Google's AlphaZero Machine Learning program learned the total human history of Chess in 4 hours then beat the strongest chess engine on the market.
https://arxiv.org/pdf/1712.01815.pdf
Some demo games are at the bottom of the doc.
Google's AlphaZero Machine Learning program learned the total human history of Chess in 4 hours then beat the strongest chess engine on the market.
https://arxiv.org/pdf/1712.01815.pdf
Some demo games are at the bottom of the doc.
Last edited by sonatine; 12-07-2017 at 09:22 AM.
"Birds born in a cage think flying is an illness." - Alejandro Jodorowsky
"America is not so much a nightmare as a non-dream. The American non-dream is precisely a move to wipe the dream out of existence. The dream is a spontaneous happening and therefore dangerous to a control system set up by the non-dreamers." -- William S. Burroughs
Yep.
No need to be concerned about AI, fellas.
I thought Todd was god?
"Birds born in a cage think flying is an illness." - Alejandro Jodorowsky
"America is not so much a nightmare as a non-dream. The American non-dream is precisely a move to wipe the dream out of existence. The dream is a spontaneous happening and therefore dangerous to a control system set up by the non-dreamers." -- William S. Burroughs
Real talk btw, these were also 1m per move games. I think if you bump that up, the margin of victory could go through the roof for AlphaZero.
"Birds born in a cage think flying is an illness." - Alejandro Jodorowsky
"America is not so much a nightmare as a non-dream. The American non-dream is precisely a move to wipe the dream out of existence. The dream is a spontaneous happening and therefore dangerous to a control system set up by the non-dreamers." -- William S. Burroughs
I haven't studied chess in years, so I'm not familiar with Stockfish, but I read that Google took away its opening book for this study. Which might level the playing field and all, except Stockfish was designed to use that opening book, so they deliberately picked a gimped opponent.
I mentioned before how OpenAI basically lied about the competency of their video game bot, and I also observed some shenanigans with the HUNLHE bot at Carnegie Mellon.
Machine learning is fascinating, don't get me wrong, but these researchers need to chill the fuck out and stop pressing their thumb on the scale.
"Birds born in a cage think flying is an illness." - Alejandro Jodorowsky
"America is not so much a nightmare as a non-dream. The American non-dream is precisely a move to wipe the dream out of existence. The dream is a spontaneous happening and therefore dangerous to a control system set up by the non-dreamers." -- William S. Burroughs
Also I ran these through stockfish personally and by the time middle-game started, these positions were often in perfect equilibrium, which was frankly haunting.
If the opening books had been gimped in favor of AlphaZero, we would be seeing consistent and significant positional if not material advantages manifesting well before middle-game started to evolve in earnest.
"Birds born in a cage think flying is an illness." - Alejandro Jodorowsky
"America is not so much a nightmare as a non-dream. The American non-dream is precisely a move to wipe the dream out of existence. The dream is a spontaneous happening and therefore dangerous to a control system set up by the non-dreamers." -- William S. Burroughs
https://www.chess.com/news/view/goog...100-game-match
For one thing, Claudico got crushed by the human team by 9.6bb/100, but the researchers called it a draw, claiming that the winrate wasn't large enough to be statistically significant. lol give me a break.The player with most strident objections to the conditions of the match was GM Hikaru Nakamura. While a heated discussion is taking place online about processing power of the two sides, Nakamura thought that was a secondary issue.
The American called the match "dishonest" and pointed out that Stockfish's methodology requires it to have an openings book for optimal performance. While he doesn't think the ultimate winner would have changed, Nakamura thought the size of the winning score would be mitigated.
"I am pretty sure God himself could not beat Stockfish 75 percent of the time with White without certain handicaps," he said about the 25 wins and 25 draws AlphaZero scored with the white pieces.
Libratus was, by all accounts, horrifyingly good. Probably unbeatable. But it did apparently have a limping strategy with some glaring weaknesses, and the human team felt that formulating a strategy around this was their best chance of winning, or at least keeping it close as possible. The AI team caught wind of this, so they just reprogrammed the bot to never limp. In the middle of the competition. lol
Fucking Naka.
Of the 10 games, I just randomly picked 4 of them, making sure 2 of them included Stockfish as white.
In every single game, by move 12, Stockfish considered the game to be in either perfect equilibrium or white to have a .1 or .2 piece advantage, and at least one of these was with Stockfish as white, who went on to face a significant disadvantage when entering end game.
If Stockfish's opening books were in any way crippled here, by move 12 it would have found itself screwed, blued, and tattooed.
If someone has the CPU power to let early middle game positions in these games really cook for a few hours and discovers that somehow, AlphaZero was sitting on some deep-future advantage, that would be a significant booster to Naka's argument.
Otherwise really he's just being an attention whore, the case for which is only strengthened by the deficit of sympathetic opinions.
"Birds born in a cage think flying is an illness." - Alejandro Jodorowsky
"America is not so much a nightmare as a non-dream. The American non-dream is precisely a move to wipe the dream out of existence. The dream is a spontaneous happening and therefore dangerous to a control system set up by the non-dreamers." -- William S. Burroughs
So I read GM Larry Kaufman's comments which make a bit more sense than Naka's, which probably werent reported particularly well. Basically he seems to be saying that somehow AlphaZero built its own opening books and that somehow Stockfish couldnt hang because it's weak without 'established' book openings? Again I dont see how Stockfish can be so comfortable with its middle game positions if thats the case.
What I found more interesting was the inference that the distributed processing powering behind AlphaZero gave it an advantage. I think there is much to be said there in a 1m per move match. But again without more information, it really could just be sour grapes.
"Birds born in a cage think flying is an illness." - Alejandro Jodorowsky
"America is not so much a nightmare as a non-dream. The American non-dream is precisely a move to wipe the dream out of existence. The dream is a spontaneous happening and therefore dangerous to a control system set up by the non-dreamers." -- William S. Burroughs
Honestly I don’t think chess would be that hard for a human to master. There’s 16 pieces on a total of 64 squares. That equates to a total of 1,024 different possible combinations. While it’s much easier for a computer to memorize 1,024 combinations I don’t think it’s impossible for a human to do it. Not easy, but not impossible either.
AI pushed Kobe into stardom.
Last edited by Brittney Griner's Clit; 12-07-2017 at 08:59 PM.
There are currently 1 users browsing this thread. (0 members and 1 guests)