HIKARU TRICKED THE STRONGEST CHESS ENGINE: HOW???
ฝัง
- เผยแพร่เมื่อ 20 มี.ค. 2024
- An epic story about @GMHikaru Nakamura tricking a 3200-rated chess engine and winning a game against it in almost 300 moves. If you've never seen this game, your mind will be blown.
My podcast with Fabiano Caruana: • FABIANO CARUANA: "I co...
My podcast with Nepo: • NEPO on cheating, Magn...
My podcast with Levon Aronian: • LEVON ARONIAN: chess p...
My Twitter: / mustreader
My Twitch: / gregmustreader
Collabs: mazdrid@gmail.com
#hikarunakamura #hikaru #chess
---
My name is Greg Mustreader, I am a Candidate Master in chess, an entrepreneur and host of several podcasts.
In this channel, I interview top chess players (my guests include Nepo, Fabi, Aronian, Dubov and many others - the unreleased podcasts will be published in January-February), play against chess AIs, analyze the most interesting chess games and combinations of today and of the past, as well as show you my path in chess and the way I work on my chess improvement (including recaps from the tournaments I play).
I live in Almaty, Kazakhstan and travel the world extensively (been to 50 countries). I'm 30 years old, I have a girlfriend, two dogs and two cats. I speak English, Russian, German, and a bit of French and Chinese.
Don't forget to subscribe!
My podcast with Fabiano Caruana: th-cam.com/video/1LMPYFf1iLo/w-d-xo.htmlsi=uCudy...
My podcast with Nepo: th-cam.com/video/sW-i4CPCa_w/w-d-xo.htmlsi=AbSRc...
My podcast with Levon Aronian: th-cam.com/video/THyL4D2Clbw/w-d-xo.htmlsi=F_RHV...
My Twitter: twitter.com/mustreader
My Twitch: twitch.tv/gregmustreader
Collabs: mazdrid@gmail.com
Rybka means little fish in polish so it's kinda funny that now we have stockfish
"He must have cheated"
Bro he was facing the cheat source
This is Kramnik speaking🤣
Surely if you're facing the source of the cheating, it's not cheating, even if you cheat?
@@AURON2401 You missed the point so hard that I had to caught it for you lmao. He meant to say that if he had cheated by using the program (which runs using the exact same algorithm), he wouldn't made it. Because it was programmed to find the "most efficient move to win" instead of the most logical one.
@@monke4044there is truth to what he is saying tho', since the engine can and has beaten itself when playing both sides.
Not saying that's the case here, but It is possible.
“Thou shalt not make a machine in the likeness of a human mind.”
- Butlerian Jihad
So basically hikaru forced a completely closed position, then sacrificed material 2 times because he knew the engine is programmed to go for a win even if it's in time trouble & the engine does not understand rooks are pretty useless in completely closed positions and large pawn chains.
Yes, that's quite an accurate summary!
Even humans will think they are winning if they have two rooks against two minor pieces
@@charlesa1234 Human 'might think' like that however they are not programmed to think like that. The ai will always prefer the rooks but humans could think otherwise
@@wallnut7624It already happened for an AI to not let their bishop get traded for a rook, AIs can very much decide that their bishop is better than a rook, they just don’t understand completely closed positions
Closed positions are really hard for computer yeah, mainly because the gamestate is position based and not move based.
So it needs to go by experience which the Ai lacks.
Hikaru clearly saw mate in 250 despite that closed position.
Thanks for the laugh mate hahaha
Yeah of course, its from move 27, it was forced mate in 250, there WAS amother in 357 but, ohhhh well
I’m surprised he didn’t premove it.
It's just takes, then takes, then if takes then take, take take, shuffle a bit... then takes, takes takes...
Cringe
Hikaru took Rypka into a deep, dark forest. haha
I didnt know you like chess as well
@@kinggloxinia5091 eh, it's ok. I'm not good at it.
Where a little fish is not comfortable.
Where Mr hoodieguy waits for his finisher
The engine is called Rybka which means little fish in Slavic languages
Hikaru exorcised the demons of A.I. chess with the help of 6 bishops.
LOL
5 bishops. ;)
@@Alexxx95_HOne was lost to Devil's devious deceptions.
I didn't know Elton John played chess. Thanks for the video.
I'll take it as a compliment! xD
@@MustreaderChess Fwiw I think you look cool as heck. Love your shirt/top.
