A fair fight

[peter]

Hi Ronald!

What is important is that, apparently, the general level of play of current top engines can be reached (and most likely be far exceeded) by an approach to computer chess that is completely different than how all leading engines have worked since Claude Shannon wrote the first paper on computer chess.

What we've been shown by now is just, that the level of current top engines playing without book and therefore repeating the very little opening- repertoire of a 1'/move bound to play top engine plays, can be reached and exceeded.

From the 10 games shwown (and I'd really be very interested in seeing the rest of them as for some kind of doublettes more) the two lost with White are identical for the first six moves, the eight ones lost with Black are identical for the first 4 moves of White.

If the big achievement of A0 is learning, even if there wasn't any chance for it to learn from the known entity of SF on known hardware known to have always exactly 1' for each move (which isn't said in the paper that it wouldn't have happened maybe too, just a little bit after selfplaying only and before the match, just to be sure the selfplaying Elo- level, that otherwise wouldn't have had any calibration at all, would have "outperformed" SF indeed already ), in 100 games A0 should have learned especially to play against

1.e4 e5 2.Sf3 Sc6 3.Lb5 Sf6 4.d3 Lc5 5.Lxc6 dxc6 6.0–0
with Black and against

1.Sf3 Sf6 2.c4 b6 3.d4 e6 4.g3 Ba6 or ...Bb7
with White, which it obviously did quite well and so exceeded the top engine.

I'd say that was as for fair chances in chess the much bigger disadvantage then the hardware disadvantage of the given kind could have been, if it wouldn't have helped especially in this very special learing- achievement mostly too.

So to say A0 is the master of the universe in chess, I'd at least would want to see it win against SF with reasonable TC (to let it do its own timing, which is a programming- achievement of importance for a top engine too) and the engine playing with a reasonable book, at least broad enough to avoid this kind of really bad thematic tournament again with this kind of doublettes in this quality and quantity (just supposing it will have been as much related to the amount in the 10 shown games in the whole pool).

The fine graphs of A0's opening repertoire in selfplaying don't mean a thing as long as it hasn't had any chance to prove it right or maybe fully wrong yet against human theory of the last centuries at all till now in practice, playing at least once against a good book too.

Much more fun of course would be a fine Freestyle tournament with some good teams, A0 being just one of them.


In CSS we are just dreaming about building a manpower-software- hardware-cluster with some of the top computerchess- programmers


Anyhow, DeepMind, be challenged for some more fight, we want revenge or at least rematch

[Uri Blass]

But what do you think?

I think that it is completely irrelevant whether some AlphaZero prototype is or is not stronger than a perfectly tuned Stockfish setup.

What is important is that, apparently, the general level of play of current top engines can be reached (and most likely be far exceeded) by an approach to computer chess that is completely different than how all leading engines have worked since Claude Shannon wrote the first paper on computer chess.

And this approach is not just completely different from a programming point of view... it does not even need any programming (apart from the initial programming of the AlphaZero software and a bit of cleverness to adapt it to the rules of chess). They just decide how much hardware they want to throw at a problem, they push a button, and some hours later the thing has programmed itself. That is really superhuman.

Finally, a Post by Ronald de Man I agree with.
Very fair-minded, giving AlphaZero the Credit it deserves, unlike most of the old Pros here.

AZ's main advantage is twofold. It's computing power and reinforcement learning. If that much computing power was allotted to SF then the results would have been much better for SF. If SF had reinforcement learning and it was trained up it could play 1,000 elo stronger. You think that is bs?

