The Institute - September 2021 - 65

OTHER GAME-PLAYING AI
IBM Watson, developed in 2011
The machine won a game of
Jeopardy! against two of the U.S.
game show's most successful
contestants in 2011.
Q-learning, developed in 2013
This supercomputer used a
convolutional neural network, which
analyzed visual imagery and enabled
Q-learning to master several games
such as Pong, Space Invaders, and
Seaquest on the Atari 2600 console.
The computer garnered higher
scores than human players.
World champion Garry Kasparov [left] playing against IBM supercomputer Deep Blue
in 1996 during the ACM Chess Challenge in Philadelphia
route to take based on the information
gathered by the chips. Deep Blue could
explore up to 100 million possible
chess positions per second, according
to the IBM article.
" Hundreds of millions of people
around the world play chess, " Campbell
said in a 2017 Scientific American
interview. " It's known as a game that
requires strategy, foresight, logic-all
sorts of qualities that make up human
intelligence. So it makes sense to use
chess as a measuring stick for the
development of artificial intelligence. "
The team knew chess was the
right game for Deep Blue to play, but
the researchers had little experience
with chess themselves. The team
brought in grandmasters such as Joel
Benjamin, who, at 13, had become the
youngest-ever U.S. chess master.
The grandmasters helped the team
in two ways: assisting in putting
together a library of moves for the
machine to access during games and
playing against the machine so the
team could pinpoint its weaknesses.
" Humans have been studying
chess openings for centuries and
developed their own favorite moves, "
Campbell told Scientific American.
" The grandmasters helped us choose
a bunch of those to program into
Deep Blue.
" Chess is an enormously complex
game, and that's why it took us, as
a field, 50 years of development to
finally beat the world champion. "
" Chess is an enormously complex
game, and that's why it took us, as
a field, 50 years of development to
finally beat the world champion. "
SEPTEMBER 2021
THE INSTITUTE
65
Kasparov vs. Deep Blue
After the machine lost its first match
in 1996 against Kasparov, the research
team went back to the drawing board.
According to Campbell, the team
doubled the system's speed by developing
a new chess chip-one with the
enhanced ability to evaluate positions
the pawns can take. The new version of
Deep Blue was able to search up to 200
million options per second, depending
on the pawns' position on the board.
The researchers also increased the
machine's knowledge of the game by
enabling the chess chip to recognize
and evaluate chess concepts including
positions and lines of attack. The chips
could then search through the possibilities
and figure out the best move.
" Part of the improvement is we
detected more patterns in a chess
position and could put values on
them and therefore evaluate chess
positions more accurately, " Campbell
said in the interview.
Deep Blue and Kasparov squared
off again in 1997 in a six-game match.
The grandmaster won the first game;
AlphaGo, developed in 2015
The supercomputer beat the world
champion-Lee Sedol-of Go, a
Japanese board game, in 2016.
AlphaStar, developed in 2019
It was the first AI program to defeat
professional players of StarCraft II,
a real-time strategy game created
by Blizzard Entertainment, in 2019.
SOURCE: BERNARD MARR AND CO.
the machine won the next one. The
following three ended in a draw, and
Deep Blue won the final game and
thus the match.
Campbell said he and his team were
" confident that the 1997 Deep Blue was
much better than the 1996 version, "
but they still hadn't expected it to win.
According to IBM, the development
of Deep Blue inspired researchers
to create supercomputers that
could tackle other complex problems
such as evaluating marketplace
trends and risk analysis in finance;
mining data; and analyzing molecular
dynamics-which helped medical
researchers develop new drugs.
TOM MIHALEK/AFP/GETTY IMAGES

https://www.ibm.com/watson https://www.arxiv.org/pdf/1312.5602v1.pdf https://www.deepmind.com/research/alphago/ https://www.deepmind.com/blog/article/alphastar-mastering-real-time-strategy-game-starcraft-ii https://www.scientificamerican.com/article/20-years-after-deep-blue-how-ai-has-advanced-since-conquering-chess/ https://www.bernardmarr.com/default.asp?contentID=1923 https://en.wikipedia.org/wiki/Joel_Benjamin https://en.wikipedia.org/wiki/Joel_Benjamin

The Institute - September 2021

Table of Contents for the Digital Edition of The Institute - September 2021