Great video! I remember hearing Hikaru say that you more-or-less deal with cheaters this way too, close the position, play for time. If only it could be as effective as it was in this game all the time.
@@mrjoe5292 Thanks! That’s an interesting point about cheaters!
Oh, now I see it.
The glasses look like Elton John but everything else reminded me so much of Ferris Bueller (Mathew Broderick).
"You see, killbots have a preset kill limit. Knowing their weakness, I sent wave after wave of my own men at them until they reached their limit and shut down."
Aah, glad to see people still remember the great Zap Brannigan!
Where is that from
@@onatkorucu842 blasphemy sir! :) It's from Futurama
@@onatkorucu842 Those are the words of Zap Brannigan, describing how he became a legend. From the gem that is Futurama.
hikaru farming best moves💀
Imagine if he stalemated with all those bishops...
It would have happened to me 😂
Few people know that if you promote 8 Bishops while not losing the starting two, that's enough to hold a Papal election.
🤣🤣🤣 totally underrated comment!
This was not a defeat...it's annihilation
The bishops at the end was just disrespectful
@@Osmotic I bet Rubka's fans sounded like a jet engines, because he was rage heating so hard.
I remember watching this live on icc,,,,was epic!!!
I can imagine!
Has he put it un a video as yet?
Thanks Eastern European Elton John!
*drops LSD*
*gets LSD sunglasses*
LOL
@@MustreaderChess I got 150ug dr seuss d.s.3s
Lore accurate humans.
You might be strong, or run fast, but in a game of endurance, we are unmatched.
I've seen this game before and I have always been impressed for both sides (in one way or another) multiple years later I don't remember too much about the game, but a human beating a chess engine is +1 for humanity
Fun fact:
Rybka in polish means fish and maybe rybka is stok fishe's beta version
Yes, it's interesting! In Russian, it means the same, BTW
Isn't rybka like a small fish ? A fishlet, if you will ? With "ryba" being "fish" and -ka being cutesy/hypocoristic
It is!@@paulbarbat1926
My chess machine, back in the 1980s, won the first (and only) computer correspondence chess championship. Its name was Piranha. 6502 48K RAM, 1 140k floppy drive, used Monte Carlo to play many thousands of full games while playing (deep analysis ca 60-200 games per move). It also had a contempt factor, so Piranha bad sacced a pawn. Opponent was cheating (human moves) and tried to win with the horizon effect (which would have beaten any stock computer chess machine of the time) but since my bot played full games with a bit of Hans Kmoch (pawn structure database) it refused to gobble a "free" pawn, which would give the cheater a passed pawn (it would have been protected). Closed position, single bishop (his) vs single monster central knight (Piranha). My Apple 2 died shortly after that, and all the code was on thermal paper. The computer magazine that sponsored the contest went out of business later in the 1980s. Piranha means fish, very hungry.
Wow! An impressive story! Where can I find the games? Can you email them to mazdrid@gmail.com?@@Galahad54
It’s been known to happen. In the game of GO, Lee Sedol took a game from the AI Alpha Go (great documentary BTW)…according to the AlphaGo programmers he basically took the computer into a deep hole and it became confused.
Also many years ago Marion Tinsley was the only human to beat the Best Checkers program (AFTER, it was fine tuned). Checkers has since been solved.
I've heard about Tinsley. His life story is very impressive!
The top go programs all got beat by an amateur after Lee Sedol match (and a 60 game win streaks vs pros) exploiting a flaw in their understanding of the game
The only frustrating thing is that we didn't see a 7 bishop checkmate. That would have been instructive.
This is true. I don't know how to do a 7 bishop checkmate either. Every time I'm in that position, I stick 6 of the bishops in my butt and end up stalemating with the last bishop. But what other moves are there?
LOL
In summary: constipate the position, run the opponent's clock down to handicap its search function, wait for a mistake, pounce, checkmate with an army of bishops.
hAHAHAHA "everyday I'm shoffeling" hahaha that accent with that line....
"in the strictest sense, I did not win... i busted him up."
lt. commander data, tng peak performance
Even it was rapid game but still to win against rybyka is phenomenal achievement. Computer programs are merciless. You need some cheecky idea like premove and hope it suceed to beat computer engines.
Has anyone mentioned yet that rybka means fish in several Eastern European languages?
Yep!
