Tuesday, April 03, 2018

AlphaGo documentary

Highly recommended -- covers the matches with European Go Champion Fan Hui and 18 time World Champion Lee Sedol. It conveys the human side of the story, both of the AlphaGo team and of the Go champions who "represented the human species" in yet another (losing) struggle against machine intelligence. Some of the most effective scenes depict how human experts react to (anthropomorphize) the workings of a complex but deterministic algorithm.
Wikipedia: After his fourth-game victory, Lee was overjoyed: "I don't think I've ever felt so good after winning just one game. I remember when I said I will win all or lose just one game in the beginning. ... However, since I won after losing 3 games in a row, I am so happy. I will never exchange this win for anything in the world." ... After the last game, however, Lee was saddened: "I failed. I feel sorry that the match is over and it ended like this. I wanted it to end well." He also confessed that "As a professional Go player, I never want to play this kind of match again. I endured the match because I accepted it."
I wonder how Lee feels now knowing that much stronger programs exist than the version he lost to, 4-1. His victory in game 4 seemed to be largely due to some internal problems with (that version of) AlphaGo. I was told confidentially that the DeepMind researchers had found huge problems with AlphaGo after the Lee Sedol match -- whole lines of play on which it performed poorly. This was partially responsible for the long delay before (an improved version of) AlphaGo reappeared to defeat Ke Jie 3-0, and post a 60-0 record against Go professionals.
Wikipedia: ... Ke Jie stated that "After humanity spent thousands of years improving our tactics, computers tell us that humans are completely wrong... I would go as far as to say not a single human has touched the edge of the truth of Go."

In this video interview, Ke Jie says "I think human beings may only beat AlphaGo if we undergo a gene mutation to greatly enlarge our brain capacities..."  ;-)
Last year I was on an AI panel with Gary Kasparov, who was defeated by DeepBlue in 1997. (Most people forget that Kasparov won the first match in 1996, 4-2.) Like Lee, Kasparov can still become emotional when talking about his own experience as the champion representing humanity.

It took another 20 years for human Go play to be surpassed by machines. But the pace of progress is accelerating now...
Wikipedia: In a paper released on arXiv on 5 December 2017, DeepMind claimed that it generalized AlphaGo Zero's approach into a single AlphaZero algorithm, which achieved within 24 hours a superhuman level of play in the games of chess, shogi, and Go by defeating world-champion programs, Stockfish, Elmo, and 3-day version of AlphaGo Zero in each case.
Some time ago DeepMind talked about releasing internals of AlphaGo to help experts explore how it "chunks" the game. Did this ever happen? Might give real insight to scholars of the game who want to "touch the edge of truth of Go" :-)

No comments:

Blog Archive