Deep Mind AI Alpha Zero Refuses a Draw from Stockfish
Vložit
- čas přidán 7. 12. 2017
- Download Mproov and Improve Your Chess Today! app.mproov.me/AgadCZcams1
Follow MprooV on Twitter / mproovapp #agadmator Read more about Deep Mind Alpha Zero here arxiv.org/pdf/1712.01815.pdf
Link to the other games lichess.org/study/wxrovYNH
A chess game between Deep Mind Alpha Zero and Stockfish
Google Deep Mind Alpha Zero vs Stockfish
One of the games
1. e4 e5 2. Nf3 Nc6 3. Bb5 Nf6 4. d3 Bc5 5. Bxc6 dxc6 6. O-O Nd7 7. Nbd2 O-O 8. Qe1 f6 9. Nc4 Rf7 10. a4 Bf8 11. Kh1 Nc5 12. a5 Ne6 13. Ncxe5 fxe5 14. Nxe5 Rf6 15. Ng4 Rf7 16. Ne5 Re7 17. a6 c5 18. f4 Qe8 19. axb7 Bxb7 20. Qa5 Nd4 21. Qc3 Re6 22. Be3 Rb6 23. Nc4 Rb4 24. b3 a5 25. Rxa5 Rxa5 26. Nxa5 Ba6 27. Bxd4 Rxd4 28. Nc4 Rd8 29. g3 h6 30. Qa5 Bc8 31. Qxc7 Bh3 32. Rg1 Rd7 33. Qe5 Qxe5 34. Nxe5 Ra7 35. Nc4 g5 36. Rc1 Bg7 37. Ne5 Ra8 38. Nf3 Bb2 39. Rb1 Bc3 40. Ng1 Bd7 41. Ne2 Bd2 42. Rd1 Be3 43. Kg2 Bg4 44. Re1 Bd2 45. Rf1 Ra2 46. h3 Bxe2 47. Rf2 Bxf4 48. Rxe2 Be5 49. Rf2 Kg7 50. g4 Bd4 51. Re2 Kf6 52. e5+ Bxe5 53. Kf3 Ra1 54. Rf2 Re1 55. Kg2+ Bf4 56. c3 Rc1 57. d4 Rxc3 58. dxc5 Rxc5 59. b4 Rc3 60. h4 Ke5 61. hxg5 hxg5 62. Re2+ Kf6 63. Kf2 Be5 64. Ra2 Rc4 65. Ra6+ Ke7 66. Ra5 Ke6 67. Ra6+ Bd6
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
If you realllly enjoy my content, you're welcome to support me and my channel with a small donation via PayPal or Crypto.
Link to PayPal donation www.paypal.me/agadmator
Maiar Wallet @agadmator or get.maiar.com/referral/pv0mam...
BTC address bc1qckd3ut0hqyymzv33eus97ld8klj02xhk2kcwld
BCH address qzmfclyn69hqhjslls40r7r0dsttwe3tcsl946w4fr
LTC address Laarf1RmvCpLt2BcSwC1PBLG3hRC4HjBrz
NANO nano_1h1kgfaq88t1btwadqzx73rbha5hwbb88sxmfns851kwj8hnosdj51w388xx
Monero 4AdvvqmC4xhPyyRSAEDxTTAoXdxAtX2Py6b8Eh4EQzBLGbgo5rY5Khcap1x76JrDJH87yibAE9b6TPwTsvBAiFFCLtM8Be7
For any other currency address, contact me via agadmator@gmail.com
Check out ALL my videos here • "Grand Opening" - Ande...
Facebook: / agadmatoryoutube
Twitter: / agadmator
Instagram: / agadmator
Lichess: lichess.org/@/agadmator
Chess.com: agadmator
Skype: agadmator
League of Legends: agadmator :) "Watch me without ads on your Amazon devices (bit.ly/Agadmator_Amazon) and Roku TV (bit.ly/Agadmator_Roku) - Zábava
Alpha Zero wouldn't kill humanity. It would sacrifice us for activity.
