| Life In 19x19 http://www.lifein19x19.com/ |
|
| Alphago and komi http://www.lifein19x19.com/viewtopic.php?f=13&t=12820 |
Page 1 of 1 |
| Author: | Aeneas [ Sat Mar 12, 2016 5:00 am ] |
| Post subject: | Alphago and komi |
Do you think Alphago can be used in future to determine more precisely what the correct komi should be? I am thinking that it would be interesting to look at the thousands and thousands of games Alphago plays against itself and compare the winning percentages of black and white with different komis. |
|
| Author: | Galation [ Sat Mar 12, 2016 7:15 am ] |
| Post subject: | Re: Alphago and komi |
A "correct" determination of the komi is, for sure, one of the infinite possibilities that a Go skilled AI will provide to the Go community in the future. Who knows where will this led in the end?  I mean, komi would have changed the result of some famous games of the past, like the one in Kawabata's "The Master of Go". Galation |
|
| Author: | Pio2001 [ Sat Mar 12, 2016 7:37 am ] |
| Post subject: | Re: Alphago and komi |
I don't know if the final score is a linear function of the strength of the players. Think about moves that have the ability to upset the flow of a given game. In this case, a komi given by AlphaGo may be fair only for games played by AlphaGo, while at the same time being unfair for games played by weaker players. I'd prefer a komi based on human games. |
|
| Author: | Galation [ Sat Mar 12, 2016 10:09 am ] |
| Post subject: | Re: Alphago and komi |
Actually, komi determines the strategy for Black ( and White as well). As Michael Redmond stated, in the 15‘ review of the 2nd game of Alphago vs Lee Sedol challenge, there were big expectations to see how the AI could promote the fuseki to take the lead when using Black, so I think that is complex to say that human statistics are basically better than computer calculated higher number statistics. In the past, something similar proved to be effective in other game as well (e.g. best first moves in Backgammon). Galation |
|
| Author: | bayu [ Wed Mar 16, 2016 1:42 pm ] |
| Post subject: | Re: Alphago and komi |
Slightly related to the question of komi: Did somebody hear or read about whether AlphaGo used an opening book for the first 2 moves or so? It started its games very human like. So, during the last couple of thousend years, did humans figure out how to play the beginning properly or did AlphaGo play that way because of humanely influenced training or an opening book? |
|
| Author: | Uberdude [ Wed Mar 16, 2016 3:16 pm ] |
| Post subject: | Re: Alphago and komi |
AlphaGo doesn't use an opening book, but it has been trained on many human games which is why its play looks human. I've heard Deepmind are thinking about training a new AI only with self play from the rules rather than from human experts which could well lead to alien Go, or maybe it will rediscover all our josekis and opening theory. |
|
| Page 1 of 1 | All times are UTC - 8 hours [ DST ] |
| Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |
|