Testing variants on the age-old game that can jolt players into creative patterns.
Read the whole story
Read the whole story
AI ruined chess. Now it’s making the game beautiful again
- Thread starter JournalBot
- Start date
I can see how Chess960 makes the game more spontaneous, but I feel like any variation of chess that doesn't have the randomness will just as easily become a game of memorization of computer openings.
Upvote
68
(68
/
0)
AI did't killed Chess, it put at another level.
I wrote Chess Engines (and other games such as Reversi) for decades, as a hobby but also commercially, for example the "Partner" for Grandmaster & Twice French champion Joël Lautier TV show about chess.
My goal wasn't to beat humans, that was an easy feat for 99% of the player at that time already, but to provide a partner that is enjoyable to play with.
I have a small collection of Chess Computers, and the most enjoyables to play with are the one that build the game, and from what I see with many incredible AI games, they are incredibly interesting, with fundamental novelties discovered by them in multiple parts of the game!
(Donald Knuth essentially solved the endgame decades ago)
I think AI could be made to give enjoyment to human players, not only crush them.
I wrote Chess Engines (and other games such as Reversi) for decades, as a hobby but also commercially, for example the "Partner" for Grandmaster & Twice French champion Joël Lautier TV show about chess.
My goal wasn't to beat humans, that was an easy feat for 99% of the player at that time already, but to provide a partner that is enjoyable to play with.
I have a small collection of Chess Computers, and the most enjoyables to play with are the one that build the game, and from what I see with many incredible AI games, they are incredibly interesting, with fundamental novelties discovered by them in multiple parts of the game!
(Donald Knuth essentially solved the endgame decades ago)
I think AI could be made to give enjoyment to human players, not only crush them.
Upvote
98
(100
/
-2)
Kind of a catch 22 given that part of the appeal of the game is the cold hard lack of randomness. If you've done poorly it can only be through the skill of your opponent or lack of skill by yourself.
Upvote
25
(27
/
-2)
If you want to keep chess interesting, I can also recommend a different technique: A friend and I only get out the chessboard a few times a year when we're drunk enough at four in the morning that it seems like a good idea, keeping us at a bad enough level that each game is new fun.
Or you go all in on computers and do stuff like 30 Weird Chess Algorithms
Or you go all in on computers and do stuff like 30 Weird Chess Algorithms
Upvote
38
(40
/
-2)
I would love this "self-capture chess" rules.
I am so bad at development that very often I just block myself in.
I am so bad at development that very often I just block myself in.
Upvote
19
(20
/
-1)
Yes but it's only a random starting point. Reading up about it now, you randomise White's back pieces (subject to some rules such as bishops must be on opposing colours etc). Then Black's pieces are simply set to mirror White's pieces, as in standard chess. They don't get independently randomised, so they are identical starting positions. Once play begins, it'll be all skill.
In case the randomisation produces more than the usual advantage to White, for tournaments apparently it is common for each position to be used for two games, with colours reversed, so that players get games as both White and Black in the randomised position.
Upvote
49
(50
/
-1)
Yeah, but AlphaZero calls all this soulless calculation into question. That's the interesting thing about AlphaZero--not that they're now getting it to play these rule variations.
Stockfish searches tens of millions of positions per second. AlphaZero searches 80,000 positions per second, and still wins, by playing very differently than a traditional computer engine. The article seems to gloss over this fact.
Upvote
55
(56
/
-1)
"Chess has been gaining popularity for years but experienced a pandemic boost as many people sought new intellectual stimulation, says Jennifer Shahade, a two-time women's US chess champion."
Women's US chess champion? So, at the championship level it's speed chess with the chess pieces weighing 120 pounds?
Separate chess leagues based on sex?
Why?
Women's US chess champion? So, at the championship level it's speed chess with the chess pieces weighing 120 pounds?
Separate chess leagues based on sex?
Why?
Upvote
101
(106
/
-5)
Upvote
36
(38
/
-2)
Judit Polgár called it, refusing to play women's championship and instead playing on men championship, and she literally crushed a lot of males on the board.
She was one of the strongest player in the world.
As a side-note related to the article, she also had an Ed Schröder's Chess Engine Module named after her (and her sister also competitive chess player), the Mephisto Polgar.
Upvote
54
(57
/
-3)
Is there a program that teaches humans how to play chess better in terms a human understands?
One can buy any number of books that explain why it is bad to move the queen early or the importance of energy in the middle of the board. The concepts are explained in simple terms; it may seem that moving the queen to the center of the board early in the game gives you strength, but often the opposite is true, you will value the piece is highly that any attack demands you respond, giving the other player control of the momentum of the game. Now while this is useful, there may indeed be reasons to move it early, something that can only be understood by examining the board.
