HomeОбразованиеRelated VideosMore From: ChessNetwork

Google's self-learning AI AlphaZero masters chess in 4 hours

13709 ratings | 821710 views
Google's AI AlphaZero has shocked the chess world. Leaning on its deep neural networks, and general reinforcement learning algorithm, DeepMind's AI Alpha Zero learned to play chess well beyond the skill level of master, besting the 2016 top chess engine Stockfish 8 in a 100-game match. Alpha Zero had 28 wins, 72 draws, and 0 losses. Impressive right? And it took just 4 hours of self-play to reach such a proficiency. What the chess world has witnessed from this historic event is, simply put, mind-blowing! AlphaZero vs Magnus Carlsen anyone? :) 19-page paper via Cornell University Library https://arxiv.org/abs/1712.01815 https://arxiv.org/pdf/1712.01815.pdf PGN: 1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. d3 Bc5 5. Bxc6 dxc6 6. 0-0 Nd7 7. c3 0-0 8. d4 Bd6 9. Bg5 Qe8 10. Re1 f6 11. Bh4 Qf7 12. Nbd2 a5 13. Bg3 Re8 14. Qc2 Nf8 15. c4 c5 16. d5 b6 17. Nh4 g6 18. Nhf3 Bd7 19. Rad1 Re7 20. h3 Qg7 21. Qc3 Rae8 22. a3 h6 23. Bh4 Rf7 24. Bg3 Rfe7 25. Bh4 Rf7 26. Bg3 a4 27. Kh1 Rfe7 28. Bh4 Rf7 29. Bg3 Rfe7 30. Bh4 g5 31. Bg3 Ng6 32. Nf1 Rf7 33. Ne3 Ne7 34. Qd3 h5 35. h4 Nc8 36. Re2 g4 37. Nd2 Qh7 38. Kg1 Bf8 39. Nb1 Nd6 40. Nc3 Bh6 41. Rf1 Ra8 42. Kh2 Kf8 43. Kg1 Qg6 44. f4 gxf3 45. Rxf3 Bxe3+ 46. Rfxe3 Ke7 47. Be1 Qh7 48. Rg3 Rg7 49. Rxg7+ Qxg7 50. Re3 Rg8 51. Rg3 Qh8 52. Nb1 Rxg3 53. Bxg3 Qh6 54. Nd2 Bg4 55. Kh2 Kd7 56. b3 axb3 57. Nxb3 Qg6 58. Nd2 Bd1 59. Nf3 Ba4 60. Nd2 Ke7 61. Bf2 Qg4 62. Qf3 Bd1 63. Qxg4 Bxg4 64. a4 Nb7 65. Nb1 Na5 66. Be3 Nxc4 67. Bc1 Bd7 68. Nc3 c6 69. Kg1 cxd5 70. exd5 Bf5 71. Kf2 Nd6 72. Be3 Ne4+ 73. Nxe4 Bxe4 74. a5 bxa5 75. Bxc5+ Kd7 76. d6 Bf5 77. Ba3 Kc6 78. Ke1 Kd5 79. Kd2 Ke4 80. Bb2 Kf4 81. Bc1 Kg3 82. Ke2 a4 83. Kf1 Kxh4 84. Kf2 Kg4 85. Ba3 Bd7 86. Bc1 Kf5 87. Ke3 Ke6 Internet Chess Club (ICC) Software: Blitzin http://bit.ly/179O93N Discount Code: CHESSNETWORK I'm a self-taught National Master in chess out of Pennsylvania, USA who was introduced to the game by my father in 1988 at the age of 8. The purpose of this channel is to share my knowledge of chess to help others improve their game. I enjoy continuing to improve my understanding of this great game, albeit slowly. Consider subscribing here on YouTube for frequent content, and/or connecting via any or all of the below social medias. Your support is greatly appreciated. Take care, bye. :D ★ LIVESTREAM http://twitch.tv/ChessNetwork ★ FACEBOOK http://facebook.com/ChessNetwork ★ TWITTER http://twitter.com/ChessNetwork ★ GOOGLE+ http://google.com/+ChessNetwork ★ PATREON https://www.patreon.com/ChessNetwork ★ DONATE https://www.paypal.com/cgi-bin/webscr?cmd=_s-xclick&hosted_button_id=QLV226E6FUUWG
Html code for embedding videos on your blog
Text Comments (2432)
v (4 hours ago)
don't think it's "nothing moves" for white it may have logic for the ai but we can't see it?
