"Bold Sir Lancelot" | DeepMind's AlphaZero vs. Stockfish

2-time British Chess Champion Matthew Sadler takes a look at one of the new games between DeepMind's AlphaZero and Stockfish published in Science in December 2018. Replay the game on an interactive board: chess24.com/en...
Matthew analysed the game together with Women's International Master Natasha Regan, his co-author for an upcoming book on AlphaZero’s approach to chess: Game Changer www.newinchess...
Chess24 is working with Matthew and Natasha to share insights on AlphaZero with the chess community.
With thanks to DeepMind for their support.

[@Mrius86]

This is like watching someone draw an arch, then they draw a half circle, then a full circle. Then they keep drawing and you say, "hang on, what's going on", and BOOM suddenly the've drawn up the complete plan, on all floors, for a fully operational Death Star.

[@16prospero]

A lovely analogy for the 40+ somethings here. Thanks.

[@Varney_of_London]

Regardless of how you feel about Stockfish's strength or the match conditions, there can be NO doubt that Alpha Zero has an understanding of positional chess that is beyond anything we've seen from any classical chess engine to date. Alpha's long term plan for getting that Knight to G6 was amazing.

[@Varney_of_London]

@@masters.1000 I don't know about that. Just look at 'how' Alpha Zero destroys Stockfish; Alpha Zero isn't just outplaying Stockfish out of an equal position...maybe winning a Pawn, and riding that Pawn to victory...Alpha Zero is thoroughly crushing it positionally. The games that we are seeing aren't that close; Alpha is stomping Stockfish. As a result, I don't think any version of Stockfish is beating Alpha Zero.

[@red75prim]

@@masters.1000 On the other hand, DeepMind could train AlphaZero another couple of hours, if defeating SF10 would happen to be problematic.

[@Varney_of_London]

@@masters.1000 Dont stop there; tell the full truth or dont bother speaking.... Yes, this version of Stockfish won 6 times, but still lost the match... In fact, every version of Stockfish that faced Alpha Zero lost the match to it... So yeah, Alpha Zero is superior to Stockfish.

[@BB-ki7he]

@@Varney_of_London Until Alpha Zero proves it in actual competition that everyone is allowed to see in real time then i dont think you can say that. There is no proof they didnt cherry pick 1000 games. My Stockfish suggests different moves in almost all the games shown . Alpha Zero should have to compete and win competitions to recognized as the best imo.

[@BB-ki7he]

@@Varney_of_London if it truely is superior they should enter it into chess engine tournaments and prove it. Playing games behind closed doors and cherry picking games for chess KZbinrs to analyze screams propaganda to me.

[@joxxx20]

excellent recap and excellent game analysis. Short, compact but not too fast, easy , understandable and shown with passion. Thank you, perfect video

[@aleksanrnaskela]

What excellent about this analysis? Where do you see analysis in this video? Everyone can make such video: this move, that move, this move, that move. From GM I would expect more

[@mariusdafunk]

GM Simon Williams, a prophet of his time, knew the importance of the H-Pawn :P

[@TheOriginalGankstar]

LOL :D

[@jocaingles8464]

Put it in H

[@harmhoeks5996]

If only humans would care that much about the long-term

[@pasqualeuk]

amazing game and beautiful commentary. You can tell how Matthew Sadler is passionate and excited by the game as smile so much :-) thanks for all your videos!

[@mrsommer84]

hi chess24. I'm at 2'24'' right now and already now I like your enthusiasm you're presenting the game with. Keep it up! As much as Alphazero loves to fix the center 2:03, as much I can hear your enthusiasm. First time I watch a video of yours!

[@KrazyHades]

Absolutely enjoyable commentary. Your enthusiasm is infectious!

[@zombiefireman]

Wonderful analysis, Matthew. Thank you!

[@CHESSTERUK]

I enjoyed your analysis and I love your enthusiasm Matthew.

[@petergreen5337]

Beautiful lesson and demonstration.Thank you very much publisher.

[@mattherzberg991]

Your great appreciation for the beauty and art of chess is contagious. Thanks for the videos.

[@robertirimias3945]

I like how he always manages to slip the line with his co-author Natasha and the book they wrote in there

[@dannygjk]

You don't like to earn more money?

[@valeriobertoncello1809]

Well he analyzed these games for his book (?)

