DeepMind Publishes Research Surrounding AlphaZero, the Game-Playing AI: It's Becoming Smarter

Monday, 10 December 2018 - 11:59AM
Technology
Artificial Intelligence
Monday, 10 December 2018 - 11:59AM
DeepMind Publishes Research Surrounding AlphaZero, the Game-Playing AI: It's Becoming Smarter
< >
Pixabay Composite
Back in 2016, DeepMind's AlphaGo won the "Match of the Century" against Go Grandmaster Lee Sedol, conquering one of the last bastions of human superiority (at least when it came to complex board games). It's successor, AlphaZero, has since mastered chess, Go, and Shogi, and even beat machine opponents custom-built to play those games. Now, DeepMind has finally published the long-awaited paper detailing how they trained their champion AI—the "one program to rule them all," according to the study.

Putting aside the worrying comparisons to Sauron, AlphaZero's achievements are even more impressive considering how long it took it blow past world champions after learning the games from scratch. According to the pre-print research published on arxiv last December: "Starting from random play, and given no domain knowledge except the game rules, AlphaZero achieved within 24 hours a superhuman level of play in the games of chess and shogi (Japanese chess) as well as Go, and convincingly defeated a world-champion program in each case."

AlphaZero is an artificial intelligence that relies on reinforcement learning, and while it's still not as efficient as human Grandmasters, it's leagues ahead of chess-playing programs like Stockfish. In fact, British chess Grandmaster Matthew Sadler has called its style "beautiful," and even says it displays 'intuition' similar to a human player:

"Traditional engines are exceptionally strong and make few obvious mistakes, but can drift when faced with positions with no concrete and calculable solution … Impressively, [AlphaZero] manages to impose its style of play across a very wide range of positions and openings. It's precisely in such positions where 'feeling', 'insight' or 'intuition' is required that AlphaZero comes into its own. AlphaZero plays like a human on fire. It's a very beautiful style."

AlphaZero's creators don't intend to stop at games when it comes to applying their unique brand of AI learning. According to to Demis Hassabis, the CEO and co-founder of DeepMind: "Alpha Zero is a stepping stone for us all the way to general AI...Ultimately, [we're developing algorithms that can be] translate[ed] into the real world to work on really challenging problems...and help experts in those areas."

Science
Science News
Technology
Artificial Intelligence
No