130 lines
6.3 KiB
Plaintext
130 lines
6.3 KiB
Plaintext
|
|
DeepThought : The computer chess Champion
|
|
By : Anonymous Rebel
|
|
|
|
|
|
|
|
|
|
The artificial intelligence pioneer, Herbert Simon, predict-
|
|
ed that by 1967 a computer would be able to beat any human in
|
|
chess. However, he was quite far off. In 1989, the human mind
|
|
can still defeat even the greatest of computer chess machines.
|
|
Simon wasn't as far off as the critics of the Fifties, who
|
|
claimed that a computer could never master chess. A computer
|
|
called Deep Thought now plays at the level of a human grandmas-
|
|
ter, which means that only about 200 people in the world can
|
|
defeat him.
|
|
|
|
Deep Thought has not exactly been ranked as a grandmaster as
|
|
of yet, because according to Thomas Anantharaman, one of the
|
|
creators of Deep Thought, "You can't claim to be a grandmaster
|
|
unless you are a human being." However, this hasn't upset his
|
|
fellow creators or him. So far, they have won $10,000 from The
|
|
Fredkin Fund, for winning 25 straight games in world competition.
|
|
($100,000 is waiting for the first electronic world champion.)
|
|
|
|
Deep Thought follows the tradition of its computer predeces-
|
|
sors, by considering every possible move. Good human players
|
|
usually examine only a few moves, and rely on pattern recognition
|
|
and experience to succeed. This type of evaluation proves ex-
|
|
tremely difficult to program into a computer, although the former
|
|
computer chess champion, Hitech, was designed to recognized
|
|
certain programs. However, ruling out certain moves, without
|
|
fully examining all the possibilities will eventually lead to an
|
|
error in the computer's part.
|
|
|
|
Although a computer lacks the ability of insight, they make
|
|
it up in their ability to number crunch. For example, in a
|
|
typical situation, a chess player has 38 moves to consider. To
|
|
check all the responses would mean examining 1444 different
|
|
positions. To look one step ahead would mean checking over 2
|
|
million positions.
|
|
|
|
The secret of the computer's success, obviously, is its
|
|
incredible speed. Even a little toy chess computer is fast
|
|
enough to beat most people. The big time computers like Deep
|
|
Thought is even faster. In fact, the ultrafast electronics are
|
|
the source of power for all Fredkin prize hopefuls. Deep Thought
|
|
is simply the fastest around.
|
|
|
|
Deep thought is designed to examine five moves ahead. Since
|
|
this is far from seeing the end of a game, the machine must not
|
|
only check what positions it could end up in, but must evaluate
|
|
them as well. Typical considerations include the value of what
|
|
pieces will be captured, and whether the King will be put in
|
|
check. Deep Thought also checks its position on the board, and
|
|
the number of empty squares from which the King could be at-
|
|
tacked, and some 80 other items. Deep Thought gives numerical
|
|
values to all of the considerations, and uses the highest ranked
|
|
move.
|
|
|
|
All of the world-class chess computers uses certain varia-
|
|
tions on this approach. The reason for Deep Thought's success
|
|
is that it doesn't always hold itself to a few moves. If one
|
|
move seems particularly good, it searches down that path for up
|
|
to 15 moves. This strategy is called singular extension. If
|
|
Deep Thought registers an equally dangerous threat, it also
|
|
searches further, it works in the same manner, but this time its
|
|
called Threat Extension. If this feature comes into play, the
|
|
results are usually deadly for the opponent.
|
|
|
|
Deep Thought didn't used an experienced chess player to
|
|
assign relative values to its 80 considerations, like other chess
|
|
computers. Hitech had used Hans Berliner, a computer scientist
|
|
and a former world-class champion. "The fascination," Ananthara-
|
|
man says, "was writing a program that can do something I can't
|
|
do." Instead, the computer watched games played between human
|
|
grandmasters. One of the programmers "tuned" Deep Thought's
|
|
evaluation equation until it started choosing Grandmaster style
|
|
moves. First approximations were made in deciding the order.
|
|
Deep thought would use this hierarchy and replay the grandmaster
|
|
games. At first, Deep Thought played horribly. But after each
|
|
game, the program would change the rank slightly until Deep
|
|
Thought was playing like a true grandmaster. After 800 games,
|
|
Deep Thought was playing like a true master.
|
|
|
|
Anantharaman was truly surprised that the "tuning" worked as
|
|
well as it did. After evaluating the process, he decided that
|
|
this process is a better technique even if you have the expert
|
|
knowledge. The approach could also be used in other fields where
|
|
human experts aren't available. The trouble is that not many
|
|
problems are as well defined as that of chess. Some day the
|
|
thinking process would be useful to the military, but that is far
|
|
off. "Computer chess will give you an idea how to approach a
|
|
problem," says Anantharaman, "but it won't solve the problem
|
|
completely. I can't think of a single real-world problem that
|
|
has been solved by artificial intelligence."
|
|
|
|
|
|
|
|
|
|
Bibliography
|
|
|
|
|
|
|
|
"Pawn to King Four." Tom Waters. Discover Magazine,
|
|
May 1989. Volume 10, Number 5
|
|
|
|
|
|
|
|
X-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-X
|
|
|
|
Another file downloaded from: NIRVANAnet(tm)
|
|
|
|
& the Temple of the Screaming Electron Jeff Hunter 510-935-5845
|
|
Rat Head Ratsnatcher 510-524-3649
|
|
Burn This Flag Zardoz 408-363-9766
|
|
realitycheck Poindexter Fortran 415-567-7043
|
|
Lies Unlimited Mick Freen 415-583-4102
|
|
|
|
Specializing in conversations, obscure information, high explosives,
|
|
arcane knowledge, political extremism, diversive sexuality,
|
|
insane speculation, and wild rumours. ALL-TEXT BBS SYSTEMS.
|
|
|
|
Full access for first-time callers. We don't want to know who you are,
|
|
where you live, or what your phone number is. We are not Big Brother.
|
|
|
|
"Raw Data for Raw Nerves"
|
|
|
|
X-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-X
|