[@ehsansabaghian5617]

Awesome series of videos Matthew. Very interesting. I feel Alpha Zero will change the game forever.

[@gregmath314]

These videos are excellent, Matthew! Thanks!

[@badrinathchess6718]

Beautiful analysis....alpha zero a master class piece play monster!!!

[@GOVAUS1]

Beautiful to watch and listen. Thank you.

[@Detherocable]

I can't decide if Alpha Zero plays more Kasparov-like or Karpovian! On one hand, it has this terrific feel for dynamics like Kasparov and the way it swings across the breadth of the board and the way it delivers the killing blow....But on the other hand, it has these "probing" moves much like Karpov did, that induce further weakness and irritates the opponent while slowly putting their pieces in a bind and driving them into complete passivity....! One of the recurring themes in these games is that of "Two weaknesses", and before you know it, the whole position looks like cheese.

[@HighPlanes]

what an incredibly entertaining commentary!

[@aboninna]

real good show by deep understanding of a high level grandmaster thanks so much and pls continue

[@archaeopteryx7405]

I need your book Mr. Sadler. This is the greatest revolution of chess game! I'm counting the days missing to official publishing of the book..

[@umityayla5051]

Simply stunning! Excellent! Thank you for sharing this magic game, it is an initial sign of hidden treasure. Those knight moves are just incredible. Am I the only one who thought that Alpha Zero is not just an algorithm anymore.

[@OHOHOHCOME]

Alpha zero is just an algorithm just as much as human beings are algorithms too.

[@SeresTheZocker]

14:34 I love the feeling of sudden unexpected existential dreed

[@misium]

Excellent storytelling, exciting action, deep characters.

[@gg4u481]

To think that it thought of this game as a possibility and tried its hardest to persue it

[@penearth]

Great video series

[@florianlipp5452]

off topic: Mr Sadler has the most pleasant english voice I ever heard. Does anyone know what his accent is?

[@Lasselucidora]

He is a Welch sheepfarmer with their typical accent.

[@louismiller4374]

He is a gremlin

[@kamesenin298]

Isn’t there any match of AlphaZero playing against itself ? Or AZ playing against other engines (Houdini, Leela, Komodo) ?

[@Levijeh]

You probably saw it already but there is one game AZ vs AZ and it's out of this world. It's in Agadmator's channel.

[@dimitriskontoleon6787]

13:30 why stockfish lose so much time to just protect the a pawn? i mean he spent 3 moves on endgame from just protect the A pawn?

[@alieskandari6036]

Now I understand the whole concept of positional playing style of Capablanca. The beauty of this style is logic instead of mind blowing calculations. One was human , another is machine!Even super engine like Alpha Zero prefer positional advantage 🤗

[@asherasator]

This is nothing like Capablanca. The positional play here is completely different than Capablanca & very tactical.

[@alieskandari6036]

asherasator If you think positional playing is just fancy ideas, look at Capablanca vs Alekhine first five games( games before the famous world championship).As Polgár mentioned, always tactics justifies positional achievements

[@mrdraw2087]

It would be nice to see how Stockfish would fare in open positions or positions with some tension in the center. Against Alpha Zero closing the center is pretty much suicide.

[@MarcEllie31]

Love it. Thank you

[@jackmallory7996]

Matthew Sadler, what a sexy charismatic beast, the Elvis Presley of chess.

[@vazquezb2011]

The PetrosianZero Gambit.

[@genomos90]

mister knight go along your way. lol

[@KrishnaKumar-np3tw]

Excellent!

[@andreikarakozov2531]

So great!

[@janiko5017]

what about a6 on 10:35?

[@ChessStuff64]

You have nice soothing voice makes me fall asleep!

