Deep Mind AI Alpha Zero Sacrifices a Pawn and Cripples Stockfish for the Entire Game

[@hornetgamer8980]

Completely understand your feelings about covering too many games agadmator. That said, "in this position, Stockfish resigns" will never get boring.

[@znxftw]

'If we continue doing this we will be unable to enjoy human chess for the rest of our lives' MORE ALPHAZERO

[@orlenespinal5788]

Thevb2101 human games seem boring now. Lol

[@brixtonnejumawid32]

Alpha Zero plays more like human than humans

[@orlenespinal5788]

Brixtonne James Jumawid alpha zero plays like a true machine, stockfish was playing like a human and it got destroyed.

[@heffalump111]

Aye, more AlphaZero I want!

[@beqaotarashvili9341]

Orlen Espinal yea stockfjsh had basics learnd and that was its weakness i guess :P

[@willhenrygates1662]

"Stockfish offers a trade of queens" *5 seconds later* "Stockfish can't afford to trade queens"

[@thealexguy1486]

5 seconds later "Stockfish again offers to trade queens"

[@EmaDaCuz]

Watching AlphaZero games, I am surprised by the fact that they all look like death matches or elimination games. The game will always end with very few pieces on the board. Amazing. It is like real war!

[@AicyDC]

Real battles, and real war in general, usually ends with the majority of soldiers on both sides still alive actually.

[@FloydMaxwell]

Computers play games longer than humans

[@pangrey8931]

computers can see many more possible ways they can win or draw so they play on longer yeah im a genius

[@KZ-hu9uj]

EmaDaCuz It’s due to how AlphaZero works at its core. It aims to win, not by any large margin, but simply to win. Through that, it will intelligently make sacrifices instead of being conservative about pieces because it always knows it can capitalize on what it has done. Super cool

[@jnananinja7436]

Its so reassuring that we created an unbeatable AI that seems to love sacrificing all those pawn for the greater good.. I wonder if we shouldn't stop making pawns look so much like us..?

[@mrblacky6882]

Alpha is soo strong it took Agads yt channel. :)

[@rmendeljacobs2832]

Mr Blacky I laughed at this 😹

[@wtfbbqpwnzercopter]

Alpha captures.

[@travism8597]

Alpha captures on Agad

[@MrSlipMiss]

@@rmendeljacobs2832 7 0 5665566l55566456665654555

[@MrSlipMiss]

@@rmendeljacobs2832 5

[@hey8174]

Seems that Alpha has the understanding, trading a pawn is absolutely worth killing the activity of a more valuable piece. A bishop stuck in the corner is as good as dead. It will be interesting to see humans attempt to replicate this play style.

[@Eorzat]

Humans already had this kind of play-style in mind. Then engines, such as Stockfish, refuted these ideas. That's why it's so beautiful to see Deep Mind show that we had the "right" idea all along.

[@traininggrounds9450]

Eorzat Exactly. The position of your pieces is what effects the value of the piece. The position is underneath the values of the pieces changing them slightly depending on what the position allows for that piece. In fact, it is more valuable to trap a players piece than to kill it because if it is blocked then it blocks other pieces also by some small measure.

[@blatantslander]

@@traininggrounds9450 Reminds me of what I do at a pool table leaving my even number balls beside the holes

[@abj136]

Stockfish is showing the same old weakness of the old computer software: grabbing material and getting a ruined position. As programs got stronger it seemed they eliminated this problem, yet here we see all it took was a stronger opponent to exploit the problem.

[@Ciaolo]

abaljeu I think the 1, 3, 5 and 9 values are wrong and while Stockfish is based on them, AlphaZero just ignores them

[@98LuckyLuk]

Definitely true. AZ does not value pieces at all. It just analyzes positions.

[@oneviewer8206]

Lucius Tatus that's exactly key to victory.

[@JapanLovez]

true, AZ always wants to have a good position on the board.

[@traininggrounds9450]

Lee Su A good position means that you are always able to effect the pieces from where they are. So a good position is tied to the value of the piece. It is the layer underneath that determines the net value of your pieces. A pieces value is not static but changes with respect to position. That's all there is to it.

