Can AI Help Us Reimagine Chess? (acm.org) 64
Three research scientists at DeepMind Technologies teamed up with former world chess champion Vladimir Kramnik to "explore what variations of chess would look like at superhuman level," according to their new article in Communications of the ACM. Their paper argues that using neural networks and advanced reinforcement learning algorithms can not only surpass all human knowledge of chess, but also "allow us to reimagine the game as we know it...."
"For example, the 'castling' move was only introduced in its current form in the 17th century. What would chess have been like had castling not been incorporated into the rules?"
AfterAlphaZero was trained to play 9 different "variants" of chess, it then played 11,000 games against itself, while the researchers assessed things like the number of stalemates and how often the special new moves were actually used. The variations tested:
- Castling is no longer allowed
- Castling is only allowed after the 10th move
- Pawns can only move one square
- Stalemates are a win for the attacking side (rather than a draw)
- Pawns have the option of moving two squares on any turn (and can also be captured en passant if they do)
- Pawns have the option of moving two squares -- but only when they're in the second or third row of squares. (After which they can be captured en passant )
- Pawns can move backwards (except from their starting square).
- Pawns can also move sideways by one square.
- It's possible to capture your own pieces.
"The findings of our quantitative and qualitative analysis demonstrate the rich possibilities that lie beyond the rules of modern chess."
AlphaZero's ability to continually improve its understanding of the game, and reach superhuman playing strength in classical chess and Go, lends itself to the question of assessing chess variants and potential variants of other board games in the future. Provided only with the implementation of the rules, it is possible to effectively simulate decades of human experience in a day, opening a window into top-level play of each variant. In doing so, computer chess completes the circle, from the early days of pitting man vs. machine to a collaborative present of man with machine, where AI can empower players to explore what chess is and what it could become....
The combination of human curiosity and a powerful reinforcement learning system allowed us to reimagine what chess would have looked like if history had taken a slightly different course. When the statistical properties of top-level AlphaZero games are compared to classical chess, a number of more decisive variants appear, without impacting the diversity of plausible options available to a player....
Taken together, the statistical properties and aesthetics provide evidence that some variants would lead to games that are at least as engaging as classical chess. "Chess's role in artificial intelligence research is far from over..." their article concludes, arguing that AI "can provide the evidence to take reimagining to reality."
"For example, the 'castling' move was only introduced in its current form in the 17th century. What would chess have been like had castling not been incorporated into the rules?"
AfterAlphaZero was trained to play 9 different "variants" of chess, it then played 11,000 games against itself, while the researchers assessed things like the number of stalemates and how often the special new moves were actually used. The variations tested:
- Castling is no longer allowed
- Castling is only allowed after the 10th move
- Pawns can only move one square
- Stalemates are a win for the attacking side (rather than a draw)
- Pawns have the option of moving two squares on any turn (and can also be captured en passant if they do)
- Pawns have the option of moving two squares -- but only when they're in the second or third row of squares. (After which they can be captured en passant )
- Pawns can move backwards (except from their starting square).
- Pawns can also move sideways by one square.
- It's possible to capture your own pieces.
"The findings of our quantitative and qualitative analysis demonstrate the rich possibilities that lie beyond the rules of modern chess."
AlphaZero's ability to continually improve its understanding of the game, and reach superhuman playing strength in classical chess and Go, lends itself to the question of assessing chess variants and potential variants of other board games in the future. Provided only with the implementation of the rules, it is possible to effectively simulate decades of human experience in a day, opening a window into top-level play of each variant. In doing so, computer chess completes the circle, from the early days of pitting man vs. machine to a collaborative present of man with machine, where AI can empower players to explore what chess is and what it could become....
The combination of human curiosity and a powerful reinforcement learning system allowed us to reimagine what chess would have looked like if history had taken a slightly different course. When the statistical properties of top-level AlphaZero games are compared to classical chess, a number of more decisive variants appear, without impacting the diversity of plausible options available to a player....
Taken together, the statistical properties and aesthetics provide evidence that some variants would lead to games that are at least as engaging as classical chess. "Chess's role in artificial intelligence research is far from over..." their article concludes, arguing that AI "can provide the evidence to take reimagining to reality."
Wow late (Score:1)
I was going to blame Slashdot for being late, but it was just published in the Communications of the ACM, even though every other news site carried it a year ago.
The ACM has really gone downhill.
Re:Wow late (Score:4, Interesting)
In many cases, communications of the ACM is a republication of an article that appeared elsewhere. The idea is that you published a real cool article in your own scientific community, but the editors of CACM may think that the article could be of interest to a broader audience. So it is either republished or reworked for publication. Then it enters production and articles are grouped thematically so that the paper issue has a theme.
