AlphaGo will win the 5th match against Lee Sedol

Created by artir on 2016-03-13; known on 2016-03-16; judged right by PseudonymousUser on 2016-03-15.

  • artir estimated 58% on 2016-03-13
  • EloiseRosen estimated 75% on 2016-03-13
  • regex estimated 45% and said “I think AlphaGo’s end game is too weak.on 2016-03-13
  • JoshuaZ estimated 75% on 2016-03-13
  • Tapetum-Lucidum estimated 85% on 2016-03-14
  • themusicgod1 estimated 51% and said “this claim covers the exact same ground as some other claims. 2016-03-14
  • Raahul_Kumar estimated 100% and said “I think your existence is too weak.on 2016-03-14
  • JoshuaZ said “Since AlphaGo is not being updated during the match(I thought?), can someone explain how they can generally give an estimate much above the base rate of 75%? I can understand lower numbers since Lee Sedol may have learned, but higher puzzles me. on 2016-03-14
  • PlacidPlatypus estimated 75% and said “I’m not really sure but my impression is that AlphaGo can learn even without being updated by its programmers. Still wouldn’t go that high though.on 2016-03-14
  • artir estimated 60% on 2016-03-14
  • lavalamp estimated 60% on 2016-03-14
  • Raahul_Kumar said “AlphaGo uses Q learning. will converge on optimal play, and given that AlphaGo is already ranked 4 and is only 90 points away from being the strongest at Go ever. 2016-03-15
  • Tapetum-Lucidum said “@JoshuaZ I saw that the value network was (usually) far better assessing board position than Lee Sedol was…on 2016-03-15
  • Tapetum-Lucidum said “Its one major mistake occurred because Lee Sedol was able to force it into an unfamiliar situation. Given that it has played through most professional games in existence…on 2016-03-15
  • Tapetum-Lucidum said “I believe Lee Sedol won’t FIND any more holes in its knowledge, even if those holes still exist. That is, I think Lee Sedol got lucky—even if his move 78 on Game 4 was very good.on 2016-03-15
  • Tapetum-Lucidum said “I.e., I believe that the holes in AlphaGo’s knowledge come from situations that humans (and itself, in self-play) have rarely played through. on 2016-03-15
  • Tapetum-Lucidum said “Given that these ‘holes’ persist for AlphaGo after going through the games of the most talented players in history—not to mention playing against itself—I believe the probability of Lee Sedol finding another hole is < 1/4.on 2016-03-15
  • Tapetum-Lucidum said “We’ll see though :pon 2016-03-15
  • Houshalter estimated 55% on 2016-03-15
  • nshepperd estimated 80% on 2016-03-15
  • PseudonymousUser   judged this prediction right on 2016-03-15.

Please log in to respond to or judge prediction