One of the best comments I have ever read nice play on words
If Alpha Zero self-learned about other human activities would it just give up chess and watch TV?
That's seriously funny dude!
Funnnnnnyyyyyy😂😂😂😂😂
That's excellent. This is the second time I have ever commented on a CZcams video, but that comment warrants it.
"This rook is really on a mission to visit every square on the board."
I laughed so hard xD
Me too
Only Antonio can deliver such a beautiful commentary
Me too xD
that was a really good one by #Agadmator's Chess Channel
I think one day someone has to write a book of these beautiful sentences :)
That rook moved 22 times during the game and visited 20 different squares.
Dude congrats on the running. Health is no joke, wish you the best.
i dont understand why this line wasent in the compilation of best agamoments part 1... :O
THANKSS
Hell yeah
I can't get enough of these Alpha Zero games
this games are so inspiring. i try to play like alpha zero and i think i already got higher rating
me too, I love how the rook travels around the board....It has done its mission, He was chosen by the King to fulfill a mission, The mission was to attact and kill enemies while defending the Kingdom. Now I haVe witness a true chess game. I think this is how chess should be played.
i love too how alpha zero play...an ai who can think faster and better than humans
Yes
same
One possible explanation to the Rf6 vs Rd7 dilemna is that Rf6 is the best move in that position and the Rd7 is the second best move, but after two repetition Rf6 is no longer the best move because would give black just a draw and Rd7 becomes the best move
+Boldi Istvan Yes, and Alpha would rather go for the second best move, than the best move that gives a draw. It wants to win :D
In other words, alpha zero probably evaluated both Rf6 and Rd7 as winning moves first, having Rf6 slightly better then Rd7, so it choosed Rf6, but after repetition Rf6 is no longer winning, but Rd7 still winning so it choosed that one.
Boldi Istvan yeah I think you nailed it.
The win% changed, that’s why Rf6 was ignored the third time. Re7 win% did not changed though, so it became the best move.
In that sense we can say that every engine wants to win :) They are explicitly programmed to choose the move which gives the highest score. I'm pretty sure if that Rd7 were loosing than Alpha Zero would have agreed on a draw.
keep an eye on the rook, okay!!
....
looks away for 2 sec.
damn where is the rook?
+Akshay Gandhi :D
Haha😂😂😂😂😂 best comment ever
Hahahah
Lmao
everytime i see deep destroy stockfish i get this weird sense of joy. like some tough kid standing up for you against the kid who bullies you.
take that stockfish you materialistic son of a bitch.
Deep mind vindicated us. We were right all along human way of playing chess is the right one. We just weren't good enough.
Lmao this is so accurate
The revenge of the neural network! This time in silicone.
+YoutubSosetXui apprently alpha improved his game by 'playing with itself' . maybe we should all play with ourselves more.
+Gareth M haha yes. who is the dumbass now!
Hey Agadmator, while I can't contribute monetarily right now, I love what you are doing, and I hope your heart doesn't explode, keep it up! I mean the videos, and, please! take care of yourself
A nice motivational message is also a contribution :) Thank you! I'll try not to run too fast :D
Love the channel, Agadmator. Question: so your other job is a sit down job, or your CZcams chess job is your sit down job? Or both? If it's both, you need to find a hobby where you run around or walk fast or swim or climb or something. Great humor lately too.
+TAR ICO I was working for the past for years in marketing for a local company. A full time sitting job. For the past 2 months, I'm doing only yoitube, also a full time sitting job :) And thank you :)
I searched for 'greatest chess player of all time' . And Google suggested many names as well as your name. I was really happy after seeing this.
Never thought i'd say this about an engine but alpha zero is fantastic and amazingly different. It doesn't give a shit about some generally accepted laws.
alpha zero improves. maybe they could give it a month of learning instead of 4 hours
No Limits lol seems legit
@No Limits kinds of impossible because you can just sacrifice rook for a pawn
@@Alec____ not clear they got to that point. But its interesting, why does that happen?