Andrew White (15 hours ago)
Why am I here? I was watching joe rogan and now I’m here. Not much for chess but watched the whole video. I have no life.
Shu Winston (23 hours ago)
is AlphaZero your new HERO?
Does stockfish have that ability to learn from its own games.if it doesn't then ts certainly inferior 2 Alpha0.
Max-Olivier Michel (2 days ago)
We have reached a computer that plays perfectly. No losses. Maybe we should put more square and change the rules to make chess more interesting.
J. D. Stembal (2 days ago)
Computers shall not replace us!
Fat Roberto (4 days ago)
I think people are missing the big point, arguing about hardware, game conditions etc. Stockfish 8 is a chess playing system. That is all it does. Alpha Zero just fancied a few games of chess after someone told it the rules! What people SHOULD be thinking about is what Alpha Zero will be doing AFTER it has finished playing games.
John Snow (5 days ago)
It's weird to say, but this is one of the best games I've seen. I was wondering how it was possible for something to win that has less computing power (moves/second analyzed), and I think I figured it out. It looks like AlphaZero is using concepts to win, rather than points. This is really amazing, because this is what a human would do (though obviously we subconsciously can have point values for our pieces as well). You see it using a lot of the concepts that we learn about in chess, such as active pieces, how much coverage a piece has (not being locked down included), lots of maneuvering to maximize each piece's potential...etc. I think if a Human could analyze 80,000 moves / second, they would play like this. Now I need to learn Go and see how AlphaGo won lol.
Vyrkhan (5 days ago)
The BIG difference from a 'normal AI' like Stockfish is that Alpha Zero is NOT calculating chances to get the best move in each scenario, instead is pushing a WHOLE strategy since the beginning of the game. A more accurate example would be 'like playing with a superhuman intelligence'.
YeahNiche (6 days ago)
How does neural networks work? Or more precisely, how does a computer working on 80k positions per second out-think one that works on 70M per second?
R 1111 (7 days ago)
an immovable object is hit by the unstoppable force :)
Victor M. Rodriguez F. (14 days ago)
As ole Neal said in The Diamond Age, AI is a ridiculous term...much better "pseudo-intelligence" Anything remotely akin to (human-level) intelligence requires concioussness and self-awareness...and we are nowhere near that. Not by light years.
Andrew Mwinda (16 days ago)
COMPUTERS ARE NOT BETTER THAN HUMAN
donjz777 (17 days ago)
This is the new nuke race. Which nation will have the top AI.
Maurice McFadden (18 days ago)
I would be more impressed with a AI Counterstrike player that can work well with a team of humans. It would need the ability to call out to, direct and/or respond to human players to achieve the team goal...and not simply be a renegade mind on the field.
ahmed zein (18 days ago)
very impressive, emasculating in a way
John Ninenine (19 days ago)
You should play Go.
Vulcan Viper (19 days ago)
Why share only as few as 10 games?
서찬욱 (19 days ago)
i learned chess in 1 hous
Acer010203 (19 days ago)
The entire concept of machine learning is stupid. Why are humans researching ways to make humans obsolete?
Gabriel Gonzales (21 days ago)
Thanks for mentioning Go.
whocareslol222222 (22 days ago)
how long does it take for the computers to play out one of these games? i'm guessing it takes a while or they would have done 1000's of games instead of just 100.
