Forum for AlphaGo

Estimated strength after 60-0 [#3366]

Back to forum     Back to page

New reply

 
reply
188.6.73.221: Estimated strength after 60-0 (2017-01-13 03:22) [#10874]

This result seems to imply way more than the 2 stones handi mentioned in the page. Of course these were fast games, and presumably not all opponents were at Ke Jie level. Still it's interesting to speculate, especially considering the previous belief that top pros are already only 3-4 stones away from perfect play (which AlphaGo is probably still not near).

X
174.104.194.232: Re: Estimated strength after 60-0 (2017-01-14 07:35) [#10875]

It would be absurd if pros weren't within "3-4 stones" of best play. While limited information suggests pros respect AG despite the short time controls, I will until hearing otherwise enthusiastically mock the meaning of it giving its prior self 4 handicap stones, being monte carlo loony.

188.6.79.201: Re: Estimated strength after 60-0 (2017-01-14 13:42) [#10876]

Well each statement seems sound and probable, but all cannot be true at the same time:

1. pros within 1-2 of alphago
2. pros cannot win a single game in 60
(3. pros within 3-4 of perfection)
(4. alphago within 1-2 of perfection)

One possible solution is that the mathematical model for handicaps is completely different around 9d than for lower ranks (>98% winrate for lower differences). Another option is that Alphago is maybe near perfection afterall - in this case #1 above is still false. Or a third one: random variance, lucky streak. :)

174.104.194.232: Re: Estimated strength after 60-0 (2017-01-14 20:37) [#10877]

I'm alluding to AG's use of handicap stones being poor.

We do know that at perfect play, winrate goes to 100% for any difference in rank.

In Go terms, pros care about fractions of a point, and a handicap move is worth more than 10 points.

188.6.79.201: Re: Estimated strength after 60-0 (2017-01-15 02:47) [#10878]

Winrate is never 100%: even flawless play needs an error from the opponent to win (with proper komi). And even a 1-2 stones weaker opponent will avoid errors in a few, rare games by pure chance.

 
Back to forum     Back to page

New reply


[Welcome to Sensei's Library!]
RecentChanges
StartingPoints
About
RandomPage
Search position
Page history
Latest page diff
Partner sites:
Go Teaching Ladder
Goproblems.com
Login / Prefs
Tools
Sensei's Library