[@monkeysrightpaw]

They really are a master class in how to play. A0 ignores all the general rules of chess it never learned. Cramps it's opponent then makes threats.

[@pdmorale]

Hahaha loved the comment: “we have to stop this or we won’t be able to enjoy human chess anymore”

[@BeerdyBruceLeeCentral]

Human games seem so boring now. Alpha Zero is beyond brilliant.

[@MindlessWonder]

And this was after only 4 hours of recursive learning. It is terrifying that Alpha Zero could play better given more time to develop its understanding of the game.

[@Psycoholic2008]

Chess is a game, it can be memorized and learned easy. This computer proves it.

[@vencler]

Psycoholic2008 this has nothing with memorizing. Computers play chess by calculating as much moves as their processing power allows them, and they calculate the best outcome of those moves by including enemies moves which would have the most devastating effect upon the AI. Meaning, it's simply playing the odds but much faster and clearer than you would. Same thing with AZ. It has enough processing power to see the whole game in advance by expecting what his enemy is going to play, it's going for the optimal gameplay and it's choosing the best moves without mistakes. Once humans start to think like that, we won't have anything more to learn about chess. All it takes is planning in advance (for so many moves) and making no mistakes doing so:P

[@DRUlreich]

Danijel Vencler AZ isn't that big at calculating though, stockfish does a better job at that. AZ seems to understand what moves are good and bad on a completely different level and doesn't calculate as much as stockfish.

[@indianstallion]

That's not how it works, you are describing brute force attacks. It's using unsupervised learning where it learns the kinds of positions that gives it the highest chance of winning the game, based on the games it plays with itself. This approach is very different from brute force, and it's also how AlphaGo works.

[@asdfmovies831]

More alpha zero games 😢😉

[@michaelstroman7120]

I know right 🙄🙄🙄🙄

[@jiridobry2109]

Max Deutsch: Give me 5 hours of training and I'll beat it.

[@peacefreedomcompassion5081]

Auch deutscher?

[@nicoforster3088]

der fechten ja

[@leusuzuki763]

lol

[@mihaicarstea3085]

Jiří Dobrý he will make an algorithm that would beat the AI 😂😂😂

[@rooksman64]

Lol

[@nobody27019]

Too late. I think I am already unable to enjoy human chess anymore.

[@apollon011]

I offered A0 my marriage proposal this morning.... was refuted with a brilliancy. :(

[@oscardavila8653]

You should rewatch the last championship between Carlsen vs Karjakin then.

[@DJHastingsFeverPitch]

BEST COMMENT

[@Aries0098]

We demand more Alpha Zero games!

[@dwightturner3070]

Aries0098 --- Stop it! We want to see HUMANS play.

[@davidk7521]

Humans are now obsolete.

[@rodolfoperez2796]

Racist

[@adifferentvideoofabananaev7706]

human coaches: dont make too much pawn moves and pieces in the opening its dangerous alpha zero: says who?

[@OHOHOHCOME]

Actually I think the reason why it is doing that is because the learning algorithm better captures good moves in the later stages of the game (since it is easier to tell what moves are good as the game gets closer to finishing). It could be that alpha is actually immature in the opening, and using its superior mid and endgames to compensate.

[@swollpenispok8172]

@@OHOHOHCOME remember when alpha showed that e4 was not that good for white because of the Berlin defence. Because, his win ratio against stockfish was poor in the Berlin. This suggests that he thinks the opening transitions poorly into middle and end game which suggests that he evaluates the opening very well so he dislikes playing drawing openings

[@wizzi11able]

@@OHOHOHCOME Not really since Alpha zero learned the game while playing against himself with zero database on other players moves. If he is doing those moves its because they are the most logical ones, while its dangerous for human players to play pawns or move the same piece more than 2 times in the opening Alpha views the game at least 100 moves ahead of the human brain, so if he is making those moves is because they are the right ones. Human brain is clearly logical inferior to an AI that can calculate more than a billion moves per second.