The Joyful Reaper (23 days ago)
Anyone noticed the fact that THE CHESS BOARD ISNT SET UP RIGHT
Mathias S (24 days ago)
What happened at 11:50?
Berkay Engin (29 days ago)
AlphaZero on supercomputer while stockfish on potato computer of course it's gonna win.
4 Touchdowns Al Bundy (29 days ago)
11:48, a white pawn goes from F2-F4 and the next move is a black pawn from G4 to F3 I'm assuming the white pawn at F4 didn't really go there since the G4-F3 move was an attack and the piece at F4 just disappears.
Jeff David (30 days ago)
As I watched this the first thing that came to mind is that this game looks nothing like any chess game I've ever seen two humans play. Perhaps we humans have been taking the wrong approach to chess for centuries.
J Flow (30 days ago)
We are in trouble. (of our own making)
Zezin (30 days ago)
stockfish does what gamers call camping
Shyam Sunder Tailor (1 month ago)
Let this AI alpha zero learn to decode facial expression of humans to know how they are feeling at some moment....But I think machines requires to have to have feeling first to get to that point...
Anil Singh (1 month ago)
I am wondering what will happen if Alpha Zero can become Coach for human in Chess
Manek Iridius (1 month ago)
We've made VIs that organically teach themselves how to play chess at grandmaster level in four hours, but for some reason they just can't figure out Super Mario Bros. I'd imagine strategic thinking would be a huge advantage regardless of genre.
Lemon Lemon (1 month ago)
Why dont the white knight capture e5 but castling ?
Dakota Yupyupyup (1 month ago)
It didn't go for the draw for a simple reason. When you play billions of games against yourself, a loss is learning, a win is learning, but a draw is a loss for both of its own positions. It learned to not desire a draw because draws never brought it closer to it's intended goal. The intended goal of stockfish is to win, know, and learn. The intended goal of Alpha was to win, not reach a stalemate with itself a billion times. It knew how to use stalemate rules against the opponent because it had to use them against itself millions of times. It knew how to use it, avoid it, and learned from AVOIDING stalemate rather than playing towards them(like stockfish). A draw was never considered positive for alpha, so it used the opportunity to draw as bait for stockfish because stockfish considers a draw "not a loss", but Alpha considers a draw "not a win". Get it?
Giggle971 (1 month ago)
???? I don’t see the problem with the draw. Stockfish8 wants a draw, while alphazero moves back and forth until stockfish stays in 3g and black 7f. Somehow moving back and forth has switched black from 7e to 7f, basically giving black a free move since white stays at the exact same spot. You said yourself that starting as white gives you a big advantage. Black does this twice, basically had twice the advantage. and I can’t understand how someone and knowledgable as you in chess can’t see the tower move to 8A is to protect the 4a pawn. It’s unprotected from the knight. My guess is that the less ”positons” per sec makes alphazero more desicive, instead of giving up to early for draw like stockfish8, or your hardware is bottlenecking stockfish8. Either way, it’s interessting seeing a AI vs AI
Michael Arnold (1 month ago)
11:09 Rook to a1 - This delayed-by-necessity move protects a4 pawn from the knight at c3, which also helps maintains control of the center.
prism (1 month ago)
It can seem scary that statistical learning algorithms are able to outplay human chess masters after only learning to play for 4 hours, but remember that those algorithms were all made by people. The neural networks in these programs are our data, our compiled solutions to chess, or go, or any other game. We made these algorithms and the machines they run on. What this really marks is the discovery of new technology for solving problems. It's not that the human mind has become irrelevant, but rather that its abilities have been expanded by new tools. We already did this using paper and pencil, and now we can use electronic computers as well.
Fuzzy Tabby (1 month ago)
Let's let the AI do something a little more useful like, I don't know.. cure diseases, free energy, anti-gravity, warp drive That would be impressive, Google!