I read CACM fairly often. Well, I scan the table of content and mark articles that I'll read later. It is really nice!
The results of the study (Score:4, Funny)
They explored a lot of variations and the first thing they found was that except for variations that led to a forced win for the payer to move first ( or second) there was no basis to conclude whether change was good or bad. So they had to incorporate a heuristic that evaluated if the game play was "fun" or not. Eventually they evolved the rules to the the funnest game and found the overall winner was a rule set where all pieces moved only forward and only along the diagonals by one square at a time. Captures were made by jumping over an opponent if it had an open space on the other side. And you could also jump multiple hood as well as jump your own pieces. All pieces reaching the far side could be promoted to a piece that moved backward as well as forward.
Re: The results of the study (Score:2)
They also found that the new game version was best played on a porch with a glass of sippin' whiskey or a mint julep, and against a human friend rather than wasting time with a computer.
Re: (Score:2)
This would be perfect if you and your friend are fairly evenly matched. Rarely is that ever the case though. But if you're in it primarily for the social aspect, it probably doesn't matter to you as much as those who like to play a bit more seriously. I have to admit that some of my most enjoyable time playing was in a chess club that met in a German Gasthaus weekly (a few decades ago).
Re: The results of the study (Score:2)
Reimagine Scoring (Score:2)
Re: (Score:2)
The results? (Score:4, Interesting)
What a tease. Which games were more fun?
Their video doesn't say. It does say that removing moves, such as castling, didn't make the games less diverse. They argue that these changes on the whole made the game more diverse. (minute 3:12 in the video).
Figure 2 in their text shows that on average white gains an advantage over black - something white already has in the classic game.
https://dl.acm.org/cms/attachm... [acm.org]
Perhaps in future research that could look for variants that make the game more fair?
Re: (Score:1)
Torpedo chess (where pawns can move two squares, even if they've already moved before) is the most dynamic. Unfortunately, white looks to benefit greatly more than black in this variant.
Surprised removing the statemate rule made such a difference to white's winning chances (30% more wins) considering it's so rare to see stalemate in practise. Sideways pawn moves didn't seem to help the game much. Only around 13% extra to help white, and black wins
Re: (Score:2)
Surprised removing the statemate rule made such a difference to white's winning chances (30% more wins) considering it's so rare to see stalemate in practise.
Well, in real games, a the person attacking will attempt to avoid creating a stalemate since that is just a draw, and they would rather win. With this change a rule giving the attacker a win for a stalemate, there is no longer a reason to avoid that situation.
Re: (Score:2)
How the moves are distributed in terms of strength(in more extreme cases 'strength' may be direct
Re: (Score:2)
Or we could leave it alone (Score:2)
Re: (Score:2)
Change for change's sake here.
Yes!
Also, if you must, stop "re-imagining" and just make something new.
This applies not just to games but also to movies and TV shows.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Chess is actually a game that is approaching draw-death. Despite it's reputation as the number one mind-sport, it is actually a very poorly designed turn-based tactics game that in my opinion has numerous design flaws and strange rules that do not make sense:
Re: (Score:2)
Best use of AI (Score:4, Informative)
The best use of AI could be to scan wikipedia and find out that there is an entire page devoted to the different variants of chess [wikipedia.org] and there is already a metric shit ton of "re-imagined" chess games.
Re:Best use of AI (Score:5, Interesting)
I play chess very casually, usually against my phone running an embarrassingly low rating bot. One advantage the bot has is that the programmers included a database of opening moves (aka book moves) that I have not taken the time to memorize because, well, I play casually. As I've played, I've started to figure out some of these standard opening patterns and, frankly, the opening part of the game is boring because of that.
From the little I've explored in the vast online chess world, that idea seems to be popular: so much of the space of opening moves has been explored and understood that if you don't memorize a metric ton of patterns, you are doomed to lose. That does not sound like fun to me either way: losing, or necessarily memorizing lots of material so you have a fighting chance.
So, I started looking at variants, and the one I like the best is Fischer Random, invented by Bobby Fischer, where the non-pawn pieces are placed randomly (both sides get the same random starting placement). There are some constraints, for example to ensure that the two bishops are on opposite-colored starting squares, creating 960 possible starting positions. That expanded number of starting positions means (a) memorizing the standard book is now nearly worthless except for a 1-in-1000 chance of getting the traditional placement, and (b) it becomes three orders of magnitude harder to memorize the enlarged book, making that task, at least for now, impossible.
So now, it's all about chess, which is fundamentally about constraint propagation and your innate ability to search the space, rather than how much you've memorized.