[@mim8312]

I think that too many people are focusing on the game, which I also follow, as if this were an ordinary player. Since I have significant knowledge, and since I believe that Hawking and Musk were right, I am really anxious by the self-taught nature of this AI. This particular AI (and its more generalized, even more recent variant MuZero) is not the worrisome thing, albeit it has obvious, potential applications in military logistics, military strategy, etc. The really scary part is how fast these were developed after AlphaGO debuted. We are not creeping up on the goal of human-level intelligence. We are likely to shoot past that goal amazingly soon without even realizing it, if things continue progressing as they have. The early, true AIs will also be narrow and not very competent or threatening, even if they become "superhuman" in intelligence. They will also be harmless, idiot savants at first. Upcoming Threat to Humanity. The scary thing is the fact that computer speed (and thereby, probably eventually AI intelligence) doubles about every year, and will likely double faster when super-intelligent AIs start designing chips, working with quantum computers as co-processors, etc. How fast will our AIs progress to such levels that they become indispensable -- while their utility makes hopeless any attempts to regulate them or retroactively impose restrictions on beings that are smarter than their designers? At first, they may have only base functions, like the reptilian portion of our brain. However, when will they act like Nile crocodiles and react to any threat with aggression? Ever gone skinny dipping with Nile crocodiles? I fear that very soon, before we realize it, we will all be doing the equivalent of skinny dipping with Nile crocodiles, because of how fast AIs will develop by the time that the children born today reach their teens or middle age. Like crocodiles that are raised by humans, AIs may like us for a while. I sure hope that lasts. As the announcer in Jeopardy said about a program that was probably not really an advanced AI long ago, I, for one, welcome our future, AI overlords.

[@PuzzleQodec]

I am very much reminded of 'the Pearl of Zandvoort', Euwe-Alekhine: www.chessgames.com/perl/chessgame?gid=1013180

[@FloydMaxwell]

"Sign of greatness in a young player...how they move their knights" My youngest son is a MONSTER with his knights. So much so that I offered to play him with one piece removed from each side at the start of the game -- my rook, his knight -- and he refused.

[@krishrao2778]

Petrosian is reborn.

[@johnkom2339]

Shades of Karpov too.

[@64blip]

"One of the most popular lines in modern day chess". I don't recall seeing this position in any of the recent WC games between Carlsen and Caruana. d4, c4 when black didn't play d5? I'm not good at chess but this doesn't look usual to me.

[@dannygjk]

Those two aren't the only strong players in the world. Plus you are looking at a tiny sample of games.

[@adebayoadegboyega1658]

@@dannygjk The main human move I believe is f5!?. But the engines see things differntly

[@Varney_of_London]

LOL!!! I laughed out loud when Alpha played Rook H1 as I can already see that this is gonna be bad for Stockfish. :)

[@mikemcwilliams2735]

Never seen anything like this in my entire chess career. This is artificial intelligence... the beginning of it...

[@AlvaroGonzalez-tw2or]

I really hate how you walk us through teh game up to a point where you don't and assume we all understand what is going on. On 12:20 you let a whole lot happen without explaining or teaching a thing and they seem crucial, nor how the end game went sour.

[@RoulDukeGonzo]

Could a team of human chess players working together with ample time defeat Alpha Zero?

[@The_savvy_Lynx]

Maybe a few % of the games, because Neural Nets like AZ may be close to perfect, but arent 100% perfect. There will probably be some slight misjudgements in AZ's positional game, which when abused perfectly could net the humans the occasional win. Though, I think getting some draws is still much more realistic than getting actual wins vs AZ.

[@davidc5191]

The only problem with this analysis is that Sadler anthropomorphizes AlphaZero too much, ascribing to it "likes," "preferences," etc. which of course it doesn't have. In his book "Game Changer" he's more circumspect in attributing human-like behavior to this program.

[@bc_7644]

Definitely, there's a lot of folks who believe a skynet scenario is coming and the fact that people like agadmator humanise az (in his original remarks) doesn't help at all.

[@andraslibal]

You do know there is no understanding and planning there in the background just a very sophisticated filter being deployed on the possible moves and outcomes, right?

[@CobraXXVI]

Chess engines are predicated on human understanding of chess. AZ is superior because it does not have humans to interfere with its own "mind" and experiences. Think of Stockfish as the best of human ignorance within our limited understanding of values combined with brute force. The moment we assign these false values of 1,3,5,9 coded into Stockfish's database, it has been corrupted. AlphaZero is the divine creative drawing on its own intelligence unobstructed by human limitations. However, humans are more intelligent than AI in open ended systems because we draw on understanding of the natural world and assign our own values to such. AI is alien and unaware of these processes we draw on, much like how we can't understand AlphaZero in chess.

[@rdecredico]

CHESS PRON

[@StefanReich]

"General purpose AI system"? It plays chess

[@RandomPerson-yq1qk]

And Go and some other games. The goal for it is to become an AGI, but obviously it is not there yet.