This is a good example of the horizon effect, which is the only weakness of the engine. It gains two exchanges, sees 16 moves ahead and assesses it has a nominal advantage of 3 points. It's happy enough for 200 moves, then loses patience.
that does not happen any more. Best chess engines evaluate by winning probability
@@chrisvolk1762 It can still happen somewhat, but neural networks make it a lot less likely. The issue is these older engines were basically just brute force calculators. Modern engines are usually hybrid engines that also use machine learning as well as brute force calculations, which fixes a lot of the weaknesses they previously had.
@@eragon78 they still use a k-step look ahead. At least alphazero did. Just dont look st all possible combinations
@@fangiscool1 Yes, no matter what algorithm you use, its always going to have a k-step look ahead. This is inherent to how chess works, even humans do a similar thing.
The difference, and advantage humans and Machine Learning AI have over brute force AI is that they have a better positional evaluation.
They can understand specific structures in chess a lot more, and understand whether or not those structures are actually an advantage or not. They can also check prime moves to see if the structure ever significantly changes or not in the near future, which can change the evaluation. even with K-step limitations, a strong machine learning AI can see that a structure may not have changed much in 20+ moves, and realize that the position is probably more likely to draw if it cant force any significant change in it's advantage. Humans can do the same thing, but even better, even if our ability to look ahead is much lower, our positional understanding is better still.
And then brute force AI cant really realize that. It's positional analysis is really poor, usually just some sum of points based on very simple things like material advantage, and maybe only a few hand written specific values for positional advantage. (Something like a passed pawn being worth more points than a blockaded pawn as an example). But outside of very simple metrics, it's evaluation of each position is generally very poor, it just gets over that with brute forcing an insane amount of positions. But since it's ability to analyze each position is really poor, it can get tripped up in closed positions where it cant brute force enough positions to get out of the closed position. It just sees that in 30+ moves from now, it may still have a material advantage, so it thinks it's winning when its not.
But Machine Learning AI has a structural neural network which somewhat encodes positional evaluations into it's algorithm. This means it is SIGNFICANTLY stronger in single positional analysis. And it can see that in 30+ moves, the position is pretty much still equal with no real breakthroughs. This makes it give much more even results for those closed positions, and it wont just throw away the game to break out of a closed position unless it think's the position is still pretty even afterwards, which it is much better at evaluating.
Of course, these Machine Learning AI still are not nearly as strong as humans in single positional evaluation. Humans are by far the best at evaluating a single position intuitively without any thinking ahead or move considerations. But since machine learning systems are still hybrids, they can still brute force many many many orders of magnitude more positions than a human to largely make up that difference.
But a strong enough human can still play very anti-engine chess and get draws using similar methods of forcing closed positions which the Engine may be too quick to agree to in exchange for a minor advantage it can never truly capitalize on.
But beating a modern strong engine using a method like this is significantly less likely. Especially one on proper hardware in a classical time format where it has plenty of time to calculate.
There are still some positions modern engines are bad at calculating, but its usually not enough of an error for a human to be able to beat them, at least not in any "traditional" game of chess. You CAN make positions though that an engine THINKS is winning for the engine, but then beat the engine, but these require really extreme positions that arent possible in a normal game of chess.
5:55 "programming says that you should win at all costs"
This is not true. It has just innacurrate evaluation in closed positions and sometimes pushes for non-existent advantage.
It was true in case of Rybka. It had built in instructions that it should avoid draws with lower rated opponents - and you can see this very well in the movie. It run exactly 49 moves without moving a pawn or taking anything , and than, becouse next move like that would be a draw with lower rated opponent - it's evaluation of best move gets overrided by build in instruction to do anything that can prolong the game, even if it's at the cost of loosing material and worsening it's own position. So it's giving up a pawn for no reason, just to not get a draw announced. it did not had anything to do with it's evaluating algorithm - this one was fine (of course way weaker than this of the current top engines). Problem was exactly this instruction that should never exist. Under no condition engine should do a move that it's own evaluating alghorytm sees as worsening the position.
@@kkmiroslaw " It had built in instructions that it should avoid draws with lower rated opponents" I doubt that it had "instructions", but if you have some official confirmation about this please share it.
The only thing can be used in engines for this is the contempt factor which is the value of the draw itself which is usually 0 . It does not know the rating of the opponent, a chess engine just trying to find the best move in a given time assuming perfect play. They might set this to minus anything but this is risky.