Robin Smith, now deceased, ran a test RomiChess vs several top engines with the top engines using a truly humongous opening book and Romi's learning turned on. Romi gained 50 elo for every 5,000 games. A million games would not saturate with that humongous opening book. So doing the math, 1,000,000/5000x50 = 10,000 elo gain. This gain would be moderated somehow I'm sure but when would that moderation begin? Still think this is bs? Romi played 20 matches against Glaurung rated 2700+ at the time using the 10 Nunn positions. Match 1 saw Romi scoring 5%. By match 20 Romi scored 95%. That series of matches was 400 games using 10 positions. The performance gain was -512 to +512 = 1024 elo. Bottom line is if SF had reinforcement learning and was trained up it would be 4400 elo at least. Now I hope you guys can begin to understand that SF is far far superior to AZ. It is just that SF does not have Romi's reinforcement learning even after It was made available to them 11 years ago. So if they got trounced by something at least 11 years old it is not the algorithms fault. It is the shortsightedness of the programmers. Does anyone understand, yet?

I doubt Romi could really earn 1000 elo.

Maybe it could earn 1000 elo against a specific opponent in a specific opening but real gain is staying when you change the opponent or the opening.

If nobody played 1.a4 against Romichess then how Romi can learn how to play against it?

There are too many possible non losing lines in chess to get experience against all of them and you cannot teach romi not only to win against 1.a4 but also to win against 1.e4 c5 2.h4 and all the stupid lines that are not losing immediately.

[Michael Sherwin]

But what do you think?

I think that it is completely irrelevant whether some AlphaZero prototype is or is not stronger than a perfectly tuned Stockfish setup.

What is important is that, apparently, the general level of play of current top engines can be reached (and most likely be far exceeded) by an approach to computer chess that is completely different than how all leading engines have worked since Claude Shannon wrote the first paper on computer chess.

And this approach is not just completely different from a programming point of view... it does not even need any programming (apart from the initial programming of the AlphaZero software and a bit of cleverness to adapt it to the rules of chess). They just decide how much hardware they want to throw at a problem, they push a button, and some hours later the thing has programmed itself. That is really superhuman.

Finally, a Post by Ronald de Man I agree with.
Very fair-minded, giving AlphaZero the Credit it deserves, unlike most of the old Pros here.

AZ's main advantage is twofold. It's computing power and reinforcement learning. If that much computing power was allotted to SF then the results would have been much better for SF. If SF had reinforcement learning and it was trained up it could play 1,000 elo stronger. You think that is bs?

Robin Smith, now deceased, ran a test RomiChess vs several top engines with the top engines using a truly humongous opening book and Romi's learning turned on. Romi gained 50 elo for every 5,000 games. A million games would not saturate with that humongous opening book. So doing the math, 1,000,000/5000x50 = 10,000 elo gain. This gain would be moderated somehow I'm sure but when would that moderation begin? Still think this is bs? Romi played 20 matches against Glaurung rated 2700+ at the time using the 10 Nunn positions. Match 1 saw Romi scoring 5%. By match 20 Romi scored 95%. That series of matches was 400 games using 10 positions. The performance gain was -512 to +512 = 1024 elo. Bottom line is if SF had reinforcement learning and was trained up it would be 4400 elo at least. Now I hope you guys can begin to understand that SF is far far superior to AZ. It is just that SF does not have Romi's reinforcement learning even after It was made available to them 11 years ago. So if they got trounced by something at least 11 years old it is not the algorithms fault. It is the shortsightedness of the programmers. Does anyone understand, yet?

I doubt Romi could really earn 1000 elo.

Maybe it could earn 1000 elo against a specific opponent in a specific opening but real gain is staying when you change the opponent or the opening.

If nobody played 1.a4 against Romichess then how Romi can learn how to play against it?

There are too many possible non losing lines in chess to get experience against all of them and you cannot teach romi not only to win against 1.a4 but also to win against 1.e4 c5 2.h4 and all the stupid lines that are not losing immediately.

That really is not the point for the top engines because if their opponents resorted to bad opening play to defeat the training the trained engine would win anyway gaining even more elo. As far as Romi is concerned the gained elo would be of a pseudo nature that would only be revealed if non learning engines that do not play randomly at all purposefully started playing bad moves. But what engine author would resort to doing that? I guess that in the read.me file they could say please if Romi is in the tournament please use this losing book. So Uri you are correct in a technical sense but not in a practical sense.