PredictionBook is now read-only ( read more ).

AlphaGo wins 5-0 in march

Created by lavalamp on 2016-02-02; known on 2016-03-30; judged wrong by PseudonymousUser on 2016-03-13.

  • lavalamp estimated 85% on 2016-02-02
  • lavalamp said “Elsewhere: http://predictionbook.com/predictions/176982on 2016-02-02
  • themusicgod1 estimated 49% on 2016-02-02
  • themusicgod1 estimated 47% on 2016-02-02
  • WilliamKiely estimated 10% and said “https://www.reddit.com/r/baduk/comments/43xret/how_likely_is_alphago_to_win_50_against_lee_sedol/on 2016-02-03
  • ChristianKl estimated 10% on 2016-02-03
  • ChristianKl estimated 20% on 2016-02-03
  • ChristianKl estimated 10% on 2016-02-03
  • hipp0 estimated 20% and said “Could possibly be much higher.on 2016-02-04
  • hipp0 said “@lavalamp. It seems a bit inconsistent for you to believe that AlphaGo will win 5-0 with a 85% chance and only 88% that AlphaGo will win at least 3 games. Do you have a guess for the win rate?on 2016-02-04
  • davidmanheim estimated 40% on 2016-02-05
  • davidmanheim said “@Hippo; If you think that most of the uncertainty is about how rapidly the AI is improving, the probability that it’s near-par with any human is very low.on 2016-02-05
  • davidmanheim said “@Hippo also, see @Gwern’s comments on the other thread linked aboveon 2016-02-05
  • Raahul_Kumar estimated 80% and said “Without seeing the code, or how much training AlphaGo has undergone this is all guesswork, but I believe Google has both money and hardware to waste.on 2016-02-08
  • danielfilan estimated 40% on 2016-02-08
  • hedges estimated 35% on 2016-02-15
  • lavalamp said “@hipp0, I think the chance that alphago ends up +/- 300 elo of Lee Sedol is very low. I’ll be very surprised if alphago wins but still loses 1 or 2 games.on 2016-02-19
  • Raahul_Kumar estimated 90% and said “AlphaGo Distributed is 1202 CPUs and 176 GPUs vs 48 CPUs, and 8 GPUs of the original. The original was already good enough enough to beat a 2 Dan pro 5-0.on 2016-02-19
  • Temeraire estimated 70% on 2016-02-22
  • WilliamKiely estimated 35% on 2016-03-06
  • themusicgod1 estimated 33% on 2016-03-07
  • leo_gri estimated 52% on 2016-03-08
  • nshepperd estimated 65% on 2016-03-09
  • Raahul_Kumar said “AlphaGo beats Lee in Game 1. Th end is nigh, fleshbags.https://www.youtube.com/watch?v=vFr3K2DORc8on 2016-03-09
  • artir estimated 30% on 2016-03-09
  • WilliamKiely estimated 60% on 2016-03-09
  • Temeraire estimated 79% and said “Updating based on initial win; fewer opportunities to lose now, plus information about AlphaGo’s skill. Main concern is Carl’s low bet!on 2016-03-10
  • JoshuaZ estimated 54% on 2016-03-10
  • PseudonymousUser estimated 13% and said “Even if AlphaGo’s win probability is 80%, 0.80^4 = 0.41. It’s unlikely to win all four remaining matches.on 2016-03-10
  • PseudonymousUser said “Remaining games, I meanon 2016-03-10
  • PseudonymousUser said “If Lee Sedol is even a little bit better than AlphaGo, e.g., if he has a 0.56 probability of winning a game against AlphaGo, we’d have P(4 games won by AlphaGo) = 0.44^4 = 0.037.on 2016-03-10
  • PseudonymousUser estimated 23% on 2016-03-10
  • rhaps0dy estimated 56% on 2016-03-10
  • leo_gri estimated 96% on 2016-03-10
  • leo_gri estimated 42% on 2016-03-10
  • PseudonymousUser estimated 80% on 2016-03-11
  • Raahul_Kumar estimated 100% and said “I no longer find it plausible that any human player could beat AlphaGo.on 2016-03-11
  • nshepperd estimated 90% on 2016-03-12
  • rhaps0dy estimated 95% and said “Seeing the crushing defeats…on 2016-03-12
  • kuudes estimated 90% on 2016-03-12
  • artir estimated 98% on 2016-03-12
  • PseudonymousUser estimated 87% on 2016-03-13
  • PseudonymousUser said “Lee Sedol won Game 4on 2016-03-13
  • PseudonymousUser   judged this prediction wrong on 2016-03-13.
  • lavalamp said “Very surprised. Glad I didn’t update my original prediction after the first three wins.on 2016-03-13
  • Raahul_Kumar said “Shocked. I hope the Deepmind team will release what happened, and what the weights of the neural net were leading up to that mistake. I wonder if this is a drawback of training neural nets?on 2016-03-13
  • rhaps0dy said “Goes to show that I shouldn’t update so much and so suddenly. Especially not with altered emotions immediately after the game.on 2016-03-13