@@kkmiroslaw also the big mistake of rybka does not come from anything like this, it is in timetrouble and basically has no time to do any deep analyses.
@@mgoogyi I guess we don't know how it evaluated the position internally, but just on piece score if it was at +4, then a 50th move which resulted in a draw would have seemed worse than giving up a pawn that resulted in a piece score of +3. So without the ability to see far enough into the game to understand that the +3 was really just drawn as well then it would obviously favour giving up the pawn to keep the game going when it thinks it has an advantage.
@@mattc3581 It would definetly give up pawns until it sees that's better than a draw. It had 2 rooks vs 2 minor pieces and it was way before neural network based engines so it is likely does not see that's a dead draw.
It does not lost because of this whatever was the contempt value. It lost because of time trouble and most engines can't handle it properly. (My hobby was actually chess engine programming for around 8 years and all the engines are quite similar in main functionality.)
In style, loved it!
Awesome. Most the time when i watch matches of high ranking players i dont learn much. But i actually learned a lot from your video. Actually moving the pieces to show why a rook isnt good in a closed position is much better than just saying the words witch is where i think i get lost most of the time. Visualizing things really helps me understand it.
Glad that you liked it!
Really nice video, I like the way you explain the moves and the nature of the position!
I'm glad you liked it! Check out my other videos!
Great game, thanks for sharing. 👍
fantastic video, earned a sub, AND am listening to your podcast now with Fab
The engine literally played the worst first move which is ofc still almost impossible
Disrespect Speedrun by Rybka xD
In the end, Hikaru was just bullying the AI.
Thanks for this!
Hikaru made Rypka his "Bish"...op!
First time hearing about this game. Interesting that time trouble was the computer's weakness. Thanks so much! Is this the last time a human defeated a chess engine?
The greatest weakness (or, rather, bug in the system) seems to be its lack of objectivity in a dead drawn position. Time pressure probably just made it worse.
As for other cases of humans defeating engines on equal terms, I’ve heard only about such cases in hyperbullet (15 sec games)
But why on super low time, we have change? I mean almost no one can play 15 sec game. I barely play one min with just move piece
@@dimitriskontoleon6787 The big advantage these old engines had over humans was that they could crunch millions of positions per second. But the game tree is huge if you evaluate every legal combination of moves. So even the computer can run into time trouble if it searches only "randomly". Rybka used of course methods to only search parts of the game tree that look promising, but this is much more sophisticated in newer engines.
"Artificial Intelligence" is neither artificial nor is it intelligence. Richard Feymann says here on TH-cam when asked about A.I. back in 1978 that "An airplane can fly but it is not a bird."
th-cam.com/video/ipRvjS7q1DI/w-d-xo.html
Wonderful! And a very enjoyable analysis.
Rybka: I am the strongest chess computer in the universe
Hikaru Nakamura: (Closes position) Nanomachines, son! You can't hurt me, Jack!
Ver7 cool - thank you for this review
Absolutely unbelievable movie, beautiful story! Thx.
Hikaru is so slick. I remember him beating Magnus with all premoves.
great commentary, no drag and straight to the point
Whoah ! This was a clever game !
the limitations of AI are in stark display here. it can't think yet. didn't know about this game. thanks for covering it.
However, this was before the modern-day neural networks emerged, this has changed everything (in chess as well as other aspects of life)
@@MustreaderChess has anyone since 2008 beaten or perhaps drawed a modern chess AI (within the last three years for example)? or was 2008 the last time this happened?
@@MustreaderChess They still not thinking or reading. They just better at the brute force massive calculations and pattern solving after all a lot of this copied by how this process done in nature by various animals. But it stuff done by animals that can't beat the higher end predators thinking.
Computers not beating Poker or Bridge yet. In particular bluffing hard to deal with as good players will change their bluffing pattern. They are dang good at both but the top players own them.
We need Hikaru to be the Army strategy leader in the coming post Ai apocalypse era.
That checkmate was legendary
Brilliant game! amazing! 😲
Very interesting video. I learned the power of idea of pawn break.
It would have been interesting if there was two eval bars, one from Stockfish and the other from Rybka.
