I'm about 1850 in Lichess blitz, and it played at the level of an earnest 12 year old, which is actually meant as a compliment. It got crazy with its kingside pawns in the opening (no book knowledge), lost material pretty steadily and was checkmated down a queen, bishop and two pawns by move 25.
But it developed some pieces sensibly, made obvious recaptures and was not reckless about king safety. It avoided pointless toggling moves and suicidal piece jettisoning.
You've earned the right to be a proud papa on this one.
If you play a defensive setup like push D, E pawns one step ahead, fianchetto both the bishops, try to castle early, the computer just flings all the pawns forward, even from the king side: https://imgur.com/a/Nt4x7TP
It’s a pretty decent model for the size. If I played like I’ve never played before it beats me pretty well, if I play defensive at all it gives me pieces until I checkmate.
I agree. I'm at a similar chess rating level, I would add that it is very weak in the endgame. I guess if it only looks four moves into the future it's difficult to see pawn promotions.
I'm of similar rating on Lichess, and agree with the above comments. To be honest, any difficulties I encountered were due to unfamiliarity with the look of the pieces.
Still, it was fun.
I joined my girlfriend to a chess tournament of the local club once. Old vs. young. They were an adult short, and I know the rules, so I volunteered to play - otherwise there would be a kid without an opponent every round (my plan had been to write some code in a corner, but entertaining the kids, why not right?).
I lost every game. The kids were something like 10 years old. Admittedly, I know little more than the rules, but I expected to at least give one or two of them trouble by playing logical. Not so...
They had the best time, seeing in how many moves they could beat me. In my last game I tried an unusual (okay, stupid) opening hoping to at least break any pattern they have, but that just made me lose even faster. Be weary of 12-year-olds if they're playing in a chess club!
What op probably meant was that the engine knows no theory, but can See when something can be taken and can think a few steps ahead.
Not sure if I agree completly, but since you can win in under 10 moves simply focusing on an attack in the king I know what he means (wasn't quite a fools mate, but couldn't protect the Kings side)
I think that a 12 year old would play Mord reckless and would make more obvious mistakes. I think this is at the level oft a self taught novice that only played against another self taught novice (funny enough I was thinking of when I player chess against my friends when I was 12...) good times
I'm getting a kick out of this thread because, having hardly played chess since age 12, this game instantly reminded me of games in the middle school library.
It seems like there's broad agreement on skill level.
Yeah, just tried a fool's mate! It didn't /quite/ work, but pressing on still led to a very early mate, after dealing with an inconvenient pawn. (I think about 15 moves total?)
I settled for the second trick newcomers learn -- forking the Queen and the Rook with the Knight. Easy win being up a Rook and Pawn to a Knight, with the black's King uncastled and in the open.
I expected it to play like Ed's Chess on DOS, which I always had trouble beating. IIRC that engine looks 3-4 moves ahead usually, but 6-7 moves ahead when it needs to. No idea how big the equivalent JS would be.
Well the deeper you look, the more clever you need to be about pruning your branches and exploring the search space, so I would guess it does require more code.
Was extremely good at 12. But my dad who thought himself good stopped playing after I won a game by taking everything and losing none. I recall seeking so many possibilities, now I stink at it.
Sharing other people's amazement that such a tiny amount of code plays so well (I'm an inexperienced, infrequent player and it beat me once and drew once), I was pondering that there's no absolute reason to think that small programs inherently must play games poorly. For many abstract strategy games, you could write an explicit game tree search in a relatively small amount of code and it would, in principle, play perfectly if you somehow gave it enough time to search the whole game tree. (That wouldn't be feasible for most abstract strategy games that human beings like to play, but would for small games like Tic-Tac-Toe, Hexapawn or Octapawn, Nine Men's Morris, maybe Hex on a small board.)
In GOFAI (https://en.wikipedia.org/wiki/Symbolic_artificial_intelligen...) there's a clear pattern where "intelligence" in abstract strategy games represents, in part, being able to prune game trees well without having to search them exhaustively. In some cases that means having a good evaluation function for positions, and in other cases having a cleverer search strategy in another way.
Thinking about this made me realize that it's not just impressive that this program plays decently while being tiny, but especially that it plays decently and quickly while being tiny.
I've heard of the Kolmogorov complexity connection, but I'm not sure I've seen it spelled out as explicitly as this before. Definitely very interesting!
>Thinking about this made me realize that it's not just impressive that this program plays decently while being tiny, but especially that it plays decently and quickly while being tiny.
Yeah I wonder if there are subfields of chess programing that optimize for that, something like "quality of engine per unit code size".