But that is not how computers work, they are largely statistical. That, I assume, is because the goal of such programs is to win, not teach. They may indeed indicate that moving the queen out early is bad, but the explanation for that is that of the 50 million future moves it examined, 90% of them lost. That is not useful to a human.
Is there a teaching program that doesn’t just say don’t move queen in first 15 moves, but can actually say that moving the queen now is OK even before that depending on the board state? It seems that would be a best seller.
One can buy any number of books that explain why it is bad to move the queen early or the importance of energy in the middle of the board. The concepts are explained in simple terms; it may seem that moving the queen to the center of the board early in the game gives you strength, but often the opposite is true, you will value the piece is highly that any attack demands you respond, giving the other player control of the momentum of the game. Now while this is useful, there may indeed be reasons to move it early, something that can only be understood by examining the board.
But that is not how computers work, they are largely statistical. That, I assume, is because the goal of such programs is to win, not teach. They may indeed indicate that moving the queen out early is bad, but the explanation for that is that of the 50 million future moves it examined, 90% of them lost. That is not useful to a human.
Is there a teaching program that doesn’t just say don’t move queen in first 15 moves, but can actually say that moving the queen now is OK even before that depending on the board state? It seems that would be a best seller.
Upvote
16
(17
/
-1)
For chess being such a logical game, the pieces in the picture don't appear to be very logically placed. But maybe it was a game variant using those wonky rules.
Upvote
-4
(2
/
-6)
Mungus the Unhyphenated
Ars Tribunus Militum
Even before the advent of computers, tournament-level chess was on the path toward the realm of memorized openings and counters. Middle and end-game was where the challenge was, and humans were already finding ways to analyze and reduce those to statistics. Computers just amped it up.
My high school chess club didn't play competitively, so we tended to not only test ourselves in ordinary matches, we were keen on exploring variations to the classic rules for the same reasons as now -- find ways to interpret the game for fun an interesting mental challenges.
There's always been a sub-culture of exploring the logic and permutations achievable through carefully modifying how the game is played. Traditional computer programming, with what amounts to rigid statistical analysis driving equally rigid algorithmic output, is inevitably going to be boring because it relies on repeatable inputs and outcomes. AI allows for more randomness; only now can it begin to engage in what human minds have been entertaining themselves with.
Human players can inject "bad", "random", or "sacrificial" moves under traditional chess rules and throw very methodical, logical player off-track. Computers are unimpressed with flashy or confusion-inducing gambits, which makes them boring opponents still bent on crushing the human opponent. There's no life or interest in that kind of gameplay.
While I doubt that AI can truly display any kind of "temperment" or "personality" in gameplay, it will be interesting to see if it can engage with a human opponent in any sort of way that has more randomness and lively interplay in a match.
My high school chess club didn't play competitively, so we tended to not only test ourselves in ordinary matches, we were keen on exploring variations to the classic rules for the same reasons as now -- find ways to interpret the game for fun an interesting mental challenges.
There's always been a sub-culture of exploring the logic and permutations achievable through carefully modifying how the game is played. Traditional computer programming, with what amounts to rigid statistical analysis driving equally rigid algorithmic output, is inevitably going to be boring because it relies on repeatable inputs and outcomes. AI allows for more randomness; only now can it begin to engage in what human minds have been entertaining themselves with.
Human players can inject "bad", "random", or "sacrificial" moves under traditional chess rules and throw very methodical, logical player off-track. Computers are unimpressed with flashy or confusion-inducing gambits, which makes them boring opponents still bent on crushing the human opponent. There's no life or interest in that kind of gameplay.
While I doubt that AI can truly display any kind of "temperment" or "personality" in gameplay, it will be interesting to see if it can engage with a human opponent in any sort of way that has more randomness and lively interplay in a match.
Upvote
18
(19
/
-1)
Upvote
48
(48
/
0)
It would be an interesting challenge to create a machine player that taught a human by judging their skill and then consistently playing in a way that's flawed but slightly better than the human - trying to get them to spot the weakness. Theoretically you could then vary where the weakness appeared in order to teach different styles.
Of course there's no way I could do that!
Upvote
44
(44
/
0)
You don’t need randomization if you’ve opened the decision space enough that memorization is no longer feasible. Current rules you either move your pawns or knights on the first move and pawns can move one or two spaces so 20 possibilities. Add self-capture and that goes up to around 40. With the 2 move pawn all the time, on the second move the pawn that moved 2 spaces on the first move can do it again, or a pawn could take another pawn under the self-capture rule.
So just those two changes have the potential to really explode the decision space and make memorizing an opening book dramatically less useful.
