next wednesday march 9th is the 1st match between an improved google Go and the top human player Lee Sedol. it will be livestreamed on youtube with commentary by several experts.[a1]
as has been noted in a lot of articles on the subject, the recent match between european champion Hui tested a human at ELO ranking 2750. Sedol is ranked at 2940. Hui is ranked 633rd in the world, Sedol ranked 5th. Sedol [and other cohort experts] is already on record as saying he is confident he will win at least this 1st match based on analysis of the prior Hui match and apparent mistakes in the computer gameplay. the prior Hui match would seem to definitively show the Go machine is already playing substantially better than ELO 2750.
but Sedol maybe seems not to have realized/ anticipated/ acknowledged that Google is already throwing significantly more hardware and software improvements (obtained via reinforcement learning) in the new match, therefore the new system will not be the same as the prior one; it will presumably be substantially improved and maybe even its style of play will diverge some.
also one must note that “style of play” is in fact a moving target. one might even expect it adapts its style of play to the opponent and that maybe approaches it used against Hui might not match up with those against Sedol. modern AI can be regarded as supremely adaptive…
and one would expect the Google team to make some careful calculations and would not arrange the public/ publicized match anticipating to be beaten in all the games, because this would be a serious PR blow to Google. could Google possibly have some way to estimate its ELO performance of the game? ELO rating is nearly a scientific measurement of winning probability in different contexts. and even the 5-game AlphaGo-over-Hui win might have some implication for an estimated ELO score greater than Hui’s 2750 (any ELO/ math experts out there? what would that calculation look like?)
the prize is $1M and will be donated to Go charities if Google wins. its also interesting to ponder this as a bet. is Google essentially making a $1M bet they can beat Sedol, or at least win a few games? what was the behind-the-scenes analysis/ thinking that went into this “wager”?
also, its a winner-take-all prize! a little minicommentary of my own: this has strong tones of the historic 1997 Kasparov vs Deep Blue match which got a lot of internet attn/ coverage. in 1997, as is not widely pointed out, the $1.1M prize Kasparov vs Deep Blue match had $700K to the winner and $400K to the loser. essentially Kasparov was paid $400K just to play and lose! he was exceedingly bitter about that loss as reported by the media, which did not note at the time much it seems that he still walked away with $400K. it takes a world class grandmaster to express such bitterness about $400K, eh? there are some really great resources on this history, ie a lively/ engaging movie documentary “Game Over” (saw it recently, luved it, highly recommend it!), two books, one by the chief architect of Deep Blue, and another by a top chess commentator, and an excellent 17m ESPN documentary with great/ cutting commentary/ analysis.[b]
my thinking is wrt this, Sedol might be a bit overconfident in a match where he cannot really have a good idea of the power of the machine he is playing (he apparently only has a “small dataset” of only 5 prior games to go by, and the new system is presumably going to play higher than that level), and maybe Sedol, if given a choice or during the presumable negotiation opportunity, should have chosen a similar split prize arrangement to be safer (if he actually cares about the prize money, as most humans would be apt to do)! but Sedol is quite confident (possibly also having heard of the history of the Kasparov match), and yet here it reminds me of the Kasparov overconfidence expressed pre-match.
this is a groundbreaking story of how a determined effort by a large entity in this case Google can crack a CS problem quite readily with good resources, namely a topnotch 15 man team experts in AI, latest AI techniques, and lots of excellent hardware. how many other deep CS problems are waiting to be overthrown in such a fashion? this is maybe the 1st in a long line of victories by AI (actually “moving target wise” its just the latest in a very long string, but whos counting?). to me just the match itself is a huge milestone and even genuine breakthru for the field.
so while deservingly heralded as a genuine breakthru, its a bit unnerving for AI researchers and maybe singulatarians to realize that it took ~2 decades to get here from chess and that in some ways this is “only” about the 2nd major/ memorable/ milestone event of the 21st century (the historic IBM Watson jeopardy match being the 1st). AI is unequivocally advancing, but it doesnt really seem to be approaching any “explosion” or even “merely” a Moores-law like exponential trend—so far that is. and by the way even Moores law itself is now widely regarded by experts to have plateaued at least “for the moment”. another related point, we do not even have any strong evidence that a strong uninterrupted continuation of Moores law would inevitably be utilizable for AI improvements. intuitively it seems plausible of course, but science of course is more than educated guesses!
anyway, cant wait! am really on the edge of my seat!
- 1. AlphaGo v Lee Sedol match schedule announced | British Go Association
- 2. DeepMind – YouTube
- 3. Google DeepMind: Ground-breaking AlphaGo masters the game of Go – YouTube
- 4. Top professional Lee Sedol prepares to play computer in Go showdown | News | Lifestyle | The Independent
- 5. Go players react to computer defeat : Nature News & Comment
- 6. YouTube will livestream Google’s AI playing Go superstar Lee Sedol in March | VentureBeat | Big Data | by Jordan Novet
- 7. Facebook trains AI to beat humans at Go board game – BBC News
- 8. Odds favor machine over human in big Go showdown
- 9. Don’t Pass Go: AI Will Beat You at Pretty Much Everything – The Daily Beast
- 10. Google’s Go Victory Is Just a Glimpse of How Powerful AI Will Be | WIRED
- 11. ‘Huge leap forward’: Computer that mimics human brain beats professional at game of Go | Science | AAAS
- 12. Google’s AI is the first to defeat a Go champion
- 13. Official Google Blog: AlphaGo: using machine learning to master the ancient game of Go
- 14. Research Blog: AlphaGo: Mastering the ancient game of Go with Machine Learning
- 15. Artificial Intelligence: Stop or Go?
- 16. Google DeepMind will fight Go world champ live on YouTube (Wired UK)
- 17. Google’s AI AlphaGo to take on world No 1 Lee Se-dol in live broadcast | Technology | The Guardian
- 18. BitBet – AlphaGo will defeat Lee Sedol overall in March 2016 match :: 5.65 B (51%) on Yes, 5.51 B (49%) on No | closing in 3 weeks 3 days | weight: 93`611 (100`000 to 1)
- 19. A Go Grandmaster Will Battle Google’s AI in a $1M Prizefight | WIRED
b. Deep Blue vs Kasparov 1997
- Behind Deep Blue: building the computer that defeated the worlds chess champion/ Feng-Hsiung Hsu
- Kasparov vs Deeper Blue: the Ultimate Man-Machine Challenge by Daniel King
- Game Over: Kasparov and the Machine
- Signals: the man vs the machine