https://www.chess.com/news/view/google-s-alphazero-destroys-stockfish-in-100-game-match

Google's AI just crushed the best chess AI :O

next thing we know it's gonna be number one in all our ladders.

RIP queens indian defense xD

Watch https://www.youtube.com/playlist?list=PLDnx7w_xuguHIxbL7akaYgEvV4spwYkmn

Edited 12/7/2017 19:37:01

Wow, Google is rapidly taking AI to a whole new level. A couple years ago I learned about AI at university as being artificial with the connotation of being "fake" intelligence. So far I have only skipped over a few youtube videos but as I see so far this is the first "real" artificial intelligence getting cretead here, a bot which can teach himself a broad magnitude of games. I have always been kinda a sceptic on AI tbh but now stuff like the ultimate artificial soldier suddenly quite fast seems to become more than just an Elon Musk mumbo jumbo.

For those on chess.com


https://www.chess.com/club/warlight-players

@Norman: You should check out DeepMind's papers (a lot of which you can just find YouTube videos for). A solid half of the hype-getting deep learning papers right now are coming from these guys, it's almost scary. They seem to be going all-in on deep reinforcement learning as a big step toward general intelligence and have really pushed that field forward.

AI has been crazy ever since the ImageNet breakthroughs around 2012.

@Chatul

Don't forget that WZ is not really comparable, due to how the turns and fog prevent complete information. Not to mention picks. Getting 6th for example, and you can beat someone who plays perfect. So no, an AI can never really beat WZ.

@TBest:

Given my old uhm... "WarZone" bots I have put a bit of thought into how difficult / easy WarLight is. Maybe I didn't Google with the correct terms but I found quite little mainstream information on the general difficulty of AI problems, see for example here: https://en.wikipedia.org/wiki/AI-complete. Apparently AIs taking over jobs are mostly still theoretical thoughts and people have more urgent matters than finding out which jobs fall away first. As for WarLight I'm not really sure how difficult the game is however at least the fog shouldn't cause that much trouble.

I'm probably the only person who could consistently beat that thing at Warlight.

Then why can't you get better than 15th on the 1v1 Ladder?

For those who are interested on more in-depth analysis on the topic.

https://en.chessbase.com/post/alpha-zero-comparing-orang-utans-and-apples

TLDR: It turns out that there are some shades in this stunning 28:0 (=72) result. Stockfish played on small and iappropriate hardware, with its databases for openings and endgames switched off. Moreover , the used time control( 1 minute per move) make no sense , either for humans and computers, and cause huge problems for Stockfish's way of finding proper moves. I would also like to know why only 10 games were released. Sure, they are beatiful ( I have seen all of them already on YT), but considering the fact AlphaZero played against crippled version of Stockfish , having its opponent playing in way lower strength that it is supposed to be... well, you can play nice games agaist way weaker opponents anytime ;)
In overall, this looks like Google marketing way more than anything else. There was no proper match - Google experts just set up opposing program in the way it had to lose, inevitably.

I'm probably the only person who could consistently beat that thing at Warlight.

You're the last one capable at beating me, so don't even think about beating an AI.

So no, an AI can never really beat WZ.

famous last words!

PS- nice new avatar btw ;)

Edited 12/16/2017 01:39:53