alt Hacker NewsI do understand you, but you and your "conviction" are wrong. Apparently you aren't even familiar with AlphaZero.
> Adding pieces up only increase complexity up to a point.
As you said, until you reach 32 pieces. What you vastly underestimate is how much complexity is added at each level. You're like the king in the fable who agreed to give the vizier a small amount of grain: 1 grain for the first square, 2 grains for the second square, 4 grains for the third square, etc. The king thought he was getting a bargain.
> The thing is : you don't need to train it on all games from the endgame table before it converges to perfect play.
But you do, because there is no algorithmic simplification, at all. Strong chess players understand that while there are common patterns throughout chess, their application is highly specific to the position. That's why we have endgame tables, which are used to solve positions that pattern matching doesn't solve. You can get excellent play out of an NN, but that's not the same as solving it. And the absence of Bellman violations is necessary, but not sufficient ... you can't use it to prove that you've solved chess. The fact is that it is impossible within pragmatic limits to prove that chess has been solved. But so what? Programs like AlphaZero and Stockfish already play well enough for any purpose.
Anyway, you're free to go implement this ... good luck. I won't respond further.
>the absence of Bellman violations is necessary, but not sufficient
It is sufficient though. All chess game ends in a finite number of moves. If you are consistent aka you have zero violation. Thinking backward (dynamic programming), you can "color" correctly final positions. And you can color correctly all position at 1 turn before end because you are consistent. Then 2 turn before ends,... Then recursively you have correctly colored all chess positions.
You are missing the complexity collapse which can occur in games, like for example the game of Nim, where a simple function can predict the outcome. When you have 15 sticks and you can remove any 3, naively one would think that there are 2 ^ 15 game states and "15 choose 3" legal game moves by turn, but in fact there are equivalence classes, which mean the game state is reduced to 15 different states only.
Modulo the prism of a trained neural network, game states got grouped into equivalence classes and the same phenomenon occur in chess, which allow simplification by high level rules like white wins because white wins the pawn race.
I am more familiar with Stockfish than AlphaZero or LeelaChess0, but the Stockfish demonstrates that well engineered features can bring down the neural network size a lot. In particular counting usually poses problem to neural networks, and counting like how many moves before the 50 moves rule or number of moves before a pawn race are edge cases that can be simplified (DTZ, and DTM).
Also these engines are trying to compress the evaluation function which is a lot more information than just whether the position is win, draw or loss, aka just the frontier.