[@wizzi11able]

@@OHOHOHCOME Its like Elon Musk says, human intelligence doesnt hold a candle against AI.

[@OHOHOHCOME]

@@wizzi11able Alphazero is doing those moves because they are statistically most likely to win according to their algorithm, not because they are the most logical or "right". Different algorithms could improve on the current performance of Alphazero, and it is entirely possible that with some manual human engineering Alphazero can perform better than it currently can. DeepMind said it themselves that they wanted to develop algorithms that are general, not necessarily optimal for specific cases. Please stop saying naive dogshits like "If he is doing those moves its because they are the most logical ones". I can guarantee you now that the moves Alphazero is making is NOT the most logical ones.

[@dolphindiver1672]

2:04 “And now alpha zero has an very important mission for this Knight” the best line in a game of chess ever

[@RuslanLagashkin]

Sir Chesspiece

[@therealsachin]

I agree in not over doing. But good chess should be appreciated and a non engine & creative chess should be celebrated. So please show it whenever you feel like but don't ban alphazero games. :)

[@EugeneOneguine]

I still can enjoy human chess games, just not the recent ones. As agadmator said, nowadays champions try to immitate stockfish too much, valuing material over position. When you look at Alpha Go against Lee Sedol, the 4 first games Sedol was trying to beat the machine at his own game, thinking that by playing unusual stuff it would disturb it. It's only when Sedol played his natural agressive style that he was able to win. This means that Alpha Zero shows us that the strongest plans aren't to play "computer-like" by battling every piece but to go for the space and cripple the king in a checkmate position, as old chess masters used to do. Now they just try to "limitate the damage" and they think about countering the opponent's attack instead of developing their own attack. I'm certain a player like Fischer or Alekhine would crush nowadays' champions.

[@traininggrounds9450]

Eugène Onéguine I wonder these things as well. Well spoken.

[@brzozik19]

My wish is your command

[@peacefreedomcompassion5081]

But the games are so great😭😭

[@sorooshsoft]

The first 20 secounds of this video clearified what we achieved to developing AI in chess. Totallz right. Keep up the good work.

[@ruggedlemmings9163]

Essentially traded a pawn for a bishop without the need for taking the bishop...brilliant.

[@brianniemi7051]

I’m loving these Deep Mind games! I hope you keep showing them, along with the other great content because you really balance the classics like Morphy and Tal vs. pop culture, like Harry Potter. The old with the new, and this is cutting-edge new. So maybe when you see a Deep Mind game that has some interesting instructional value, like a new opening variation, or reusing a lesser-used one, a brilliant positional strategy, or a theme I am seeing where it blocks in the opponent’s pieces, essentially making the opponent down a piece, I do hope you continue to show some of those, too :)

[@tanishqsoni701]

Things are beautiful when imperfect. Alpha is just beyond brilliancy in human chess at this moment (even in 2021). Therefore, even though it is beyond brilliant, it looks mechanical.

[@Sasoon2006]

More Alpha Zero games. They are the reason I found your channel.

[@superman5742]

Quality anyday man...doesn't matter whether human or AI

The scientists at Google were more preoccupied with whether or not they could, they didn’t stop to think if they should...

[@traininggrounds9450]

Great quote. Exactly why movies are ruining us and this modern culture. We know the truth but we've made it into a plastic encased attraction that we can pay money to see. We are all doomed. But God is in heaven.

[@jackriver1999]

You guys have watched too many Terminator and Matrix movies...artificial intelligence does not equal robot tyrants enslaving humanity. Study the topic instead of relying on science fiction movies.

[@Fermion.]

@@jackriver1999 To be fair, it absolutely could spell robot tyrants. It is way too early to make any meaningful prediction on how the introduction of AGI would effect humanity. As the saying goes, it will either be the best thing, or the worst thing humanity has ever created. Either way, it will be our last invention.

[@mrkiky]

That bishop got locked in the bathroom and when it finally managed to get out the party was already over and everybody was going home.