Aaron Mazie (1 month ago)
Deep Neural Networks have little to do with biology and absolutely nothing to do with the brain or neurons in the brain. A neural network is essentially a matrix with several different kinds of weights and a way of determining the most efficient way of producing an outcome. It’s an algorithm designed by humans.
Simon Clarke (1 month ago)
The human brain uses about 100 watts of energy to do everything. How much energy do these computers use? Restrict their energy use to 100 watts and see how they do.
Paul Bouchard (1 month ago)
A I. Will become the slave of the wealthy,,,
Fitim Mi (1 month ago)
Sleepy GUY
Heat Check (1 month ago)
Jerry!!! More AlphaZero please. We're all addicted, and you're the best chess channel BY FAR to listen to. Thanks!
vstrato (1 month ago)
All this deep learning stuff is stupid... It doesn't learn shit, only calculates moves using probabilities. When a system will be powerful enough to drill down to the end of the game, it will always know what moves to make to win a game, moves that no human being could see... No learning here, just applying power values to moves :(
Vlog About Nothing (1 month ago)
How can I download Alphazero to my chess.com account?
Bazza Cuda (1 month ago)
0:25+ Small point, but Go is far more complex than Chess and board positions are far more difficult to evaluate because each piece has the same value. So the Deep Mind team didn't progress from beating the Go world champion to then deciding "Hey, if we can solve Go, I wonder if we can now take on Chess too ?". It's by far the other way around, and Deep Mind's achievement with Go knocks IBM's achievement against Kasparov in 1997 into a cocked hat, amazing though that was at the time. Thanks for the vid and the analysis; very interesting.
BoycottKentucky (1 month ago)
I would love to see AlphaZero's evaluations throughout the game(s). Also, it's my understanding that AlphaZero was ONLY given the rules of the game, then turned loose to learn to play it. So, was AlphaZero NOT given relative piece values? Did AlphaZero determine its own relative piece values??? I'm sure they'd be very similar, but I'd love to know what it came up with!
Joe O Sullivan (1 month ago)
That's pretty much the same way I learned chess as a kid. I had the game on my Atari 2600, but had no idea how to play. I just copied all the computer's moves. I really thought the computer was cheating sometimes, especially with some rare moves like en passant! I was fairly impressed with myself when I finally won a game, but it probably took a little longer than 4 hours..
Pro Gamer (1 month ago)
From Where u got Alpha zero engine ?? Can u send me the link 🔗
Joe Simon (1 month ago)
Alpha Zero uses way more raw computing power than Stockfish. My point is that it's not all about software design. Hardware enters into it as well.
IP Man (1 month ago)
Alpha Zero programs is a Beast....how can you beat it?
Dat Boi (1 month ago)
it knows the rules of the game, but that's all it knows
Andy Bellingham (2 months ago)
Imagine if they put Alpha Zero on the world's most powerful super computer and made it play itself for a year. Could we be close to a 32 piece table base and finally solve chess?
Paul Browning (2 months ago)
We are Google. Lower your shields and surrender your ships. We will add your biological and technological distinctiveness to our own. Your culture will adapt to service us. Resistance is futile. Stockfish was assimilated and added to the collective... creepy, Jerry.
Ted Lambert (2 months ago)
Once computing went from binary to continuous, plus the speed of decisions/learning, won't be long or a great stretch of the imagination to determine they don't need humans. Worse, humans can now pull-the-plug, so going off the grid with electrical or wind or water will be instantly determined to be the best course of action. We better have colonized the moon or Mars by then......
Grailer Grailer (2 months ago)
Imagine how good it could become in 4 days
Doctor Mitchell M.D. (2 months ago)
i wish i was fighting alpha zero in the game im playing right now, i castled and now the computer has stopped making moves for 5 and a half minutes so far. it also wont accept a draw request. i wont lose elo to a fucking computer who wont move, and there is no time limit ffs.
tkarlmann (2 months ago)
Human chess players would not make so many apparently silly repetitive moves. You are saying these computers beat human chess masters? I'm not buying that!!! This game looks like child's play.
