Alpha Zero question

[Leo]

I am a little out of the loop these day with the latest chess engine developments. How would Alpha Zero match up against the current Stockfish TCEC champion? Who invented Neural networks for chess? I forgot. It really was a revolution.

[Werewolf]

There are loads of people better qualified than me to answer, but I'll start the ball rolling.

1) Current SF would be stronger than the Alpha Zero that beat SF8. This is easy to test as you can play SF Dev against SF8.
2) Current Alpha Zero may be improved or not. It is unknown to the public.
3) NNUE is what SF uses and, as I understand it, it is small net that works well with Alpha Beta engines like SF. Note Lc0 and Alpha Zero are not NNUE based but use larger nets which are slower and hence they also use a different search: MCTS.

[smatovic]

[...]
Who invented Neural networks for chess?
[...]

AFAIK NeuroChess from 1995 by Sebastian Thrun was the first neural network used in evaluation for playing chess:
https://www.chessprogramming.org/NeuroChess
...
then, 2015, with more advanced hardware, we have for example Giraffe by Matthew Lai:
https://www.chessprogramming.org/Giraffe
...
then 2017 AlphaZero with TPUs, and since 2018 Lc0 with GPUs, and ~2020 the Shogi guys backported NNUE to Stockfish.

--
Srdja

[supersharp77]

I am a little out of the loop these day with the latest chess engine developments. How would Alpha Zero match up against the current Stockfish TCEC champion? Who invented Neural networks for chess? I forgot. It really was a revolution.

Well You will have to double back to the original 2018 match links/debates...there was quite alot of back and forth & push and pull about that "so called match" as no one was actually able to reproduce the results! What I recall (just like it was yesterday) was that Google....(Alpha Zeros Team) did not allow Stockfish a opening book and there was no "Stockfish Team" present during the so called "match" just a base Stockfish 8....."Match" was 100 games with 8 wins by Alpha Zero (98 draws?) but most were unpublished...and only the wins were published?!.......Alpha Zero at this present time (5 years later) remains a mystery...LC0 by most accounts may well be stronger than Alpha Zero currently (although opinions will vary) so until someone gets a working copy of Alpha Zero.......we will never know.....original arguments were CPU vs TPU....(Now GPU) and power supply issues and also time control & opening book issues...."Who Knows For Sure!!"....

[Leo]

Thanks everyone. That was very informative.

[FormazChar]

I disagree with some of the points. There have been several reproding matches by various people showing that engines today have passed AO level by 150+ elo at similar conditions vs the same sf opponent.

If 2018 A0 entered TCEC today it would fight for 3-4th place.

[reflectionofpower]

I am a little out of the loop these day with the latest chess engine developments. How would Alpha Zero match up against the current Stockfish TCEC champion? Who invented Neural networks for chess? I forgot. It really was a revolution.

This was not a technical look at AlphaZer0 but Alpha GO, but the concepts were similar, and it's a good documentary. I thoroughly enjoyed.

https://www.youtube.com/watch?v=WXuK6gekU1Y

[jkominek]

I disagree with some of the points. There have been several reproding matches by various people showing that engines today have passed AO level by 150+ elo at similar conditions vs the same sf opponent.

If 2018 A0 entered TCEC today it would fight for 3-4th place.

I'm interested in learning more about these. Do you have links to where this was discussed? If it was in talkchess I must have missed it somehow.

An exact reproduction attempt requires access to, or a server of the same specification as that used by TCEC from Seasons 8 through 16. I wonder if the efforts at reproduction you allude to went to such lengths.

[Raphexon]

There are loads of people better qualified than me to answer, but I'll start the ball rolling.

1) Current SF would be stronger than the Alpha Zero that beat SF8. This is easy to test as you can play SF Dev against SF8.
2) Current Alpha Zero may be improved or not. It is unknown to the public.
3) NNUE is what SF uses and, as I understand it, it is small net that works well with Alpha Beta engines like SF. Note Lc0 and Alpha Zero are not NNUE based but use larger nets which are slower and hence they also use a different search: MCTS.

There's not a lot of "monte carlo" left about Leela's and A0's search, it's PUCT.
It's like calling SF a pure minimax engine.

[Raphexon]

I am a little out of the loop these day with the latest chess engine developments. How would Alpha Zero match up against the current Stockfish TCEC champion? Who invented Neural networks for chess? I forgot. It really was a revolution.

Neural networks have existed since the 80s.
Blondie25 is the earliest I know (2006)