Chess now has an engine, AlphaZero, that starts off knowing
nothing of chess. It does not come equipped with a formidable database of old
games, or a list of openings. It was never briefed by the likes of Garry
Kasparov or Magnus Carlson, and never needed their assistance.
AlphaZero, "learned to play solely by playing against
itself, over and over and over — 44 million games. It kept track of what
strategies led to a win, favoring those, and which didn’t, casting those aside.
After just four hours of this tabula rasa training, it clobbered the top chess
program, an engine called Stockfish,
winning 28 games, drawing 72 and losing zero. "
This is the method by which AlphaGo taught itself Go, and in
short order became what one competitor called, "the God of Go," capable of playing "how I imagine
games from far in the future are played."
What's fascinating about AlphaZero's chess play is that it,
"adopted an all-out attacking style, making many bold material sacrifices
to set up positional advantages." This is contrary to the impeccable but
dull style of other chess engines. Without any help from chess history, except
for its own four hours of play, AlphaZero revives the original impulse of chess
— go for mate.
Oliver Roeder,
"Chess’s New Best Player Is A Fearless, Swashbuckling Algorithm"
https://fivethirtyeight.com/features/chesss-new-best-player-is-a-fearless-swashbuckling-algorithm/
No comments:
Post a Comment