I know AlphaZero (an AI) is one of the best top chess engines but Stockfish is also improving. Which one is better at chess?
1 Answer
We can't say for sure since AlphaZero is a private engine, i.e. we don't have games between it and the latest versions of Stockfish. Still, if AlphaZero hasn't improved since it was unveiled, it will likely lose to the latest version of Stockfish. That's because AlphaZero beat Stockfish 8 by +155 = 839 -6, which is an elo difference of about 50. The latest versions of Stockfish are capable of beating Stockfish 8 by about 150 elo (see also this). It's true that elo isn't transitive (i.e. engine A can beat engine B, which beats engine C, but C could still beat A in a match); however, the differences are so large that it would be incredible if AlphaZero is capable of beating the latest versions of Stockfish. You ask about Stockfish 10 (not latest Stockfish), which still beats Stockfish 8 by 100 elo, so it should be capable of beating AlphaZero as well.
Having said that, Leela Chess Zero is an engine based on AlphaZero techniques. It incorporates many new innovations not in the original paper, and therefore should be stronger than AlphaZero. Indications are it's competitive with Stockfish, and (for now at least) stronger: it won the most recent version of the Top Chess Engine Championship by five games - an elo difference of about 15.
Caveats:
- In the world of computer chess, 100 games is an incredibly small sample size. Engine developers regularly need tens of thousands of games to conclusively say that one engine is stronger than the other. These games are played at bullet time controls, which isn't really chess as us humans understand it. Still, we can't play ten thousand games at long time controls - that takes too long - so we have to be satisfied with 100.
- Stockfish continues to improve. As of time of writing, Leela has kinda stalled while a new net is being trained. So while Leela was (probably) stronger two months ago, it might not be the strongest engine right now.
- Stockfish is not Leela's only competitor. In fact just two seasons ago, Leela actually failed to qualify for the superfinal, losing out to AllieStein. Like Leela, AllieStein is a neural network engine; unlike Leela, AllieStein includes human knowledge. Plus, as of time of writing, all three of Leela / AllieStein / Stockfish are behind Stoofvlees in the qualification tournament for the superfinal.
You might be interested in the ongoing Season 18 of the Top Chess Engine Championship, live games are available here.
tl; dr: AlphaZero is probably inferior to Stockfish 10, and even more so to the latest versions of Stockfish. Leela is at least competitive with Stockfish, and there's a good chance it's stronger (right now) as well. For the question "who is the strongest engine right now?", the situation is fluid; there are several engines that could all claim to be #1 at any one time. For the foreseeable future, this situation is likely to remain the case.
Edit: Deepmind acknowledges that AlphaZero would probably lose to the latest versions of Stockfish and Leela. They don't perceive it as a challenge, however, because they're not a chess engine company and don't view having the best engine as a big deal.
Update: with wins in the last four TCEC championships, Stockfish is probably the strongest engine right now, up to small sample size effects (see first caveat above).