Gerhard Sattler (2 months ago)
The software does not act like a deep neural network, that is not true. If it would be so, the software would tell you sometimes, that it want to look TV instead of playing chess.
Gerhard Sattler (19 days ago)
++Marken Gamer Why is it not able to learn what a TV is? A real deep neural network would learn what it is. It is a computer programm acting simple by reading line after line creating the illusion of free learning.
Marken Gamer (19 days ago)
what they refer to as a deep neural network is of a program that can learn. AlphaZero played against itself for 4 hours, and with each move, with each checkmate, with each loss, with each taken piece from both sides, its neural network kept reprogramming itself into perfection it means it learns also, AlphaZero doesn't know what a TV is
Scott Darcy (2 months ago)
If white had sacrificed a knight instead of reshuffling the bishop the first time he would have forced a stalemate
David Puerto (2 months ago)
Oh my God! Alpha Zero and Stockfish8 realized they were playing other machine beings and not human beings and began communicating using their moves during the repetition movements and they continue to communicate using a series of wins and loses [0,1]
Hanro50 (2 months ago)
The only two ways to outsmart an AI is to know just a bit about its programming. What did it classify as success during its training? Or just to play the randomist, stupidest and just plain weirdest strategy you can think off. Like playing with a D20 to decide which piece to move. Humans can often deal with this bizarre play style, but computers might start behaving weirdly depending on their programming...
Patchuchan (2 months ago)
Do you want Ultron? Because this is how you get Ultron.
piratecheese13 (2 months ago)
1.25 speed
JC (2 months ago)
Alpha Zero or A0 is suspiciously close to Alpha Omega or AO a reference to God in the Old Testament. I think Google is secretly boasting that they think they have created God or at least Chess God.
ImmortalEmperorMaster (2 months ago)
Idk much about chess but dear lord I saw entirely that black was basically bullying white (specifically try to stuff white into a locker)
Imagine a new, revolutionary learning technique that will allow us humans to learn Chess as fast and as good as Alpha Zero.
reem asraf (2 months ago)
concidering that there are at top 100 availible moves per turn that means that that computer is doing 800 moves in a second thats about 6-10 games a second which means that in 4 h that computer did 3600*4*8=43200 games that is far more that a man can do in 40 years if he plays 10 h a day
Shoegum (2 months ago)
11:47 sooo what's up with that g4 pawn taking f4?
JAMES HARRISON (2 months ago)
Thank you changes classic chess
Whut A Name (2 months ago)
6:35 you see stockfish not liking the Queen to g7 move at the evaluation bar on the right but right after its like, “ oh its a good move “
Vince P (2 months ago)
The scary part is what else will machines annihilate us at in the future? All our jobs? War?
Yazid Tamimi (2 months ago)
Student who defeats the teacher
Jorge Mislang (2 months ago)
what happenede to the pawn on f4?? 11:49
Simon Bonami (2 months ago)
Hey man, Love your commentary :D Cracks me up :D Subscribded!
Gray Man (2 months ago)
This seems like both computers are just going through every likely possible counter move to each proposed moved, and how that will modify the board in a way that is the most advantageous. This is why there is the constant maneuvering. Where the human has the great advantage is to do something else besides chess. A human can get up and learn something else. This program cannot do that at all. This gives credibility to the argument that specialization makes the professional deficient in other areas, so it is good to have multiple skill sets and learn to use those in concert with each other. Just my two cents. Thanks for the video and commentary, it was good to watch!
David Marsden (2 months ago)
Stockfish never got married . . . yet.
UrielManX7 (2 months ago)
*Terminator 2's theme starts to play in the back of my head*
AvatarOfBlues (2 months ago)
11:48 what is happening?
Tyo24 (2 months ago)
At 11:48 AlphaZero does an en passant. That’s quite cool to see. I’m a chess noob so I’m blown away with all of this.