@@Alec____ this seems like AI has shown chess is about as solved as it's gonna get. Since it topped out at a level just above the best engines but not by some radical degree. Maybe chess can't be played any better
Okay, am I the only one who sees Deepmind as "sadistic"? I feel like it enjoys beating up the little first grader stockfish.
maybe deepmind is, but that first grader is beating up kindergarteners, so give stockfish a wedgie i say
Burns Firestarter stockfish has been playing for much longer, deep mind trained 4 hours, its more like a genius child destroying an experienced adult
İ mean its playing chess.. its a game whoever plays wants to win.
Throngdorr Mighty where did you get this information?
They are just programs designed to win chess. Any thoughts/feelings are probably just a reflection of your own.
That moment when you realize that probably the greatest AI who plays chess starts game with berlin defense...
one viewer I see know that AlphaZero is the beginning of a new, purer and better race!
Yuris that was a good one
* e v a n ' s g a m b i t*
I'm sure a one could make something of the morphy defense as well
another alpha zero game please..., This is the first time I actually become interested in an engine chess battle, I usually watch humans play.
Agreed. Human chess is so boring now with too many draws. Maybe the future of chess is more and more powerful computers (engines and AI) playing faster and faster games? Each played to a conclusion with no resignations. Wouldn't it be great if after the game the computers gave their reasons for each move?! It would be awesome to analyse a printout of each line it considered before deciding on its move.
Didn’t know I would find somebody who watched chess and hololive
@@Alec____ In chess, they do have reasons, because the way they calculate moves is lines. It'll say "I chose this move because if we both play perfectly for the next 20 moves (listing each move out) I'll be up material"
I for one welcome our new AI Overlord.
Hail Deep Mind!
Bill Wilson It'll spare you and keep you as a pet after it takes over the world for this comment.
Good videos Agadmator !!! you have a new subscriber , Take care of your health !! Simply mind boggling games by alpha zero. Clearly i am planning to continue my chess learning ...
The concept is called progressive/iterative deepening, it's trying to calculate the best move within given time frame (at the end of time limit it has to come up with the best move) but since the moves were repeated, it had more time go do the search, hence the "best" move changed.
am in love with this Alpha Zero thingy
3:54 that’s because Rf6 had a higher win probability than Re7 for AlphaZero. But then the draw possibility changed this value to 0% for Rf6.
Almost certainly this, though also there may be convolutions to do with the move number in the game and how the network reacts to that.
I don't understand how the move number could be a relevant quantity to be convoluted in a neural network?
I think you’re right. Rook to f6 isn’t the same move the second time around. As alpha zero showed, the game was still winnable. Accepting a draw was not the best move.
It's included as input to AlphaZero's neural net according to the recent paper, so it's certainly possible it has an effect. It's difficult to quantify this though.
No, no, no. Absolutely not. Skynet is just mean and likes to taunt its little cousin.
Stockfish: Yo why'd you do that
Alpha Zero: Activity.
Stockfish: You sacrified an entire knight -
Alpha Zero: *Activity.*
Congratulations on going to the gym, keep it up. Bruce Lee would train 10 hours a day. 7 days a week. No joke. You'll need that strength when Deep Mind and his friends decide to attack us pathetic humans.
It's worth noting that Bruce Lee's death was a result of his overtraining.
Bruce Lee died from internal injuries after his fight with Chuck Norris.
Doing that much training does more harm than good it would be better if he trained for 2 hours 3 days a week
Beerdy - Bruce Lee Central I concur.... PREPARE FOR WAAAAAR!!😡😠😡😠😡😠😡😠😡😠😡🤔😡😠😡😎
Bruce Lee died from an allergic reaction to a painkiller.
source- en.wikipedia.org/wiki/Bruce_Lee#Death
What makes the game more interesting is the commentary..