Seems to me like the human would settle for a draw and the program would not, or the human new the prime directive of the program and used it against itself, which is one of Sun Tzu’ Art of war tactics; using the opponents desire against itself. The human decided to turtle, giving the program a lot of choices, but none of them were good choices. Because the human new the program would have to follow its prime directive, the program would have to take the first sub optimal move in which against a good defense, it would never get its turn advantage back, but because if the prime directive, it kept trying to take the advantage back, but the only choices it was given were bad trades. In the animal kingdom, this is like getting constricted by a Boa/Python. Once it coils you, you can’t win, but the moment you exert energy to try to get out, instead of losing slowly, you just loose faster.
When my opponent refuses to find the resign button, I usually do the same crazy stuff but with knights. I mate the King in the corner with a row of knights on the 3rd or 6th row/column and I put all my other pieces on the 2nd or 7th row/column.
Great to see hikaru improve since then now he checkmates with 7 knights XD
fantastic!!
As Chris Smoove would say “No sportsmanship for the Hall of Fame CPU” 🤣🤣 god tier level of disrespect on the chess bot
I used to get draws against droidfish with locking the pawns, triple defending everything and shuffling the king. 200 move games by fifty move rule. If the computer elo 2900 got a pawnbreak it was over. But theyll take the space and lock it up sometimes.
Ooh know your opponent! Very nice
in any situation where you face a vastly superior opponent you often have to play the opponent against himself
This is the funniest chess video I’ve ever seen. Hikaro is the GOAT
He may have exploited a few bugs in the pcs programming but he has done that many times agains humans.
Humans also have their own bugs! xD
Yep it how you beat human look for their bugs in play.
This guy is a legend! 😲
nice video! Great story telling!
Thanks! Glad you enjoyed it, check out my other videos!
Brilliant !!
Bro hikaru isn't satisfied trolling with human with a titles, he even humiliated the a.i
This is not a chess engine bug. Even professional chess players will try and go for the win if they have two rooks against two minor pieces
Yes, but in this case, any human pro will know it's a dead draw
@@MustreaderChess yes but atleast they will try to win considering you are playing against a player significantly weaker than you
Me unplugging the power source, wins by timeout
That whole outfit is awesome
Thanks!
Engines have no feelings, they dont feel humiliated.
That's awesome!😀
No. AI won't ever be conscious. The mind isn't reduced to mechanics.
war against the machines have started and chess is first battleground
Sheer brilliance for Hikaru. He knew the engine would not be able to resist capturing a lower piece even though the position was closed. After all two rooks vs a knight and bishop means a superior position right?
No, you don't know what "bug" means. A bug is a fault not a weakness.
This is why I love Hikaru. Chess should be fun, and Hikaru keeps it fun. I hope he wins the Candidates this year.
😥
Pretty cool. Hikaru is awesome
Absolutely beautiful!
Hikaru has used another engine.
That's awesome!
such a cool checkmate!!!
Your my wonderwall....
Kramnik would find this very interesting…
Wow Hikaru is stupid; he could of just pressed the power button.
It was experiment with Grob's opening to induce the closure of position as soon as possible Strong engine will not choose this one, as it's extremely strong weakness in Indian game d3-d4 for the whites, without further facility to sustain a pressure on critical black square f4.
Ever seen 6 bishop’s?
IT IS LIKE CATHOLIC PARTI
What happens if you try to do this against stockfish now?
I swear you look like a batman villain
It took me 6 nights to watch this video because I fell asleep each time.
I like his attitude. (dont jus win, win in style)
Those old superhuman engines were so weak though.
Hikura use which engine
In 2008 rypka elo rating was 3238 impressive 👏
bro made a disrespect speedrun against a bot
Very interesting, Hikaru's ending was hilarious - and the comment - a catholic party! lol
Awesome. That was an inquisition lol
So in total only magnus can help you against hikaru
When this game was played, Hikaru had played around 500 blitz games against Rybka and Crafty, the top engines at that time, and scored 5 points. In other words, his win rate was about 1%. The fact that he managed to win this one isn't exactly a testament to humanity's superiority over machines. Anyway, Rybka has since been tweaked to avoid the errors it made in this game.
Go ahead and gloat, humans. One game that supposedly makes an older version of the engine look stupid proves nothing.
Absolutely goga Chad hikaru
I'd like to see someone beat an engine in a *real* games of chess (not speed) where the engine does not play one of the considerably worst openings in the game (Grob)
Its impossible. Happenend thousands times and why do you want that? You can practic against stockfish always you want but you will feel bullyed
@@flacsomtodosclas2165 skill issue
The ESET anti-virus robot can play chess? Was not aware of that. What edition do you have installed?
LOL