Robotics Reference
In-Depth Information
so he designed a special purpose Chess chip to replace some of the time-
critical aspects of his program. Instead of having a program module that
generates a list of all the legal moves in a Chess position, this task would
be undertaken by the silicon chips,
13
which could do it much faster.
Likewise, the evaluation function was also made in silicon for much faster
execution. By executing these tasks in hardware rather than software the
whole system was speeded up significantly.
The first version of the BELLE Chess hardware, in 1977, had incor-
porated 25 chips but produced virtually no speedup over Thompson's
earlier, purely software version of BELLE, which searched some 200 po-
sitions per second. The next version of BELLE, a year or so later, con-
tained 325 chips and searched approximately 5,000 positions per second.
The version that won the World Championship in 1980 employed 1,700
chips and searched 160,000 positions per second. BELLE's success in
Linz vindicated Thompson's conviction that speed is almost everything
in computer Chess. In 1983 BELLE's rating surpassed the 2200 mark,
earning it the title of U.S. Master.
The 1980s saw the first truly impressive results by programs against
grandmasters when playing at tournament speed.
14
Having won my bet
in 1978 I was happy to repeat the challenge for another five years, this
time making a $1,000 wager with Dan McCracken, a leading author-
ityonComputerScienceandtheauthorofmanybooksonthesubject.
In 1984 I defended this second bet in a four-game match in London
against the then World Computer Chess Champion program, Cray Blitz,
and crushed it 4-0. By then I could see that steady progress was being
made, with many more enthusiasts joining the ranks of the competing
programs at the various regular computer championships, and I there-
fore resisted the temptation to make a third bet. But rather than remove
myself as a target altogether, I joined forces with
Omni
magazine to an-
nounce a $5,000 prize for the authors of the first program to win a match
against me, whenever that might be. I put up $1,000 of the prize money
and
Omni
contributed the other $4,000. Soon a program called Deep
Thought, developed at Carnegie Mellon University and running on spe-
13
Thompson's design required several silicon chips. Later researchers were able to take advantage
of smaller silicon chip technology to cram all the necessary functionality onto a single chip.
14
At fast games, sometimes called blitz Chess, in which each player might have only five minutes
for all of their moves in the game, programs perform considerably better against strong human
players than they do at tournament speed (in which the players have an average of three minutes per
move or thereabouts). This is because, when playing at speed, humans are more likely than their
computer opponents to make careless oversights.
