- My Forums
- Tiger Rant
- LSU Recruiting
- SEC Rant
- Saints Talk
- Pelicans Talk
- More Sports Board
- Fantasy Sports
- Golf Board
- Soccer Board
- O-T Lounge
- Tech Board
- Home/Garden Board
- Outdoor Board
- Health/Fitness Board
- Movie/TV Board
- Book Board
- Music Board
- Political Talk
- Money Talk
- Fark Board
- Gaming Board
- Travel Board
- Food/Drink Board
- Ticket Exchange
- TD Help Board
Customize My Forums- View All Forums
- Show Left Links
- Topic Sort Options
- Trending Topics
- Recent Topics
- Active Topics
Started By
Message
Another win for AlphaGo
Posted on 12/6/17 at 7:35 pm
Posted on 12/6/17 at 7:35 pm
AlphaGo played Stockfish 8, the strongest chess program available. Well, the second strongest apparently. Out of 100 games Alpha go won 28 and never lost, drawing the rest.
What's more, AlphaGo was "turned on" not knowing anything other than the rules of the game and a multimillion game database to learn from. In only four hours it was ready to take on Stockfish.
Story here and academic paper with full description here
Interestingly, AlphaGo decided to mostly play the English and Queen's Gambit with white and the Caro-Kann with Black. It gave the French defense a whirl early but gave up on it despite never losing with it.
AlphaGo did it's thing by pure machine learning. As a calculator it wasn't that impressive, seeing only 80k positions a second while Stockfish saw 70 million a second.
If you like, you can download Stockfish for free but it will kick your arse.
Insert obligatory Terminator pic here.
What's more, AlphaGo was "turned on" not knowing anything other than the rules of the game and a multimillion game database to learn from. In only four hours it was ready to take on Stockfish.
Story here and academic paper with full description here
Interestingly, AlphaGo decided to mostly play the English and Queen's Gambit with white and the Caro-Kann with Black. It gave the French defense a whirl early but gave up on it despite never losing with it.
AlphaGo did it's thing by pure machine learning. As a calculator it wasn't that impressive, seeing only 80k positions a second while Stockfish saw 70 million a second.
If you like, you can download Stockfish for free but it will kick your arse.
Insert obligatory Terminator pic here.
2
Posted on 12/7/17 at 11:29 am to foshizzle
Well, white has the initial advantage, both in tempo and having the majority priority on deciding what game is going to be played.
But just like when it played Go, Alpha Go isn't bound by typical chess progressions and is a bit more "creative" than the standard plays.
I imagine that Stockfish followed the classical programming and more than likely used current meta, which in terms of the English and Queen's Gambit, does allow for white wins. On the other hand, Caro Kann is very drawish. AlphaGo probably realized that the initial disadvantage of being black dictated the approach to not lose when playing black and to utilize the advantage when playing white to try to secure victories.
The most interesting parts of the Go experiment was that AlphaGo did not care about margin of victory. It just cared about probability of victory. And, it would chose to win by 1 point if it had a 100% chance to win versus winning by 300 points if there was only a 99.999% chance to win. Most humans prefer the buffer and obtaining a larger 'perceived' advantage... even if the engines dictate that there is a finite reduction in probability of winning. I believe a lot of this had to do with the psychological effects it has on the participants, but that doesn't really apply to the computers.
I'd like to understand why AlphaGo decided to choose those particular variations.... and if it would continue to do so in the future. A lot was learned about Go theory after the exhibition. I'd love to see some evolution in Chess as well!!
But just like when it played Go, Alpha Go isn't bound by typical chess progressions and is a bit more "creative" than the standard plays.
I imagine that Stockfish followed the classical programming and more than likely used current meta, which in terms of the English and Queen's Gambit, does allow for white wins. On the other hand, Caro Kann is very drawish. AlphaGo probably realized that the initial disadvantage of being black dictated the approach to not lose when playing black and to utilize the advantage when playing white to try to secure victories.
The most interesting parts of the Go experiment was that AlphaGo did not care about margin of victory. It just cared about probability of victory. And, it would chose to win by 1 point if it had a 100% chance to win versus winning by 300 points if there was only a 99.999% chance to win. Most humans prefer the buffer and obtaining a larger 'perceived' advantage... even if the engines dictate that there is a finite reduction in probability of winning. I believe a lot of this had to do with the psychological effects it has on the participants, but that doesn't really apply to the computers.
I'd like to understand why AlphaGo decided to choose those particular variations.... and if it would continue to do so in the future. A lot was learned about Go theory after the exhibition. I'd love to see some evolution in Chess as well!!
Posted on 12/7/17 at 12:49 pm to foshizzle
Page 6 of the academic paper shows the winning percentage of each game! 
Including White's win/draw/lose games and Black's win/draw/lose games.
Super cool!!!
I'm doing a bit of analysis on why it chose the games it did.
Including White's win/draw/lose games and Black's win/draw/lose games.
Super cool!!!
I'm doing a bit of analysis on why it chose the games it did.
Posted on 12/7/17 at 1:36 pm to Mahootney
Not seeing why it liked Caro Kann over Ruy Lopez as a response to King's Pawn.
Only 6% net victory for black vs 12% with RL. (4w-1L vs 6w-0L) in the sample (50 as black).
Although CK has a lower percentage of allowing white wins than the RL (50% vs 52%), the Sicilian games are wayyyyy better for that (32%, 30%, & 22%).
In fact, the CK gives white the second highest win percentage when including the full 100 game sample (50white and 50black). So, why in the hell play it as black?
But I guess as black, there are more potential loss scenarios with the Sicilian.
However, variation B30 is by far the best game for Black in minimizing White's chance to win versus KP opening.
Additionally, the use of English Opening and Queen's Gambit are shocking based on the statistics.
Middle of the pack in wins. Bottom half in overall win percentage. Bottom half in wins and overall percentage based on openings without a loss.
It's just very hard to see any particular statistical reason for its preference.
Additionally, the EO gave it the most wins when playing black!!!! So, why would it choose to play it as white?!?!
Could it be that it has "human-like" preferences. Where it is just more interested in those games?
Or... that it is studying those games more deeply, or assessing/exploiting weaknesses in the Stockfish gameplay?
I'd love to see what the Grand Masters say about it!!!!
Only 6% net victory for black vs 12% with RL. (4w-1L vs 6w-0L) in the sample (50 as black).
Although CK has a lower percentage of allowing white wins than the RL (50% vs 52%), the Sicilian games are wayyyyy better for that (32%, 30%, & 22%).
In fact, the CK gives white the second highest win percentage when including the full 100 game sample (50white and 50black). So, why in the hell play it as black?
But I guess as black, there are more potential loss scenarios with the Sicilian.
However, variation B30 is by far the best game for Black in minimizing White's chance to win versus KP opening.
Additionally, the use of English Opening and Queen's Gambit are shocking based on the statistics.
Middle of the pack in wins. Bottom half in overall win percentage. Bottom half in wins and overall percentage based on openings without a loss.
It's just very hard to see any particular statistical reason for its preference.
Additionally, the EO gave it the most wins when playing black!!!! So, why would it choose to play it as white?!?!
Could it be that it has "human-like" preferences. Where it is just more interested in those games?
Or... that it is studying those games more deeply, or assessing/exploiting weaknesses in the Stockfish gameplay?
I'd love to see what the Grand Masters say about it!!!!
Page 1 of 1
Popular
Back to top
Follow TigerDroppings for LSU Football News