"It's quite a drama"
"This rook is really on a mission to Visit all squares on the board.."
6:01 - "Alpha Zero doesn't mind", yeah Tony, Alpha Zero doesn't mind, but Alpha Zero does Deep Mind. Okay-okay, I'll close the door from the outside... :(
+AGEMO I like it :D
AGEMO 😂😂😂
Lol that was actually brilliant
Like it.
Rook is really on a mission to visit every square on the board. LOL
Great video. Thank you for doing another one I'm loving these.
Thanks for posting more Alpha Zero games.
These are absolutely fantastic, keep the alpha zero vids coming
Love that opening and the channel. Glad it popped up in my youtube recommendations. Used to love chess and find the high level games fascinating.
Congrats on the running improvement! It's awesome to hear little things like that.
Hi, new fan here. I'm loving your channel. It's the best balance I've ever seen of games analysis with kind of tutorial, showing the other variations at various points. Your pace is just right, not too fast nor slow.. Great videos.. Keep it up.. Thanks for your time...
You have got to keep posting these videos with deep mind vs stockfish! It gives me something to watch and take a break from my studies. Thank you and keep up the amazing content I'm a big fan
Love this series! Thank you
congrats on the running man, keep it up! loving the recent content
praising AI and quoting the BORG, the machines have compromised agad :(
+Brandon Nobles We need young Misha, the chosen one. He will lead us against against the machines :)
by the time Misha reaches his 20s, if he doesn't do a Josh and give it up, he should be able to handle this electric Tal
almost 400,000 subs - Congratulations ,love your vids ,doing a good job
I enjoyed this commentary very much ! THANK YOU
Congrats on running! Love to hear it!
I just got done trail running. Restarting after several years and totally feel you there! I was once able to run a 7:15 mile 8 years ago and past-me is now my inspiration!
Awesome video, that was a closer game than I thought!
Love your quote "this rook is on a mission to visit every square of the board" ... great commentary, even if you initially didn't really enjoy the game!
Love how you always start your intro(*looking at your side then** "Hello Everyone"**)
Getting into the gym is always one of the best parts of my days. Thanks for the great videos
I subscribed. Been watching your videos a lot lately, you are definitely my favourite analyst. Please do as many good videos about Alpha Zero games as possible. They are just astonishing to me. Espacially the one where it locked up stockfish´s queen playing white. That was insane.
You are probably the modt genuine youtuber i have ever watched! Nice work on the running
I really enjoy your channel man I never start my morning without watching your videos ! Great work and where is the dog, I did not see in this video ?
+Sergej Krypi He is in the video, you just don't see him because he's sleepig behind me :)
I really like that dog. I have a #suggestion for a game between Kasparov and Nigel Short game 6 when Kasparov loses in a total wining position.
Love the fitness progress! Keep it up my friend.
Your alpha zero game review is being watched by many folks who are following/working in AI, and perhaps that could be the reason why you got lot of traffic from non subscribers.
+Sachin Shenoy No, it's been like that for months now :)
agadmator's Chess Channel Well they should be like me then. Yeah, I am subscribed but I still visit your channel everyday to ensure I have not missed any video. :)
Beautiful lesson and demonstration thank you very much publisher.
Glad to see you feel in shape, and hope you're feeling better from your illness in the last Alpha zero videos
Good job getting some exercise. Glad to hear it. I have been a runner for a long time, and your videos have helped me improve myself by studying the strategy of chess. Thank you. I hope you also can find self improvement through exercise!
Keep these alpha Zero games coming. They are awesome to watch
Scariest computer out there, but magician-like board manners. Love it!!
i have seen a lot of your videos and without subscribing. good thing you mentioned it. im now a subscriber. keep it up
Imagine how many "wrong teachings" chess teachers have said through the years.
Some genius kid probably had something figured out, but was told he was wrong.