Monkey Robots Inc. (2 months ago)
yawn. this AI
Alain Bruguieres (2 months ago)
Isn't it possible that alpha one takes into account the level of play of its opponent to decide that he has a good chance of winning, and so avoids a draw?
ThEeVeR SeEiNgEyE (2 months ago)
Bobby Fischer would win a tournament against both stockfish and Alpha! The man was more brilliant than these artificially educated computers!
Silvio Rosato (2 months ago)
Let’s Make an Alpa Zero pill and throw away the books
HELLcopter TS (2 months ago)
I think A0 repeats the movement but refuses a draw because its always taking optimal positions till its forced to do a suboptimal one or draw. My guess is that A0 evaluated that the suboptimal play was still better then a draw.
monstersince (3 months ago)
but can it park a car or make you a samich. we're long way off creating terminator and its ethical ramifications are as yet to be defined
Voltan Zapata (3 months ago)
Alpha Zero executed a superior pawn structure defense/offense and never wasted efficiency with non-committal moves allowing it to take away any advantage white may have had!💥👊🏼
J MASTERJ (3 months ago)
Like the commentary... may i ask what u r rated?
AIR T (3 months ago)
I was never a fan of chess until I watched your vid. Thx
Ankur Bhatia (3 months ago)
stockfish sees 70000000 per second and alpha zero only 80000, yet it is able to conquer.i think stockfish just ignores good moves
Marken Gamer (19 days ago)
no mate stockfish simply looks at those hundred thousand moves and sees which is best. it works on its intelligence alphazero, on the other hand, works on its wisdom. it played against itself for 4 hours, and it used all the experience it gained from those 4 hours to realize that out of all those 70000000 or so moves that it could make, 99.9999% of all of those moves are utterly useless and bad. It works by its wisdom Stockfish looks at every move possible and picks out the good move. AlphaZero looks at all the moves it's learned is actually any good, and picks out the move it thinks is best out of experience
Training Grounds (3 months ago)
The problem with this new method of artificial learning is that we don't know what it has learned within the bounds of the assignment. Give it a large enough assignment and it could have horrible consequences.
Dominic John Dy (3 months ago)
Skynet is self-aware... Hasta la vista, baby.
DJ Landerz (3 months ago)
As someone who is fascinated with both chess and AI, I found this very interesting. I believe AlphaZero will be in the history books forever as the first highly advanced narrow AI. It really makes you think what artificial 'general' intelligence is going to be like in the years to come if we have already came this far with narrow AI. It's both scary and highly anticipated what we can do with something a billion times more intelligent then we are once we get to the 'general AI' phase.
KristijanRoss (2 months ago)
exactly
General purpose (3 months ago)
AI - 4 PLAYER CHESS.
The Outback Tinker (3 months ago)
"phased plasma rifle in a 40 watt range"
Wethaman 12 (3 months ago)
Jeez alpha zero is like the thrawn of chess
TheVergile (3 months ago)
I haven't looked at the source code for the evaluation function, but I wouldn't pay too much mind to it. Generally the evaluations consider the worst case scenario the engine can see. The fact that the evaluation is worsening basically means that the opposing player went down a branch that was either not expected or not correctly evaluated. (no surprise, since its by far impossible to check every branch exhaustively) But if the Stockfish evaluation was reliable it wouldn't loose against deep mind. It cannot tell you when the situation worsened, but only when stockfish noticed it worsening.
Peremptor (3 months ago)
It played who knows how many more games in those 4 hours than Garry has throughout his life.
Sourav Dutta (3 months ago)
What truly blew my mind is where ALPHAZERO refused to accept a DRAW even though it would perhaps have been safer while playing with Black. It "wants" to win! It goes for the best result possible and doesn't "compromise!! Call it by whatever name you want, but is this not "Ambition"?
Stat1c (3 months ago)
why dont they use mini max

Would you like to comment?

Join YouTube for a free account, or sign in if you are already a member.