AlphaZero AI beats champion chess program after teaching itself in four hours

[Gollum]

https://www.theguardian.com/technology/2017/dec/07/alphazero-google-deepmind-ai-beats-champion-program-teaching-itself-to-play-four-hours

 

Scary ****, AI freaks me out no end. 1700+ years of chess theory (right from Gupta period) and techniques explored by some of the greatest geniuses to walk on this planet. AlphaZero learns the game in 4 hours and demolishes the World Champion Chess Program Stockfish. 

 

Go and Chess have been conquered by AI. When will Poker fall to this new beast?

Edited December 8, 2017 by Gollum

[Malcolm Merlyn]

Can it beat Kasparov or Carlsen?

[Gollum]

1 hour ago, Malcolm Merlyn said:

Can it beat Kasparov or Carlsen?

If you go by ELO most of the top engines are rated 3300 plus while Carlsen is 2830-2840 range. Strictly based on ELO ratings difference if Carlsen and a top engine (eg Stockfish) were to play, the machine would score 96 out of 100 (according to the maths formula), it could be 96 wins for the machine and 4 for Carlsen, 95 wins for machine, 2 draws and 1 Carlsen win....92 wins for machine and 4 draws etc.  Here AlphaZero beat Stockfish 64:36 with 28 wins, 72 draws and no loss !!! So yes it will crush Magnus. It has not entered the rating list but my guess will be 99-1 for the machine or 99.5-0.5 (assuming it is for all practical purposes 150 ELO points above Stockfish)...........Magnus himself has told that he hates playing the engines because he always loses. 

 

Kasparov lost to Deep blue in 1997 and it was an insect compared to modern day engines which are like dinosaurs. In fact Kasparov's style is worst possible match up against a machine because he is a speculative player who takes many risks, against computers you need to be very accurate. Tactically unsound sacrifices, sharped edged dynamic play (Alekhine, Tal, Kasparov, Topalov) etc can throw human opponents off guard and * their psychology but against a machine which is a non emotional, cold hearted calculator and perfect tactician no chance. Only a positional master with extraordinary endgame technique (Capablanca, Karpov, Kramnik, Carlsen) can even draw against these engines. In fact before Deep Blue Kasparov had also lost to Chess Genius some time in the early 90s but in the next round Chess Genius was obliterated by Anand. Similarly Kramnik (proto Carlsen) performed much better than Kasparov against stronger computers (various versions of Fritz). Kramnik and Anand were not better players than Kasparov but were better at playing anti computer moves. IMO only in the mid 2000s did computers become better than man, 1997 Deep Blue was just a publicity move by IBM and Kasparov played like a paranoid clown. So yes if Deep Blue could beat Kasparov in 1997 I see no reason why AlphaZero won't crush him today.

 

Carlsen is the most suited player to play these engines not just because of his positional mastery, endgame specialty and endless stamina but also because he has grown up in the chess engine era, constantly exposed to computers right from the moment he learnt the rules of chess. So compared to earlier masters he has a better chance of resisting a chess engine because he can predict their modus operandi better. But still he will lose very very badly. Endgame is one area where computers can be outplayed by the best humans occasionally, but Magnus doesn't have the opening knowledge or tactical level required to get such favourable endgames. I could wager that even if the 6 living world champions (Spassky, Karpov, Kasparov, Kramnik, Anand, Carlsen) rediscover the fountain of youth and join forces to take on Stockfish together they will still get smashed.

Edited December 8, 2017 by Gollum

[Old guy]

Isn’t to much more complicated than chess ?

atleast from point of view of ai it’s much more difficult to program or make it learn so this makes sense .

[maniac]

Can we buy it and then make it program to teach itself to be a phhhasht bowler

 

 

[maniac]

skynet like scenario is inevitable....time to buy a leather jacket,sun glasses and hit the gym people!!!

 

 

[Muloghonto]

3 hours ago, Gollum said:

https://www.theguardian.com/technology/2017/dec/07/alphazero-google-deepmind-ai-beats-champion-program-teaching-itself-to-play-four-hours

 

Scary ****, AI freaks me out no end. 1700+ years of chess theory (right from Gupta period) and techniques explored by some of the greatest geniuses to walk on this planet. AlphaZero learns the game in 4 hours and demolishes the World Champion Chess Program Stockfish. 

 

Go and Chess have been conquered by AI. When will Poker fall to this new beast?

Poker will fall to AI if and only if AI is advanced enough to completely replicate the human brain- including impulsive, irrational thought. Unlike chess, Poker is not a game of logic only, there is bluffing, aka the art of winning with a losing hand. 

 

[MechEng]

2 hours ago, maniac said:

Can we buy it and then make it program to teach itself to be a phhhasht bowler

 

 

Remember the movie Real Steel? I think around 100 years from now, we'll see AI humanoids playing competitive sports instead of real humans, they may end up doing unreal stuff such as a fast bowler clocking 300 kmph (if cricket still exists at that time).

[sandeep]

1 hour ago, MechEng said:

Remember the movie Real Steel? I think around 100 years from now, we'll see AI humanoids playing competitive sports instead of real humans, they may end up doing unreal stuff such as a fast bowler clocking 300 kmph (if cricket still exists at that time).

Chess doesn't require mechanical engineering, only software.   That's why computers can compete.  The latest cutting edge robot can barely walk on uneven ground without falling on its metal face.   Lagtaa hai Friday drinking jaldi chalu kar diya Manju-niac ne.  

[MechEng]

13 minutes ago, sandeep said:

Chess doesn't require mechanical engineering, only software.   That's why computers can compete.  The latest cutting edge robot can barely walk on uneven ground without falling on its metal face.   Lagtaa hai Friday drinking jaldi chalu kar diya Manju-niac ne.  

I guess Manju-niac would be more apt if the post is rephrased to "teach itself to be Imran Khan".

[Gollum]

6 hours ago, Muloghonto said:

Poker will fall to AI if and only if AI is advanced enough to completely replicate the human brain- including impulsive, irrational thought. Unlike chess, Poker is not a game of logic only, there is bluffing, aka the art of winning with a losing hand. 

 

Another reason why chess engines have become very powerful is because for many decades elite players have been developing chess playing algorithms/decision tress and collaborating with the coders (got a sharp boost from Botvinik's era onwards) . In poker there has hardly been any effort by the top players towards developing equivalent algorithms and let's face it even if they do put in such efforts in the future their ability to make algorithms pales in front that of the of the chess players.

 

Besides is there incentive to develop poker engines? Computers have defeated the best chess players in the world, but the game goes on as before. If people cheat during tournaments with the help of chess engines and are caught, they will get banned. Poker is a different beast altogether because of the money involved. If computers crack poker, there is the possibility of cheating at the tables, in online versions it will be even worse and everybody starts losing money.