Chess 960 is an interesting variant in chess. It is basically like regular chess, except you remove the factor of memorization and let the players decide the result by skill.
It contains many random positions that might not occur in regular games, you can tell the difference just from looking.
One of the things that I have thought about is whether chess engines can also play well in chess 960. I want to know how far they have mastered this format and whether we stood a chance.
In regular chess, the gap between engines and humans is almost insurmountable, maybe we have a sliver of chance in this format. Or so I thought, it seems that chess engines are actually stronger in chess 960.
You will understand what I mean by this. Without further ado, let’s get started.
What is chess 960?
Chess 960 is a variant of chess created by Bobby Fischer. Bobby Fischer is a former world champion that is known for going against the powerhouses of the old soviet union, he is a famous chess player.
One of his main complaints against modern chess is the excessive use of memorization, games are usually decided by preparation rather than creativity. This has led Bobby Fischer to quit chess as he does not find it beautiful anymore.
To counter this criticism of modern chess being filled with too much memorization, he created chess 960. Basically, in chess 960 the starting positions are randomized for each game.
This means that it is significantly harder to prepare (almost impossible) for a chess 960 game.
However modern chess engines are becoming stronger and stronger every day, do they have what it takes to play chess 960 well?
We will answer that, but first I need to explain where you can find the engines for chess 960.
Does chess 960 require chess engines that are different from the standard one?
No, you don’t need a “new” specialized chess engine in order to analyze a chess 960 game. However you do need to customize the standard chess engines in a way that they are able to analyze chess 960 positions.
Some of the standard chess engines that can analyze chess 960 are stockfish, leela chess zero and komodo.
If you are wondering why you cannot analyze chess 960 positions from the public server of stockfish, it is because the custom is not released to the public yet.
What is known though are the methodologies that chess engines use to analyze the positions given to them. I have written an article about this.
Looking at these methodologies, there shouldn’t be a lot of problems in making the jump between standard chess to chess 960.
In the official tournaments for chess engines (tcec) there have been matches between engines that are played in the chess 960 variant. The developers have to perform some tweaks in the codes in order to make the engines play chess 960.
Chess 960 has a lot of rules that are not applicable in standard chess, the standard chess engine has to be customized.
The customized versions of these engines are not available for public yet, although that may change in the future. As chess 960 becomes popular, these customized chess computers will be introduced publicly.
Are humans better than chess engines in chess 960?
Chess engines are still significantly better than humans inches 960. It is understandable if you think that humans will stand a chance against engines, at least in this variant.
Unfortunately chess engines do not really get weaker in chess 960, in fact they are probably stronger in this mode of chess.
You see, the strongest chess players in the world get into their decision because of memorization. They have analyzed thousands of games and have memorized many opening lines. Chess 960 is created to minimize this factor in the first place.
You can think of it as a direct nerf to today’s modern super grandmaster, we can play well if it is on regular chess. Chess engines on the other hand do not rely on preparation and memorization, they can play well on their own.
Some chess engines have opening books that are forced into them, but it is not the source of their strength.
Stockfish and leela chess zero for example are shaping our modern opening theory since they keep finding moves that are out of our opening books.
Chess engines win games by using brute force calculation, basically they analyze as many possible positions in a short amount of time. They will not be affected by the changes in format.
They can still rely on the brute force approach even in chess 960. While humans get weaker in chess 960 positions, the engines remain the same. This is a scary thought if you think about it, but it is true.
How good did the chess engines performed in the tcec (chess 960)?
In the tcec chess 960 variant, things the chess engines performed as much as you’d expect. They played the best moves in every single position and won’t seem to slow down.
Some chess engines are better and have outperformed the others, it seems that the neural network modification is important in this.
Those that have the neural network upgrade outperformed chess engines that are only using the brute force calculation approach. If you compare the games played by chess engines to humans, the difference is clear.
If you want to know the complete history of neural networks, you can read my other article about this. Neural networks are the future of chess engines.
In chess 960 it seems that humans become weaker, while chess engines remain relatively strong. The human reference that I am using is the world championship match between Wesley So and Magnus Carlsen.
Wesley So won the match and became the first chess 960 world champion. They have played incredible chess, however the games between the engines just seem much more advanced.
By looking at it side-by-side I can tell that the engines are still stronger than humans even in chess 960.
The games played between engines are clearly much more precise, they contain less errors than the games played by the super grandmasters.
And if you have watched the live match between Carlsen and Wesley So, you would know that they have committed some mistakes during their match. In fact the chess 960 games played even before that match contain a lot of mistakes as well.
The engine games on the other hand are much more refined, they are still clearly better than humans as of the moment.
Can humans ever surpass chess engines in Fischer random?
You might not like the answer to this, but it seems that we would never surpass chess engines even in chess 960.
The learning capacity of humans is almost at its peak, super grandmasters have become so strong because of the engines in the first place. Super grandmasters utilize engines in their preparation and memorize many engine lines.
Engines on the other hand, well, do not really rely on us to become stronger anymore. I can give them updates but with the introduction of the neural network upgrade, it seems that they can now learn on their own.
After alpha zero crushed stockfish, the neural network upgrade became more popular than ever. Even stockfish itself has now incorporated the neural network upgrade.
If one side is going to be stronger in chess 960, it is likely to be the engines and not the humans. They can now learn and improve on their own, I suspect that they can still be stronger than they are right now.
Their improvements also come significantly faster than humans (one update is all that is needed). In the future only the engines can be stronger in chess 960, not humans.
Are the chess engines good at analyzing chess 960 positions?
Okay, we have now established that they are good at playing chess 960, but how good are they at analyzing it?
The thing is we don’t know, we don’t really know if their analytical power is as good as their playing power. We did have a look at their analysis, but we can never really tell if their analysis is good or bad.
I mean we can see that the recommendations are decent looking moves, but how can we know that those recommendations are the best?
It is the engine after all, any move that It suggests will be hard to disprove. In my opinion, their analysis is just as good as their playing power.
If they are good enough to incorporate the rule changes in chess 960, then they are good enough to analyze it.
Since chess engines play chess 960 better than humans, their recommendations will always be held in a higher regard. In chess 960, the chess engine’s analysis is always respected, as it should be.
Chess engines can play chess 960 well, the difference in the rules can be overcome by a few modifications.
The chess engines that are used to play chess 960 rules are the same old regular engines, namely stockfish, komodo, and leela chess zero.
Since the engines use what’s called the brute force calculation approach, they are not as affected by the random positions.
Humans on the other hand find it difficult to excel in chess 960, the variant is literally created to make modern grandmasters less reliant on memorization.
Chess engines don’t really need memorization in order to play well, this makes the gap between humans and engines significantly larger in chess 960.
The chess engines can also analyze chess 960 positions well, the gap can only become bigger and not smaller. That is all for this article, thank you for reading.