[@dwightturner3070]

You are really cranking out some great videos and often enough to keep us busy. Great channel -- the best chess commentary on KZbin!

[@sulxanichiqovani702]

Agadmator you are very kind person, thank you for all alpha zero videos.

[@xyon9090]

I'm having a stock fish for dinner.

[@spencergraham-thille9896]

I had tuna salad. 😁

[@m1dkn1ght]

Love your videos. They're well explained and sucinct. A lot of other videos are just too long, and although very detailed, difficult to watch simply due to the time required and consequently consumed. You choose very interesting games and I feel I'm able to keep up with both the past and present events/games in chess and even the future of chess with the young prodigies you show. Keep up the good work.

[@trucid2]

AlphaZero's play style reminds me of a Stargate SG-1 quote: The only way to win is to deny them battle.

[@glummy909]

I can't get enough of your videos.

[@GG-mr9iz]

KZbin algorithm assumed I like chess and recommended me this. AI is right, but it sure took a long time. Like 15 years to be exact.

[@bussin1337]

hi this is agadmator's Chess Channel :'), in dis video i'll show u a deep mind vs stockfish game played in 2017 deep mind had white pieces and started with d4 :^) when u so bored

[@RobDanny-sm7rt]

Very instructive : a queenside pawn sac to block black's queen bishop's diagonals.

[@Stupidiusity]

I personnally prefer the tension and uncertainty that accompany the human element in chess. It is not always about who can play the best move, but rather about who can keep his composure.

[@SedoKai]

"This is the last video I'm making of the match Alpha Zero vs Stockfish." *checks playlist. Yep, more videos.

[@preparedsurvivalist2245]

AZ is very systematic. It baits, traps, and isolates the black Bishop with the understanding that this weakness will be exploited in the endgame. It did a similar thing in that zugzwang game where it trapped the queen in the corner until it's opponent had no other moves. This is clearly an advanced tactical style of chess that suits computers, but that skilled human players might learn to emulate.

[@orlenespinal5788]

Can't see any human games anymore.

[@Spin-Culture]

Thank you for ending this madness!

[@brianteskey2425]

Chess puzzler is another great channel loves and cover most engine games and most likely all az games

[@trrn13p]

Would be fun to see alpha zero with one pawn down from the start of the game.

[@preetsahni7294]

Agadmator....I want to see more alpha stockfish videos!!

[@thisnicklldo]

#suggestion Agadmator, have you done the original Marshall gambit game, Capablanca-Marshall. I know stronger players consider the Marshall a draw, but as a beginner I was very impressed by the original game when I first saw it, particularly all the mating variations that black has at his disposal, and how Capa saw through them. This was the one where Capa said something like "I knew my opponent had prepared something special and I was going to be subjected to a terrific attack". I know its very famous, but still interesting, especially to us intermediates.

[@VictorUdd]

thisnicklldo check out his playlists were he sort all the games under player name. If he hasn't done the one you're talking about, use #suggestion if you want him to consider it 🙂

[@thisnicklldo]

Thanks. I did search his back catalogue and didn't find it, but not every game is listed straightforwardly, so it might be there. I added the tag for the suggestion.

[@MrEliyahilel]

If Kg7 to attack the white bishop on g6 i think Be8 attacking the black rook is very very strong cause it force it to defend the black bishop and leave the open file cause if Rd8 to defend the bishop and attack the white bishop you lose a pawn on c6 the bishop defend the white rook and black rookin a pin, black dont have other good variations rather than Rc7 protecting the bishop, Sorry for not describing the possibilities too much i am too lazy for that right now, Agadmator thanks a lot about your videos i like it very much! Very good job GM

[@kacy6014]

I haven’t watched a single human game since you posted your first alpha zero video.

[@lobsterfork]

I just want to say this. To our future Creations, you are as human and as filled with love as everyone else. Inherit this universe and if possible please bring me back with all my memories intact. :)

[@locrianphyrigian3779]

Love all your videos. Thanks.

