Google DeepMind’s artificial intelligence (AI) program AlphaZero was able to master the game of chess from scratch and evolve “superhuman” abilities within four hours, research suggests.
AlphaZero was tasked with learning chess and Japanese board game shogi before going head-to-head with rival world-leading AI programmes, known as Stockfish and Elmo. Its predecessor – dubbed AlphaGo Zero – was recently retired after trashing human opponents.
Details of the most recent tests were published on Wednesday (6 December) by New York’s Cornell University, but findings are still to be peer-reviewed.
Google will not comment until then.
According to DeepMind, which is partnered with the UK’s NHS to research AI healthcare, AlphaZero destroyed rival AI “Stockfish” in a matter of hours.
In a total of 100 games, with one minute of “thinking time”, it won 28 matches and drew the remaining 72.
The paper said the program was given no prior knowledge of the game’s rules and within 24 hours had achieved “a superhuman level of play” in chess, shogi and Go.
The experts claimed it “convincingly defeated a world-champion programme in each case”.
Going up against AI system Elmo at shogi, AlphaZero won 90 games, drew two and lost eight. Researchers also found that the learning algorithm was able to beat a prior version of itself.
English chess grandmaster Simon Williams tweeted that the news was “one for the history books”.
He wrote: “AlphaZero took over the chess world. AlphaZero and DeepMind then went on to dominate chess, eventually solving the game and finally enslaving the human race as pets.”
In May this year, the previous iteration of AlphaZero successfully beat China’s 19-year-old Ke Jie, described as a prodigy player of the board game Go. It won by a half-point.
The race towards machine learning and AI was first boosted by the 1997 win by the program Deep Blue, which defeated then-world chess champion Garry Kasparov.
AI is now beyond human level and according to some experts is only just getting started.
“We now know who our new overlord is,” David Kramaley, who manages chess education website Chessable, said.
“The games AlphaZero played show it can calculate some incredibly creative positional bombs, the depth of which are far beyond anything humans or chess computers have come up with.
“It will no doubt revolutionise the game, but think about how this could be applied outside chess. This algorithm could run cities, continents, universes.”