“We describe the combination of two search methods used to solve Nine Men’s Morris. An improved retrograde analysis algorithm computed endgame databases comprising about 10¹⁰ states. An 18-ply alpha-beta search then used these databases to prove that the value of the initial position is a draw”
So they built a database containing all all 7,673,759,269 possible positions in the endgame phase (i.e. after all stones are on the board), and then did a full-depth search for the “place stones” phase.
Paper is from 1996, so one might think current hardware could do a full search, but they show (page 10/110) a position with “White to move and win. Mill closure in 187 plies.”
So, I don’t think this game is small enough to search the full tree. They made it small by first creating an endgame database with 7,673,759,269 endgames, but that database requires over a gigabyte of data (“The hash function we decided to use maps the 7,673,759,269 states into a range of 9,074,932,579 indices.”)
Thanks for the information! That game is definitely bigger than I thought. (I knew it was solved, but I'd wrongly imagined it was solved by complete game tree search.)
Speaking of rules : I noticed the AI was able to castle even though one of the square in the path was attacked by one of my piece. I'm not a chess expert at all (this AI seems much better than me ^^) but I thought it shouldn't be possible.
I wouldn't be surprised if it doesn't cover that, I can imagine it'd be tricky to include without adding tens of bytes (10-20 bytes could be freed up by using slightly hackier function calls to/from the display, so perhaps Óscar will add more features/intelligence one day)
Was it a long castle? In that case, the squares the rook moves through don't block the move, it's only the squares the king moves through that block the castling move, so f8 and g8 for short castle and d8 and c8 for long castle.
In my experience having small webapps, it hardly helps. I literally added a huge modal popup on my feedback button specifically saying "DO NOT MESSAGE ME ABOUT X" before continuing and I still get dozens a week.
Even on reddit, when you get a high ranked comment, you will often get the same reply over and over again, even though a similar response is already the highest ranked reply to yours. There are people who just want to blurt out their opinion regardless of whether it's original or not..
Reddit sadly went the way of mostly write-only microblogging a long time ago now. I'm not sure what the solution is between invite only subs or heavy and time consuming moderation, and random comments that are just people thinking out loud, as using any sort of karma metrics would be useless when you can get 20-30k off a no effort post in a popular sub, but I also doubt that reddit the company spends much time worrying about it.
I can't remember who I stole "the internet is write only" but it's felt very true as audience size exploded.
Perhaps the worst part is that it's self reinforcing in either direction, and few people ever leave the agora unless they delve into niche subjects, so they bring the agora mentality with them.
I think this is the key point. As with most issues on the internet, it's mostly an issue of scale. I assume these problematic users are actually only a very small fraction overall, but as the overall size grows, so does the number of people who belong at the very edge of the bell curve in terms of "normal" behavior.
With scale, any rare issue becomes common enough to be annoying.
I noticed chess.com started adding a little pop-up notification about en passant when you use it. I imagine with the huge chess surge going on right now they were getting quite a lot of complaints. Stalemates also prompt the little pop-up notification.
lol I remember when I was little in the 80s, pre-internet, and I played Chessmaster 2000, and I was like, wtf, what just happened? when I first encountered the en passant rule.
I'm a novice player but I reached a stalemate in my first game. I am not a big fan of the graphics, they're really really cool but I made at least two big mistakes because I misread the pieces - and now I have a headache :D
Ok, maybe if you have fond memories of playing CGA games... for everyone else, those colors (especially the cyan/magenta palette) are just a garish eyesore. But I guess the engine needs every advantage it can get ;)
It's common to draw too, as it tends to do 3 move repetition, perhaps to avoid losing material.
When I finally beat it, it seemed to let me fork it's pieces in the corner, and then in endgame, it knows nothing about the positional strategy to avoid forks or mating combinations, since it only has 4 move lookahead. It let me material with king and rook even though it had a knight that could have interfered.
Don't get me wrong, I dig the retro art style, but at that resolution, dithering gets in the way of legibility. I think for the chess pieces themselves it makes more sense to trade off some of the realism (i.e., shading, specular) to better communicate the shapes, particularly as it's so important to game play
I'm aware of that composite signals tended to bleed pixels horizontally. I used to draw my own pixel art when I owned an Amiga (though to be fair, the monitors were often sharper, and I wasn't always using composite). I grant perhaps it's more legible on a composite display, though I think that's a hard thing to appreciate when delivered on the modern web.
Wow it took me 40 or so attempts to beat it. I’m a chess novice but you’d think a human novice should be able to beat a small chunk of JavaScript.
This code must be pretty compact even if expanded to readable form? Is there a repo anywhere with this type of chess engine explained and expanded to human readable commented code?