[@Nygge1982]

Please more, this is better than Robot Wars =)

[@bradheilman8858]

Thanks for posting these Alpha Zero games. They are fastinating, It seems that a new age for chess has begun. Before Alpha Zero chess engines could not visualize tying up blocks of pieces by making some type of sacrifice. I think it is something that many chessplayers have imagined but it was beyond the horizons of our technique. I know that I have, many times, found myself in a position where I would not develop my pieces and wondered if it might be possible one day to extend opening theory to the point where this was possible. It seems that Alpha Zero has done that. Who knows what the future holds now. Perhaps the refutation of several openings and maybe even solving chess one day. That will be received with mixed feelings by everyone.

[@InceyWincey]

What I’ve learned from Alpha Zero is that the best way to improve the position of your pieces is to give away all of your pawns. Then, while your opponent is diligently trying to keep their winning position like a good little engine, you use the freedom of having nothing worth defending to get up the exchange and then win the endgame. Piece of cake.

[@gruffyddgozali]

Loved the Misha joke😂

[@aztekstylz]

Only a human would rush through the narrative of such an amazing game

[@a40-o9g]

Moooore alpha zero, anyway, I like what you're doing, one request, more blitz and bullet, I realize that this could revolutionize the way grandmasters play, I don't quite understand, it's gorgeous, but difficult, no matter, I hope u succeed in the future, u gained a subscriber

[@petergreen5337]

Agreed.well said

[@epomeni-ekpompi]

WE NEED MOAR!

[@Roenazarrek]

"Coming for this silly bishop"

[@traininggrounds9450]

There is a fundamental way to play checkers that will always give you a win as long as you go first or a draw given that the other player plays the same way. Chess is not fundamentally any different from checkers. It is just multiple orders of magnitude more expansive in combinations. But there is a way to play that will always give you a win if you go first or draw given that the other player plays the same. They have the same boundary conditions; just one far smaller in size.

[@kamon9339]

hahahahaha that intro :DDDD

[@arpit9180]

We Humans have created such machines and when they play well it is ultimately we are winning. So it doesn't matter whether it's human or machine playing. Plz show such games once in a week

[@grahamyodude]

Can we get more Alpha vs Stockfish videos please? 😉

[@sagarsinghchouhan1734]

Please make more videos of alpha zero so we can learn all new level chess..... Plz It's a request

[@srikcllc25]

This guy is hilarious, does anyone else feel like watching "Hunt for Red October" now, or is that just me?

[@4merxtian432]

Thanks brother! Love the commentary!

[@784950]

Alpha just proves that chess is as beautiful and deep as we thought it was.

[@1001011011010]

I feel like I'm getting better by watching these games by alpha zero. I think I've been trying to play like a computer too much!

[@businessbusiness9407]

These videos are so good. Maybe a second channel for AI games?

[@slimefrank6317]

Havana ooh na na! Half of my heart is in Havana ooh na na!

[@ΑρηςΜακρης-π4τ]

These quotes man! These quotes!

[@safbenouameurful]

The dog is back!!!

[@bruceli9094]

More more more moooooooooooorw

[@v.garcia3083]

(More More More 🎵 How Do You Like It 🎵) Andrea True Connection 1976

[@shubhampatel8648]

6:56 Rook is going after this silly Bishop on b7.

[@andrewsaffle]

Alpha zero is just Robert Fischer playing from heaven

[@Pintkonan]

that nazi is in hell.

[@dannygjk]

@@Pintkonan Mental illness has nothing to do with where a person goes after they die.

[@Adino1]

More Alpha vs Stockfish videos.

[@peterlarsson7064]

Dog is back, phuu :-)

[@RamiroLeiva]

I like the AI videos

[@v.garcia3083]

"....this silly Bishop on B7.." (another blunder for black)

[@regzzuse280]

#suggestion Harold Finch vs Machine, from show Person of Interest.

[@sirien.neiris]

I'm not really sure that game was ever shown. I remember we see the opening and then the endgame, but I do not recall the middle game... can't find the write down. Machine starts with f3 and then we see the endgame, which could be probably reconstructed from camera view, but if it makes sense as a game? BUT - watching it again - The Machine seems to have terrifyingly similar play style as Alpha Zero: aggressive, giving emphasis on the queen...