Re: (Score:3)
if you don't memorize a metric ton of patterns, you are doomed to lose
That does seem to be the case.
I started looking at variants
Interesting. What I did instead was stop playing chess years ago.
Re: (Score:2)
>> if you don't memorize a metric ton of patterns, you are doomed to lose
> That does seem to be the case.
Chess player (Fide Master) here. What you are saying is simply wrong. A few simple rules (put a pawn in the center, develop your pieces, castle early) and paying just a little attention (chess is all about tactics, after all) will get you through the opening.
If you don't believe me, run computer analysis on all the games played on sites like lichess and others, and you'll notice that the vast ma
Re: (Score:2)
I don't think Fisher Random chess will achieve its goal for the following reason. If it ever replaced the standard game, an opening book would accumulate on each of the 960 possible starting positions. Then in order to become a world-class player, instead of memorising as much as is humanly possible about one starting position, you'd have to memorise about a thousandth as much on each of the 960 starting positions.
Re: (Score:2)
so much of the space of opening moves has been explored and understood that if you don't memorize a metric ton of patterns, you are doomed to lose.
If you memorize openings, then you will have a 50-100 point rating advantage. Intuition, calculation, and understanding the game matter more: https://www.youtube.com/watch?... [youtube.com]
The reason to learn openings is so you can get into interesting positions. For example, you might want to spend a little time learning the Stafford Gambit. It's a little risky, but leads to fun games.
Re: (Score:2)
Its unsound. Accept it, then d3, then be2, c3 and nd2. Straightforward pawn down for no compensation.
Worth explaining that to play as an amateur you don't have to memorize openings. The reason is that after you get through the basic main line, you'll be faced on both sides with a lot of different choices all of which have been extensively analyzed.
However, the difference between the better and the worse is not winning and losing immediately. Its a slightly better or worse position.
Now at Grandmaster lev
Re: (Score:2)
Its unsound. Accept it, then d3, then be2, c3 and nd2. Straightforward pawn down for no compensation.
There's compensation: initiative. It's definitely not straightforward.
More importantly, it's fun.
Re: (Score:2)
Re: (Score:2)
There are a handful stupid mistakes you can make in an opening. That is the only thing super important to avoid and know.
The next thing are counters, someone playing a certain opening, and you counter with a well known defensive opening. While that is pretty basic, it is already a kind of case as you mention. For most hobby players already at the edge to memorize.
Then comes the endgame. E.g. how to beat a single king with a tower and a knight (no idea if that is even possible, but expert players memorize ho
Re: (Score:2)
Well.. that's chess.
Brain scans of grandmasters [scientificamerican.com] have shown the big difference between them and amateurs is they're using their memories a lot more.
Re: (Score:2)
Re: Best use of AI (Score:2)
Try playing against real opponents at your level. It makes it a lot more fun because you both make the same type of mistakes at about the same rate. It helps you get better and finding your opponent's mistakes as well as avoid making them yourself.
The mistakes computers make are usually not at all like the ones you'll make yourself.
thank you (Score:1)
for posting this link
I doubt it will ever happen (Score:3)
Re: (Score:2)
The rules of standard chess have changed many times in history.
Re: (Score:2)
Re: (Score:3)
Threefold repetition: 1883 (https://en.wikipedia.org/wiki/Threefold_repetition)
En Passant was not standard until 1880 (https://en.wikipedia.org/wiki/En_passant)
The current version of castling was established in 1620 (https://en.wikipedia.org/wiki/Castling) - 400 is inside 500
Free Castling was abolished in the late 19th century (same ref)
I'd call these pretty major changes.
Re: (Score:2)
I will
Re: (Score:2)
Just not in our lifetimes.
Re: (Score:2)
True, but only just. There are people who have died inside the past 5 years who were alive before the last major change to the standard rules.
What about 11 dimensional chess ? (Score:2)
AlphaZero and Superhumans (Score:4, Insightful)
>AlphaZero's ability to continually improve its understanding of the game, and reach superhuman playing strength in classical chess and Go ...
To me that sounds off, or at least rawkward.
Like saying hydraulic systems have superhuman powers.
Re: (Score:2)
Re: (Score:2)
>Hydraulic systems do have superhuman powers
I meant a superhuman would have to exist before we could talk about the kind of capabilities he had.
Of all those variations (Score:4, Interesting)
I'm not much of a chess guy, haven't played in years and was only casual so take this with a grain of salt. I'm thinking that of all those variations, being allowed to capture your own pieces is the most interesting. It's so much against the grain of the game, I'm thinking that if you can formulate some strategy where you knock out one of your own pieces that's blocking a move and checkmate the other guy it'd catch them off guard. Of course they know you're playing that variation, and you have to watch for it too; but it really does seem like the kind of thing that could make somebody want to flip the board upside down.