I imagine you could do some sort of optimization problem where rather than select the layout at random, you chose the layout that leads to the largest viable decision space - the most positions at move 10 that could lead to a victory. Use ML to chose the opening arrangement of pieces that makes for the most varied game.
So just those two changes have the potential to really explode the decision space and make memorizing an opening book dramatically less useful.
I imagine you could do some sort of optimization problem where rather than select the layout at random, you chose the layout that leads to the largest viable decision space - the most positions at move 10 that could lead to a victory. Use ML to chose the opening arrangement of pieces that makes for the most varied game.
Upvote
14
(14
/
0)
This is pretty much it in a nutshell. Chess seems to be one of the few things that elicit this response. Weight lifters didn't stop trying once the forklift came around. Nobody has any existential angst because a motorcycle is faster than Usain Bolt.
For many decades now the popular views on AI have put chess on this strange pedestal. The prevailing logic was that if we could build an AI that could beat the best humans at chess we will have implicitly passed some arbitrary point where computers would become truly intelligent.
The truth is, our assumption turned out to be fundamentally flawed. It turns out that things that we considered intellectually difficult (Chess, physics, etc.) are actually relatively easier to generate AI for. What we have found to be exceedingly difficult are actually things that a human can do quite naturally (spatial orientation, visual processing, pattern recognition, etc).
It is still possible to play and enjoy chess as a human. Just as it is possible to see how fast a human can go in a race under certain conditions. I think it's silly to say that AI ruined chess, but it gets the clicks I suppose (it got mine).
Upvote
61
(61
/
0)
Wrestling, more likely.
Upvote
2
(2
/
0)
No, to my knowledge there have never been separate chess leagues based on sex.
There are additional tournaments for women--but the women are still in the same league as men (USCF, FIDE, etc.) and nothing prevents them from competing in the "regular" tournaments.
Why? Why not? If they think it's fun, let them have it.
If you think it's discriminatory and unfair to men and you want to participate in a women's tournament, you could probably take some sort of legal action... but... why...
Upvote
37
(38
/
-1)
Reminds me of when I was in a makeshift Chess club in middle school. It was really fun and although I was in the better end of the spectrum it was still always a bit of a toss up because everyone improvised and came up with strategies themselves.
Once people started reading chess books and looking up stuff online it became a lot more robotic and boring because people weren't making up their own stuff as much. I'm also a bit bias because I never looked up things and so I started to lose more often, but I didn't really mind losing before that.
Once people started reading chess books and looking up stuff online it became a lot more robotic and boring because people weren't making up their own stuff as much. I'm also a bit bias because I never looked up things and so I started to lose more often, but I didn't really mind losing before that.
Upvote
10
(11
/
-1)
Once the game becomes deterministic, the target demographic (?) changes. It becomes less adaptive strategy and more rote memorization. I really enjoy playing chess with other novices, since we're both just playing the board. But watching two grandmaster level players is kinda like watching a much more involved game of tic tac toe, where each move has an optimal response that shouldn't be deviated from.
Fischer Random sounds extremely compelling, I'd love to play it some time.
Fischer Random sounds extremely compelling, I'd love to play it some time.
Upvote
17
(18
/
-1)
I regularly play and follow chess, Alphazero has breathed new life into chess. It has shown the value of active/positional play/king safety - alphazero/Leela routinely sacrifices pawns/exchanges to launch amazing attacks a la Morphy.
Every GM has now incorporated this into their games and they're better off for it.
Having said that, classical chess is almost dead. This pandemic has been terrific for us chess lovers in giving us some amazing non-classical tournaments and showing how human these chess players are the moment they play faster time formats.
Rapid /Blitz chess keeps fidelity to the age old game while allowing players to play creatively and brilliantly.
Upvote
25
(25
/
0)
I started playing chess in high school. Chess is deep, elegant, and a lot of work.
Last year I discovered another game played on an 8 by 8 board that is not as deep.
Randomness, environmental differences between squares, and the ability to cause friendly fire damage between opposing pieces makes this game thousands of time more interesting than chess.
It's called "Into The Breach"
Upvote
4
(5
/
-1)
I suggest creation of a new kind of AlphaZero called AlphaFun, where the reward function is the pleasure it generates for another player.
But, please, don't call it the AlphaDopamine or people may realize too early where this is all going.
But, please, don't call it the AlphaDopamine or people may realize too early where this is all going.
Upvote
4
(5
/
-1)
It's not about strength/ability of course, it's to address a visibility problem. If an activity is male-dominated, a larger percentage of girls/women decide that it's not meant for them at a competitive level. That reinforces the sex disparity, and the imbalance continues. Womens' tournaments increases visibility of women players, bringing more people to the game. Eventually single-sex leagues would be unnecessary, but we're not there yet.