[@dannygjk]

If you research neural net systems you will understand why it is general purpose.

[@StefanReich]

@@dannygjk My life's mission is to show people that there are better ways for AI than neural networks.

[@RandomPerson-yq1qk]

@@StefanReich For many specific cases neural networks have been the biggest advancement yet and their practical use is astonishing. I do not doubt that we are still missing a major piece for actual intelligence, but neural networks look very promising as one piece of the puzzle by trying to replicate the process in our own brains which is the biggest blueprint for intelligence that we have. Also Alphazero does do Protein folding. It is quite general purpose compared to 99% of all AIs we have created yet.

[@kiksumas1152]

I guess you don't know the definition of outpost, though it seems to be one of your favourite words, cheers

[@dannygjk]

What!? That is what an outpost is what he said.

[@ChessQuizToday]

Alpha zero plays awful in this game. Alpha zero is not without some serious flaws. Here Alpha Zero loses in 22 moves to an outdated Stockfish 8. Stockfish 8 2018.01.18 AlphaZero London, UK 1-0 Computer Match (255) 1. e4 book 1... e6 book 2. d4 book 2... d5 book 3. Nc3 book 3... Nf6 book 4. Bg5 book 4... Be7 book 5. e5 book 5... Nfd7 book 6. h4 book 6... Bxg5 book 7. hxg5 book 7... Qxg5 book 8. Nh3 book 8... Qe7 book 9. Qg4 g6 10. Ng5 h6 11. O-O-O Nc6 12. Nb5 Nb6 13. Rd3 h5 14. Rf3 a6 15. Qg3 Nd8 16. Nc3 Nd7 17. Bd3 Nf8 18. Rh4 Rg8 19. Bc4 Qd7 20. Nce4 dxe4 21. Nxe4 Nh7 22. Rxh5

[@paranchoy6077]

He doesn't show you how Stockfish could have defended better, so he gives you this false impression that AlphaZero just played magic and Stockfish lost for no reason. As a result, you don't learn anything because there's no logical explanation of what mistakes were made. We all know chess is objectively a draw, so we need to know how an advantage was obtained and how it grew, not just hear a GM try to reach for vague explanations of the moves.

[@valeriobertoncello1809]

No, chess is objectively a win for white

[@paranchoy6077]

@@valeriobertoncello1809 hahaha funny attempt at trolling. See the 99%+ draws in the correspondence world championship

[@Mortizul]

Weird that a British person can't even pronounce 'lieutenant' properly.

[@rick9125]

its pronounced differently in different countries. In aus its pronounced Lef-ten and in the US its pronouced loo-ten

[@grantgre]

Do you have to say “I am GM Matthew Sadler” as if it’s one word. I didn’t know what the hell you were saying because you said it so fast. It sounds a little pretentious man. I mean, is your first name GM or is that your title? Why don’t you just say your first name and then say I’m a GM. Otherwise, I think your analysis are very good keep them coming.

[@Detherocable]

If you use your goddamn brain for once and read the bloody description, you'll know exactly what he said.

[@grantgre]

@@Detherocable Ok, another pretentious Brit!

[@Detherocable]

@panagranit pana Another uneducated Indian who likes his own comments. And by the way, I'm not British.

[@grantgre]

Detherocable You’re just ignorant person. I don’t know of any other chess personality that introduces themselves like that . Plus he is saying an abbreviation for grandmaster so sounds like his first name could be GM or jim so people who don’t know him and how he talks won’t really know if that’s his title or not. If he said I am grandmaster Matthew S.... then I would know what he’s talking about.

[@conjured_up_skeletons6178]

What if this just A0 playing against itself? DeepMind can say it played against anyone-StockFish, god, E.T., etc. Show me actual footage of the match taking place. Putting out a video with a GM certainly does not give me enough indication that this is all credible. Takes more to make me a believer.

[@jeffwads]

If I had to guess...anyone could check the SF moves using SF...what a concept!!

[@conjured_up_skeletons6178]

Sure doofus... Just like I checked the moves with SF from the 2017 match and they totally differed. My foot is down-it's alpha vs alpha!!

[@RobertSela]

Wow what a completely ignorant comment. No one cares if you are a believer or not, it wont make a difference.

[@encrust1]

"...in the way a young player handles his OR HER knights" please