Go Grandmaster Loses Tournament Against Google’s AlphaGo

Getty Images
Getty Images / Getty Images
facebooktwitterreddit

Google’s AlphaGo is on a roll. In January, the artificially intelligent Go-playing robot became the first computer to beat a professional player at the Chinese board game Go. Now, WIRED reports that AlphaGo has made its most decisive victory, beating Go Grandmaster Lee Sedol four to one in a five-game Go tournament.

While computers have vanquished professional chess players, bested humans at Jeopardy, and even solved Connect Four, this is the first time a computer has competed professionally in and won a Go tournament against a player at Sedol’s level. Go, which is played on a 19-by-19 grid, has more possible opening moves than perhaps any other board game. For instance, while Connect Four has seven possible opening moves, and chess has 20 possible openings, Go has 361. The sheer number of possible moves in a Go game has historically made developing a Go-playing computer hugely challenging.

AlphaGo, however, is a triumph in AI technology. The computer system used a neural network to not only learn Go moves from professional players, but to develop its own strategies and skills by playing against itself, CNET reports.

AlphaGo, the "player" at left, was aided by a human assistant who moved the pieces on the Go board. At right, Grandmaster Lee Sedol. Image credit: Getty

The tournament put the differences between humans and computers into sharp relief. On the one hand, AlphaGo played an idiosyncratic game, according to WIRED, making moves throughout the tournament that no human would choose. (As you can see in the photo above, AlphaGo got a physical assist from a human helper who moved the pieces on the Go board.) It also made mistakes that seemed amateurish to human observers. Sedol, on the other hand, made at least one uniquely human choice: After beating AlphaGo while playing with white pieces in the fourth Go match—his only victory—he chose black pieces for the fifth and final game, a decision he knew would put him at a disadvantage.

Throughout his first four games, Sedol had noticed AlphaGo struggling more when it played with black pieces, and so, rather than play it safe in the fifth game, he decided to see if he could beat the computer at its strongest. “I really do hope I can win with black,” he told WIRED before the final game, “because winning with black is much more valuable.”

Though Sedol ultimately lost, the game against AlphaGo was nevertheless a win for human innovation.

"The game showed that AlphaGo is far from infallible," explains WIRED. "There are holes in its education. But, able to draw on months of play with itself—on a corpus of moves that no human has even seen—it also has the ability to climb out of such a deep hole, even against one of the world’s best players."

[h/t WIRED]