Upvote
48
(50
/
-2)
That's because we already knew our bodies were terrible. Just looking around in nature and the animal kingdom, we see plenty other creatures that are better, faster, stronger, pretty much in any conceivable aspect. Bacteria can eat us alive, whales dive deep under water and birds soar high into the sky. It makes sense to wrestle other humans, because doing that with a bear would... not yield a productive outcome.
Our brain, however, is the one thing that differentiates us from any other living being, making us unique on the planet and maybe even the galaxy, for all we know. Our intelligence is what defines us, and chess was seen as a way to quantify that.
Now, the moment we started losing at chess against machines, it meant we had lost our spot at the top, we had find our betters -- at least in as much as chess was a good proxy for general intelligence. Fortunately for us and for our existential angst, it turns out it's not.
We've then moved the goalposts, since chess was too deterministic and small in scope, to games like Go with their orders of magnitude more complexity and decision space. When that one fell, we moved them to self-learning, and after that to non-deterministic, imperfect knowledge games such as StarCraft. At each point, we had shocks forcing us to redefine what it means to be human, how we measure intelligence and where our limits are.
https://meincmagazine.com/science/2019/10 ... rcraft-ii/
StarCraft and similar games are where the battle currently rages, with top humans barely holding up against the best machines have to offer. We'll likely raise the stakes again when we lose that beach, but a day will come when we'll have to look up at our creations and realise they are our equals or betters in the only realm we ruled so far, that of the mind -- where do we go from there?
If that doesn't give one existential angst... well, here's a motorcycle-riding, machine-gun weilding bear for ya'
Upvote
39
(39
/
0)
Apparently a lot of us were in the Chess Club in High School. I was too, but it was back in '75 and we started playing D&D instead of Chess.
If we're thinking up variant chess games, how about Vampire chess? When one piece captures another, it absorbs their powers and can make the same moves.
If we're thinking up variant chess games, how about Vampire chess? When one piece captures another, it absorbs their powers and can make the same moves.
Upvote
13
(13
/
0)
"AlphaZero initially doesn’t know it can take an opponent’s pieces."
Huh? Isn't that a rather fundamental rule? Otherwise how would it know how to win other than to just block the opponents king from moving? And if its a case of anything goes then whats to stop it "learning" that it could take every piece in the path of say its queen, rook or bishop all the way to the edge of the board and win in no time? Also moves such as en-passant are not something you can just stumble over by "learning", they're specific and need to be known from the start.
Huh? Isn't that a rather fundamental rule? Otherwise how would it know how to win other than to just block the opponents king from moving? And if its a case of anything goes then whats to stop it "learning" that it could take every piece in the path of say its queen, rook or bishop all the way to the edge of the board and win in no time? Also moves such as en-passant are not something you can just stumble over by "learning", they're specific and need to be known from the start.
Upvote
3
(4
/
-1)
I think you've rather missed the point. Our physical strength and speed are never what made us human. We've known from the time we could stand up and see a lion coming that compared to other animals our size we're pretty wimpy so a forklift or any other physical machine is no existential threat to us , its just another thing we can't out compete in the physical realm.
Its *mentally* where we excel and what raises us above other animals and now AI has come along and in narrow areas is threatening our perceived dominance. Thats what makes people so uncomfortable about it.
Upvote
12
(12
/
0)
It learns by doing. You don’t hard code the rules. You just let it try a move and then tell it “allowed” or not. Eventually the AI finds its own way of encoding the rules without you needing to dictate the structure.
Upvote
20
(21
/
-1)
In other words the training software DOES have the rule that taking pieces is allowed programmed into it, which is not the same thing as saying they're the rule isn't coded in anywhere.
Upvote
-3
(4
/
-7)
The article meant to say that, initially, AlphaZero doesn't know that taking an opponent's pieces is (usually) advantageous.
AlphaZero definitely "knows" that it can take the opponent's pieces... the software does a Monte Carlo Tree Search, and the branches of the tree are the legal moves from the nodes (positions) in the tree. So yes, it's definitely programmed with a function that generates legal chess moves, and that could definitely be described as "knowing" what moves are available to it.
I believe DeepMind did a separate project where the algorithm was not pre-programmed with the rules of chess, and had to figure out the rules on its own. Presumably every time it tried to make an illegal move, it would forfeit that game. I believe that algorithm did successfully learn the rules. But that's a separate thing from AlphaZero.
Upvote
25
(26
/
-1)
Wrestling, more likely.
That'd be a cool new sport: Chestling!
Edit: typo
Upvote
6
(6
/
0)
That's some creative interpretating on your part. It's a bit like saying a pupil knows some rule beforehand, because the teacher says "that's wrong" when the pupil breaks it.
Upvote
4
(6
/
-2)