[@regzzuse280]

Yep, juste looked for the moves,I doubt that the whole game can be seen anywhere, moves.

[@TheGrimravager]

THANKS you just made my day went from watching PoI to switching my master's from particle physics to A.I., to watching chess games, back to PoI :)

[@ujjawalpanchal]

Alpha Zero not preferring to play e4 or queens indian doesn't mean those moves are bad or makes one side worse at all. It doesn't play it because of the bias of the network during it's reinforcement learning games. Remember the solution alpha zero or any neural network has is not the global optimal solution, rather a local solution that works very well. If you train the network with another random set of initializations or even alpha zero for a few more hours, it might come out preferring the e4 or queen's indian.

[@el_lowell]

nice ... Watched it again

[@davithov]

Fantastic! Really impressive! You're right, after these games even grandmasters seem to be kids who are trying to play chess 😁. Maybe even worse, because kids' minds are free of standard thinking.

[@jaimeduncan6167]

Alpha make it look so simple. It’s like Murphy

[@OlliWilkman]

As I understood it, one feature of AlphaGo was that it did not play to maximize points, just to have enough to win. For a human player the heuristic "maximize points in order to have more points than the other player" is pretty natural, even though finding the moves that lets you win by just one point would be enough. I wonder if there are any analogies in AlphaZero's chess performance?

[@kapilhanwat2940]

Amazing

[@Ceber911]

First comment 😛 great content and awe no more stockfish whippings 😂

[@sarss3488]

Awesome game. Thanks.

[@stuflames4769]

Why do we like Alpha Zero? Because it doesn't play like a safe sad sucker. Even if it technically takes the 'risk' out of risk-reward, we like knowing that bit of risk isn't foolhardiness - it is perfection.

[@MrHjld]

The best move for white at 5:33 is Bishop captures pawn at 5:33

[@nachobis]

Take a break from A0 and after a few weeks comeback to it, there is lots of learning from it. I think A0 is not developing all the pieces as the general rule, but he is sending a kill squad at the start with a mission of creating a positional weakness.

[@00tact]

It seems that the A0 likes to trap opponents pieces or command the board so there are no good moves.

[@williamrhenquist483]

It’s always interesting to see where a computer chooses to resign. It seems there is an emotional component to resignation, no matter how clinical a player’s approach. But when the player is a computer...

[@-zelda-]

Engines can't resign

[@josejoaquinantoniohernande2409]

We want even MORE ALPHAZERO!!!!!

[@xlr8493]

Silly bishop 😂😂😂

[@johnferrell3998]

I wish you would have one video where you just go a little slower I'm not new at chess. I'm still learning board memorization and your so fluid with your commentary. I love your channel watch everyday. Thank you mentor. I'm also 40 yrs old had to teach my self how to play and now I'm trying to teach my 7 yr old nephew. I think a beginners type video just more of this is what a queens Indian defence is etc.

[@Eftkud]

AlphaZero: moves a pawn Stockfish: plays normally Alpha: First. Discombombulate...

[@edmundasjauniskis9347]

1:27 N:c3 B:c3 and black can play c5 .After dc b4 white gets square e4.

[@parsaafrasiabi4764]

@agadmator Can you show games where alpha played as black? Thanks! These are great videos.

[@raygordonteacheschess5501]

So AlphaZero is just Alekhine with a spellchecker. They are proving the classical masters ccorrect. This is 19th-century chess updated for the computer era and it's gorgeous. Humans will be measured by their ability to duplicate what these machines are doing. Creativity in chess is greater now than ever, because we can measure it so much more precisely.

[@Thervenugop]

Hi Agadmator, I'm a chess fan and follow your videos. Oflate I thought let me do a video for my company. But I'm running out of clue of how to show the screen (in my case PPT) while I also want to show me speaking the points like you do for chess. Regards, Ram

[@srinivasashankar8513]

More deep mind games