Makes me think
true much
Benjamin G, pretty far by order of several magnitudes to be able to check every move.
Relevant scene from "Person of Interest": czcams.com/video/iz5EFsSbQ7U/video.html
They can't check every given move in the given time, the difficulty of calculating deeper is exponential.
I totally disagree with you Benjamin.
Yeah, sorry i wasn’t clear with « in a given time ». I meant every possible move, except for time contraint. I know it’s exponential, but it’s exponential for both humans and computers, the only difference is they compute many orders of magnitudes more moves than us. So they almost never got caught by weird combinations, compared to humans. No human can win in tactics against a computer, they’re just computing too many moves to fall for traps.
LOL no. Just because LeBron James tells you it's best to dunk the ball, doesn't mean you should go for the dunk as well. Why? Because you are not LeBron James and can't even fucking dunk.
Congrats on your running! I went from not exercising in about 5 years as well to running my first 5k recently (took a while). Great feeling indeed!
+James Fleming Thank you and congrats to you as well. I may do it in time as well :)
Dude I just want to say you are a genuine bad ass and I appreciate you
Liking for the running and the chess, much love!
U have 1 more subscription here....I really like ur clear & simple explanation of chess games....take care of ur health & hope to see more of ur videos...thanks
Bro I love how agad feels guilt for asking for his non subscribed viewers to subscribe. Most youtubers have it in their intro. Such a wholesome guy, you can ask for subscribers whenever buddy!
Glad to hear you're becoming healthy and being well.
Keep up the good work and good health.
Thanks for the Great content! :)
Adagmator, your videos are so enternaining that when they are reaching the end I feel as sad as a child when the cartoon is reaching the end! Keep that work, it is fantastic!
I the way you tell us about what’s going on without taking too long. Most CZcams’s take about a minute sometimes two telling us to rate comment subscribe etc. You just tell us how you are and then on with the game. Good stuff
I like the note at the beginning. Congrats man :)
The repeated positions could be an artefact of AZ using MCTS: starting with the superfluous move has all the benefits of directly Rd6 (or wherever it was) plus the occational simulated wins from imprecise play from white. So this search algorithm might be useful for giving the opponent the opportunity to play wrong.
6:58 “And while white does have 4 pawns for the piece, alpha zero will use the Bishop pair like a GOD” 😀😀😎😎
Thanks for sharing about the exercise. It can be a challenge to do for sure.
That was great, sharing that exercise is important! Using your mind all day might keep you from physically moving to stay fit.
Great videos Agadmator :) I watch every video. Enjoy :)
It's great that you're running! Exercising your body instead of your mind 😂 And don't feel bad about asking people to subscribe, you deserve more subs.
10 seconds in, YOU GO Agadmator! Keep up the grind!
White's pawn structure at 6:00 is so visually pleasing, but alpha will surely stomp on the perfect sand sculpture.
I hope you do all the Alpha Zero videos, great games!!
Also I'm a subscriber, look all big youtubers ask at the beginning or end of every video to subscribe. Just do it every video briefly it'll improve your numbers, won't hurt ya (y)
Thank you. You are probably only youtuber I watch who does not ask to subscribe and like after every video(sorry if I forgot someone who does not do that too)
thank you for the effort you put into your videos …
youre a great man , thank you for your videos
Congrats on the run @agadmator !!!
Congrats on the running! It can be hars to find the motivation to go to the gym, but I have never regreted going to one ever.
But how many chair spins can you do in a minute though?
Man I got so dizzy after 10 seconds. I was spinning like a fidget spinner
Thanks for the encouragement on the health man, I need to start running or doing some sort of cardio.