Thanks, I hate it. I'm losing every single game (I think I tried like about 30-ish times). If that's how 12-years old are playing them I should probably apply for disability benefits. My naive approach by trying to define safe squares to move pieces to and naive capture goals by exchanging figures of lesser value is not working against this algorithm, I guess I'll probably have better chance if I'd just run the same logic on paper.
You are using something like the same logic as the algorithm itself and describing how a newcomer to the game would play chess. I think this is what they mean by "12 year old". In fact the algorithm probably beats this strategy as your approach probably doesn't identify 2 move 'tactics'.
If you fancy learning more about the game take a look at tactics, and maybe try a tactics trainer. The next level of play is about understanding things like forks and discovered attacks where you aren't assuming that the other player is just going to make huge blunders that give you pieces.
It took me 4 tries to beat it, and I have a rating of 1300 on lichess.org and 1100 on chess.com.
In an effort to guestimate its strength a little more accurately, I had it play the first 4 levels of bots on lichess.org. It comfortably beat the level 3 (1400) bot and got crushed by the level 4 (1700) bot. Combined with beating me 3-1, I would estimate its lichess equivalent Elo rating to be somewhere in the 1400-1500 range.
Oh I didn't realize they depended on the device they were running on, that's too bad :(
I can reliably beat the 1100 bot and lose to the 1400 bot so they at least seem to be in the ballpark of matching my rating on the site for my hardware though.
No idea what my rating might be. I haven't played in years. I creamed it on the first try. It seems pretty aggressive, but doesn't seem to "think" things through. Pinned his king and forced him to block with his queen after a dozen or so moves, which only delayed the mate by one move. He should have resigned.
Glad I'm not the only one. My vision isn't bad but the way the black pieces are shaded and shaped makes me have to put conscious effort into identifying them. (Not that I'd be able to beat this thing anyway)
Yup, my first try I made two major blunders due to confusing knights and bishops. Beat it easily on my second try. (When my allergies aren't acting up I'm 1600 on Gameknot.)
I consider myself at a ~beginner+ level. Beat it on my second attempt. IMO this is the right amount of difficulty without being overwhelming for a beginner(or novice), which TBH is the majority the world.
This is a great program that I could use as a litmus test for checking if my newbie friends of chess has made satisfactory progress without me playing with a handicap or artificially lowering my gameplay. This program appears to be in the delightful zone for the casual player. Bravo!
It was great to play against the AI this weekend, rekindled some of the love I have for chess. I know the rules but I'm a very novice player. Took me 6 tries to checkmate. I cheered loudly. The style is very neat as well!
It looks 4 steps ahead with a points system for pieces & positions, but has no sense of broader strategy.
(The original code looked ahead 3 moves, but browsers & devices have improved a lot in the last 11 years)
At it was expected the program is very very weak. The elo should be around 1100-1200. Based on my game it doesn't see more than 1-2 semimoves ahead and doesn't have an opening repertoire. The evaluation function is weak because the program starts the opening by moving lateral pawns and neglecting good piece development and central control
That doesn't answer the question. Yes, it's random, but why? Is it just to avoid repeating move sequences so you can't replay the same game multiple times?
Ah, then yes, likely for two reasons.. to prevent the exact same game from happening if you play the same moves, and also as a way to choose between equally weighted (in terms of expected value) moves
I assume it's making a value-weighted move. To keep things interesting, you might choose from several moves that have similar EVs, esp. since the system only looks 4 moves deep, so there's uncertainty around how much better a given move actually is.
Nice, it's impressive that it can make some coherent move sequences in such short code. I beat it pretty quickly on the first try (it doesn't stand up well to tactics) but the degree of game sense it does have is impressive.
I lost the first game, it goes hard (charges forward acquiring material seemingly without any strategy). I'm a relative noob, but that's the first time I've been able to just adjust my style and easily win on the follow up. Impressive feat, thankfully I wasn't too 'humiliated' by it. Personally didn't spot any sequences in its play.
Loved the pixel art and all too, lulls you in. It is a little hard to tell from the pieces alone which they are.
There's some very basic move sequences I saw (moving into position for an attack and then executing, for example) but I also saw some odd moves such as opening with 1. ...a5
Because it doesn't have depth it can be tricked with tactics setups (forks, discoveries etc).
I can't speak for OP, but I doubt this is intended to replace LiChess. It's a uniquely designed proof of concept and a great one. We can give it a creative licence.
it went with a non-standard response to a queen's pawn opening and lost most of the pawns on the kingside falling into checkmate after getting baited into a trade that was good for it on material but devestating on position
I've played that a bunch of times this afternoon -- and succeeded in beating it in a few games.
