For the first time, a computer has beaten the human champion at the strategic game of Go. Beating humans at Go has long been viewed as a virtually impossible task for artificial intelligence because the game relies on ostensibly human characteristics like intuition and imagination.

The computer, whose name is AlphaGo, was developed by Google's AI branch, a company called DeepMind. DeepMind representatives say that after years of development, but still ahead of schedule, their computer has mastered the game. In fact, the machine beat Fan Hui, the biggest Go champion in all of Europe, with a score of 5:0. It was also matched against other Go-playing computers, and won 99.8 percent of games against its competitors.

AlphaGo uses two algorithms to choose its moves: a "Policy" network and a "Value" network. The Policy network considers all possible moves, and decides which are the most promising. Then the "Value" network plays out various possibilities should any of those moves be made.

While other AI programs attempt to "play" the games in their machine minds all the way to the end, AlphaGo knows when to quit. It analyzes a move only about 20 paces ahead, which has actually turned out to be advantageous in the game of Go. For once, overanalyzing doesn't pay off. In fact, the referee of the match called the computer's gaming style conservative.

The computer is also scheduled for a match against Lee Sedol, a Go champion from South Korea, in March. DeepMind co-founder Demis Hassabis expressed cautious confidence that his creation would win that match, too. Hassabis explains in a video produced by Nature (see below) that he got into Go research when he discovered the game at university and his friends attempted to best him at it.

Go is often compared with chess (which also made news today) because both involve intense strategy and highly competitive gameplay. But according to Nature, Go is far more complex. "The average 150-move game," says Nature writer Elizabeth Gibney, "contains more possible board configurations than there are atoms in the universe."

Source: Nature

ⓒ 2024 TECHTIMES.com All rights reserved. Do not reproduce without permission.
Join the Discussion