It makes sense to test if the opponent will find the best move if you don't lose anything from it. Whenever there is a move that can only be "refuted" with a repetition, if you play it then the opponent will either make the best response that is going for the repetition sequence, but in that case you have the chance to get back on track and not repeat and play a different move, or the opponent may not play the best move and give you an advantage. It could be possible that, during its self training, alpha zero developed a strategy of going into 3 fold repetitions and giving up if the opponent decides to play the repetition sequence. I can see that strategy being useful in the early times of alpha zero training, when errors in evaluations could lead to miss-evaluating the position
Happy that you are starting to get progress with exercise! It will only continue!! 🤗
I love your videos, you have one extra sub
Using the bishop pair like a god. Loved that one! Keep healthy! :)
I think we can assume that rook to F6 was considered the best move at the time according to Alphazero, but then when it would only lead to a draw it was changed to E7, which at first was considered to be the second best move.
About the decision to draw or not: It depends completely on how wins and draws have been "weighted" while the evolution of the neural network has taken place. If AlphaZero has been trained to think a draw is 50% as good as a win (which would be logical), then it must simply think its chances to win is higher than 50% regardless of whether the rook is on F6 or E7.
The real question is how it makes that evaluation. Does it recognize its opponent making poor decisions, and give a higher evaluation to its chances to win than if it would be playing against someone who didn't make mistakes? Perhaps that's an important "evolutionary" part in performing as well as possible, when facing versions of itself which are obviously sub-par throughout evolution?
It would be very interesting to hear someone who programmed the neural network talk about it.
It study's other games people have played to understand slightly better why certain moves are played... It keeps all of those in mind when it evaluates each move.
Wonderful AI game, indeed.
[Agadmator, you are just excellent in this chess endeavor.]
Awesome review!
You were so cute with the subscribe thing at the end! Subscribed :)
Great to hear you're going to the gym and being healthy :D The best thing you can do for your health is to make it a habit and do it consistently. I take care of my health by going on a morning walk every morning, and by doing stuff like pushups, kettlebell swings and curls every monday, wednesday and friday. Habits are really important!
omg I love Deep mind please keep making these
Man, For some reason I’m so hyped with this alphazero games
I subscribed because you mentioned it🙂. Nice work
"This rook is on a mission to visit every square on the board" - nice commentry 👍👌
Nice attacking game ..equally good explanation! :)
dude your fitness message hit me hard haha. you inspired me
@agadmator it's one thing to play a sub-optimal move hoping your opponent chooses from one of many likely sub-optimal moves and misses the optimal move--I'd argue this makes no sense to do--but totally different to take a line where the opponent can choose between 1) a sub-optimal move and 2) a draw offer. There is no cost to check if they're willing to take a sub-optimal move. I never thought of this before, thanks for bringing this up! I also never saw any similarity between poker and chess before, since there are no tricks in chess. Only vision. But now I kind of see what you mean about the "bluff"! Nice!
I think the rule about moving pieces more than once is a little like thinking about consecutive probabilities, if you calculate the chances of rolling 4 6s in a row for example you would simply multiply 1/6 three times and get the correct probability. But if you roll a 6 the chance of rolling another 6 is always 1/6. So in the same manner it could be true that the probability of moving the same piece several times being good is small, enough for a human to correctly make a principle against it, each move in itself still has the same chance of being good and Deep Mind is just inhuman enough to see the difference and look at each move in itself. Therefore it does not exactly come to the conclusion that moving a piece several times in a row is the best strategy, but rather for each position it concludes that moving the (same) piece is the best move, and does so several times in a row.
I'm impressed. Most people think that past outcomes affect future outcomes. (with regard to random events).
I hope that after your surgery you do more exercise, like here.
These games are really interesting. Would really like to see more of this content.
Even I'll start exercising now
Introvertedpawn 99 me too
re your question on the rook/draw decline. imo - A0 did this because the rules say it can - simple as that. i.e. it knows it can do this, and gain information from opponent (in this case their willingness to accept draw) before deciding on whether to 3 fold repeat. it was just exercising it's free option.
We better all be very nice to Alpha Zero... it will soon be our Robot Overlord.
Awesome job on the run today, keep that heart healthy so I can keep watching your videos!