I've been able to grab a few of its pawns early in the game. (I think its weakness is it assigns a low priority to protecting pawns. Or maybe it isn't able to look ahead far enough to see that I've got two attacks on the pawns instead of one.) It also didn't seem to be able to anticipate a "discovered check." It's good at setting up forks -- although at least once it forked me and then when I'd moved my King, didn't take the other piece!
Beautiful work.
I've been doing a lot of chess puzzles this last year and apparently it's paid off quite well. An easy mate. Though I've been inspired to make a (less beautiful) chess game.
I'm a "meh" chess player, but I do know proper openings and endgames, and I was bested by this engine. But for some reason, when I was only left with a King and a rook, it started blundering all its pieces: Bishop, a rook, 3 pawns, and a knight. It was left with only a king and knight.
And by blundered, I don't mean "made a bad move that I was able to take advantage of", I mean literally placed their rook infront of my pawn kind of blunder.
I think I've seen it start blundering more when the prospects are otherwise very bleak - if the best available move (with a depth-4 search) leads to a mate, who cares if you sacrifice a rook? I wonder if there was a quicker route to a mate you'd missed...
Yeah, I managed to trap it into giving up its queen. I was pretty happy. Then I moved my queen to a spot where I did not realize its bishop could just scoot straight through the forest and take my queen. Oops.
I'm about 1600 lichess, and this was a pretty easy conversion for me. It started by moving the a and h pawns in response to 1e4, which let me get center control and some advantageous pawn pickups.
Second game I got threefold while down a rook in a completely lost position (which I'll blame in part on the graphics, but was mostly me blundering).
Pretty cool game, nice job. It would be significantly better with just a little bit of opening knowledge, but obviously that isnt the point of this anyway. Im 1050 on chess.com and beat it first try, but the program sorta blundered the opening. I played the Queen's Gambit (very fitting)
Well, I did beat on the first try, but it's really impressive for such a small program! It didn't make really silly moves and was confused only by rather complex attacks with double threats.
Graphics are horrible though, very easy to confuse bishops and knights.
Lost one, won one, I'm a 1350 rapid player on lichess.
Of course I would say I lost the first game because of graphics and being hard to tell pawns from bishops, but it definitely wrecked me on the first try. Second try it blundered much more.
Well this is certainly very cool. It reacts well enough to my moves but I get the feeling that it does not know any of the established openings so each game starts a bit weird. Will keep going and try to beat it though.
Chess isn't a solved game and the established openings are good enough that playing without prior analysis is extremely difficult for even the best human players
This was surprisingly hard for such a small program. I haven't played chess in a long time, but was able to beat it on the first try, after dancing around the computer's very irritating queen for a while.
Knowing an opening helps a lot. I beat it on my first attempt with a london system opening. I've also been playing for about a year and a half and I'm at a 1600 rating on lichess in blitz.
I'm 1200 rating on chess.com and I lost the first game and won the second, the estimated rating of 1225 I've seen on other comments felt about accurate to me as well
Loved it! It gets crazy with the pawns, I was so surprised. The visuals of the game are so and so, but they are really cool nonetheless. Two thumbs up from me!
This statement doesn't make much sense to me. Is there a theoretical limit to the length of DNA, or do you mean that human DNA is ~100MB?
The human genome is about 6.4 billion base pairs, where each base pair is represented by the letters: Adenine (A), Thymine (T), Cytosine (C), and Guanine (G), i.e. 2-bits. 6,4 billion is then 6.4E9*2 = 1.28E10 bits = 1.28E10/8 bytes = 1600 MB. Or 800 MB if you count the bases as pairs of two.
Polychaos dubium's DNA is over 200 times as long as human DNA[0].
Thanks for doing the calculus however given that less than 1% of DNA actually has a coding role then it means that a human can be usefully encoded in 8MB.
Incomprehensible scheiße code. I looked around and I like this one because it has "meta-level" definition of movements and liitle bit of strategy. You could implement context-free chess games with varying rules for us congenitally lazy and dull-witted. https://github.com/thomasahle/sunfish/blob/master/sunfish.py
I'm about 1850 in Lichess blitz, and it played at the level of an earnest 12 year old, which is actually meant as a compliment. It got crazy with its kingside pawns in the opening (no book knowledge), lost material pretty steadily and was checkmated down a queen, bishop and two pawns by move 25.
But it developed some pieces sensibly, made obvious recaptures and was not reckless about king safety. It avoided pointless toggling moves and suicidal piece jettisoning.
You've earned the right to be a proud papa on this one.