *because* it's
suboptimal.) The probability of drawing against B is so small that
C's rating is clearly many, many hundred points (perhaps several
thousand points) lower.
B is an interesting case. I claim that we can approach his rating by
looking at the play in elementary-school tournaments where some real
live players approach B's style! :-)
The lowest rating I recall ever seeing was in the five hundreds. If
we posit that this youngster knew that capturing and checkmating were
good ideas, but that he saw only the current move, then (handwave
handwave) I claim that there is about enough room for a statement of
the form "this kid could 90% of the time beat someone who could 90% of
the time beat B" so we can place B's rating at -300 plus or minus
several hundred points.
You know, it wouldn't be too tough, and it'd probably be Really
Amazingly Humorous, to cook up some extremely weak programs that had
the styles
(1) Capture the most valuable piece available; failing that, make a
random move (I claim this player would be within shouting range of the
aforementioned 500-rated youngster; stronger because it never misses a
capture, but weaker because it doesn't even have a rudimentary concept
of openings, control of the center, or "go get that guy with the cross
on top")
(2) If captures are available, take one N% of the time; failing that,
make a random move
(3) Make a random move
and measure their performance. It'd be painful as all hell to watch
them play but in sort of a cathartic way, don't you think?
------------------------------------------------------------------
J. Michael Hammond The MathWorks, Inc. | Testers do it
Manager of Software Quality (voice) 508 653-1415 | over and over!
jm...@mathworks.com (FAX) 508 653-2997 |
------------------------------------------------------------------
J Michael Hammond
>You know, it wouldn't be too tough, and it'd probably be Really
>and measure their performance. It'd be painful as all hell to watch
>them play but in sort of a cathartic way, don't you think?
I forgot to mention that there exists (or existed) an honest-to-
goodness research laboratory somewhere in the United States called the
Artificial Stupidity Laboratory. I read an article about them in some
popular publication; one of their projects was a program called the
"Artificial Paranoid" and they did some other funny but possible
useful stuff. Does anyone know where the Artificial Stupidity Lab is,
and perhaps the name of a contact person? A stupid chess program is
just prosaic enough to possibly have been done by them already.
--JMike again
Jason D Corley
Artificial Stupidity is, believe it or not, a rather advanced form
of Artificial Intelligence. One of the most famous test for AIs is
the Turing test, in which you put an AI on one teletype and a real
person on the other, and an ordinary man gets to "talk" to both
of them. If he can't tell which one is the machine, the AI is said
to have reached true intelligence. But realise that this is extremely
easy to foul up. For instance, if the man says "Divide 82393.4389 by
3874833.39384", and one machine instantly gives back the right answer
and the other machine waits 20 minutes and gives a wrong answer, it's
pretty obvious which is the machine and which is the person.
Enter Artificial Stupidity. The machine makes mistakes. It flubs up.
It takes a long time on some problems and a really short time on
others. If it plays chess, it occaisionally screws up a knight sac or
turns a draw into an instantly resignable position. It may even
refuse to play at all "I'm sick of chess. I don't want to play chess
anymore. Lets talk about something else."
So even though it has a dumb name, Artificial Stupidity is working to
make computers more likeable, understandable, and when it comes right
down to it, more human.
I bet they do have an artificially stupid chess program, but it
hardly makes any sense outside the context of an entire AI based
on stupidity.
--
"Seriousness of mind was a prerequisite for understanding Newtonian physics.
I am not convinced that it is not a handicap in understanding quantum theory."
------Connie Willis
Jason "cor...@gas.uug.arizona.edu" Corley Southern Arizona Gizmonic Institute