> These computer AI chess programs have weaknesses.

> ... have collected statistics on which openings will lead to situations where Stockfish (or LeelaZero, or Komodo, etc. etc) will inevitably fail.

In case of Stockfish and other "conventional" engines, yes.

However, with engines like LeelaZero those weaknesses gets much harder to exploit, since you can create your own version of Lc0 by training a new model with tensorflow.

LeelaZero struggles with endgames. I'd probably use LeelaZero through the beginning and middle games, but switch to Stockfish whenever I feel like a tactical situation comes up, especially if the position starts to drop down to 10-man or 9-man. 7-man Syzygy Tablebase are less than 10TB, a RAID-array of M.2 drives will access those quickly. If Stockfish evaluates any 8-man trade, it will simplify into a 7-man situation and you have guaranteed perfection after that.

No one knows how to integrate the endgame tablebases into LeelaZero by the way. So if we both play LeelaZero for the first 40 moves, but I switch to Stockfish + 7-man Tablebase for the ending, I'm probably going to beat you. LeelaZero's endgame really is quite weak.

---------

If you are a programmer, and can figure out how to integrate the Neural-Net evaluation into MCTS to evaluate tablebases, you probably would do great. Its currently unsolved (to my knowledge).

It seems solvable... I don't see any reason why MCTS couldn't have a if(popcnt(bitboard) < 8) doTablebase() statement somewhere. The question is how to do it as quickly as possible, and how to do it asynchronously and integrate the statistics to the rest of the MCTS tree.

BTW: A research team has created the 8-man Tablebase, although it takes probably 100s of TB to store. This means that 8-man Tablebases will remain as Hard-drive only for now.

I like the opposite: only with physical presence, no consult, no studying openings just you playing your best game against your opponent doing the same according to the classic time rules.

It comes down to what do you value? Do you value playing "your" best game vs "your opponent's" best game? Or do you value playing the best game possible, using whatever tools necessary to do so?

This philisophical split is also in the speedrunning community. The tool-assist players aim for absolute perfection: what would a "perfect run" actually look like? They make recordings at 1/60th of a second, and have a computer input the sequence of buttons with absolute perfect accuracy.

Or, are you a human-only player, who values skill and training?

There's something to be said about the two sides. But I find myself preferring the "theoretical perfection" side... as just my personal taste. I can see why people prefer the skill-and-hard work approach though.

Sorry for the belated response, I've been thinking about this for a bit before answering because I think it is a subtle matter. I think the reason why I don't go for this 'theoretical perfection' angle is that it takes the fun out of the game. That's where computer programs, cold and emotional will not feel the joy of a win, their programmers might but they might just as well not even be aware of the game being played. The only way to get that feeling on both sides is for unaided humans playing other unaided humans.

The 'perfect game' is something that a computer will one day find, a curiosity but just like a math proof ultimately just a thing waiting to be discovered, not achieved.

I forget who this is attributed to, but someone proposed a method of beating chess - correspondence chess that is. Simply set up a chessboard, mail a grandmaster for white, and a grandmaster for black, then have them play against each other.

I remember this variation of the story:

An unknown chessplayer bet a lot of money, that he could play live & simultaniously against two grandmasters and achieve at least 1 point. The two matches were played in different rooms next to each other. No one except the unknown player was allowed to switch rooms during the matches (to not disturb the grandmasters). Obviously he played White against one grandmaster, Black against the other, and won his bet.

As I sometimes tell this anecdote, I wonder what its actual source might be?

I'm pretty confident, this never happend in real, as I expect no grandmaster could get scammed that easily. But the story could be rooted in a "catch me if you can"-like movie or novel or something.

If someone knows more, I would be happy for a hint...

It's a very old trick I'm sure. As for it never happening or fooling anyone in real life: the magician/mentalist Derren Brown challenged 9 English chess players, including 4 GMs, to a simul on his TV show, and got a plus score - he secretly had 8 of the players playing each other, with whom he thus had an even score. The 9th was president of a university chess club, and a very weak player, who Derren played and beat, winning the match.

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

Might not be exactly what you're thinking of, but Derren Brown did similar against a group of players

https://en.chessbase.com/post/derren-brown-s-che-trick-once-...