There's only one big question. (Score:2)
If grandmasters are correct that it is board combinations that matter, a strong enough AI should be able to play perfect* chess with a lookahead of a single move. If you can't do this, then grandmasters are using a simplification to make playing at that level easier but it's a dead-end approach.
*Since it's a full information game, there is a strategy that, if followed blindly, will result in a guaranteed win against any defence by one of the sides.
Re: (Score:2)
In theory that is possible, but in practice it requires too much RAM. Computers can only use that strategy if there are 7 or fewer pieces. In that case, the computer does play perfectly.
Re: (Score:2)
I'm curious why it would require so much RAM. I'm assuming you're thinking of storing every possible board combination, which would certainly require too much RAM, but is that necessary? (It's also technically looking ahead, merely in advance.)
I'm thinking in slightly different terms, representing a state in such a way that you can input two states to a function and determine how the probability of being on a path in which you can win against any defence will change with any given move without reference to
Re: (Score:2)
I'm thinking in slightly different terms, representing a state in such a way that you can input two states to a function and determine how the probability of being on a path in which you can win against any defence will change with any given move without reference to any other states and without recourse to a full expansion of the entire graph of possible moves.
I'm not sure I understand the algorithm described here. By "state" do you mean board position? If so, why would you input two states?
OK, fine, but... (Score:2)
Re: (Score:2)
kick the king chess (Score:2)
On the chess variants page there is a variation which is a simplification of the laws of chess:
https://www.chessvariants.com/... [chessvariants.com]
In kick the king chess, the king is permitted to move into check and the object is simply to capture the king, not to bring about checkmate. Because it's simpler, it's quicker to teach beginners.
It's essentially the same as the "Stalemates are a win for the attacking side (rather than a draw)" variant mentioned in the TFA. The difference would be the rare stalemate where there isn'
Some of these variations are already in Xiang Qi (Score:5, Informative)
1. Castling is never allowed, mostly because the king (called a "general" in Xiang Qi") can't leave a central area called "the palace". If the king wasn't confined to a small area, the game would be almost impossible to win at the higher levels.
2. Pawns can only move 1 space always. No exceptions. Thus "en passant" capture does not exist.
3. Stalemates are wins for the attacking side.
4. Pawns can move sideways one space but only after crossing the river. Note that pawns can't promote to other pieces so in Xiang Qi, if a pawn reaches the last row, it can only move sideways 1 space at a time. You might note too that in Xiang Qi, pawns can capture the space directly in front of them, not diagonally in front of them as in Chess. So if you grow up playing Chess and you put one of your pieces on the space in front of a pawn in Xiang Qi to try to block it, you may be saying goodbye to your piece on the next move. Note too that I keep saying "space" not "square" because in Xiang Qi the pieces move on the intersection of lines and not on the squares of the board.
Re: (Score:2)
There is also a japanese variation, Shogi, with the prime difference that captured pieces can be put back on the board by the one who owns them.
Wikipedia says, it originated in India https://en.wikipedia.org/wiki/... [wikipedia.org] - most likely how ever it developed from Xiang Qi.
While modern chess also originated in India, I doubt "chess" on one hand and Xiang Qi and Shogi on the other hand have much to do with each other. I assume they are simple co-developments. Or what is next? Checkers also a variant of chess? I don
Re: (Score:1)
My favourite bit is how the rooks take pieces: not the one they can see directly, but the one *behind* it. It doesn’t matter whether the first piece is yours or the opponent's. This creates all sorts of chaos with pins, and doubled-up rooks are nasty.
N-dimensional (Score:2)
...explore what variations of chess would look like at superhuman level,...
Superhuman? None of their variations even had more than 2 dimensions!!
10 years ago (Score:1)
New rules? (Score:2)
Basically every chess club I ever heard about has kids education. And not only kids but also adults constantly change the rules on the board.
So this is absolutely nothing new.
Human edge? (Score:1)
Is there any research into patterns of board or card games that AI is poor at compared to humans?
Defeats the purpose of why so many play (Score:2)
The reason many people are drawn to chess is that it has been the same for so long, and is expected to remain the same for a long time to come. People are not going to put a lot of time and effort into learning about a game that will be obsolete in a few years.
I think very few regular players think there's a need for a change. On lichess alone, there are millions of games played per day, the overwhelmingly majority of which are not draws.
Who cares? (Score:2)
So, let's see, we can change some game, and get another game. The other game might be similar or different. Let's measure.
So what?