Peak Chess

[swami]

The discussions about singularity and peak AI in the AI forum have prompted me to run a poll on peak chess. How will Elo improvement in chess end?

If Chess is not solved, meaning there is not an entity that exists, that almost never fails to solve a puzzle (or loses a single game against itself or the next strongest entity) then #4 is more likely.

[towforce]

I'm surprised so many vote for the first option. I don't think humanity will continue to exist forever.

Hmmmm... any possibility you might be overthinking this a tad?

No. No such possibility.

Notice to all persons working on peak chess: your deadline is 500 million years. At this time, the expansion of the sun will heat up earth to a level which will make it uninhabitable (the end of its "Goldilocks" era).

Given my track record, I'll probably start in 499999999 years and 360 days then.

[Jouni]

It's practically ended already. This is CCRL blitz list:

Stockfish 17 64-bit 8CPU 3812
Stockfish 16 64-bit 8CPU 3810
Stockfish 16.1 64-bit 8CPU 3808
Stockfish 15 64-bit 8CPU 3806
Stockfish 15.1 64-bit 8CPU 3803

[swami]

Are you saying that the top engine can solve all puzzles effortlessly? I remember that there are just too many puzzles it failed to solve even at higher depth, including the one about endgame related to Qa4 sacrifice posted recently.

[towforce]

It's practically ended already. This is CCRL blitz list:

Stockfish 17 64-bit 8CPU 3812
Stockfish 16 64-bit 8CPU 3810
Stockfish 16.1 64-bit 8CPU 3808
Stockfish 15 64-bit 8CPU 3806
Stockfish 15.1 64-bit 8CPU 3803

Maybe there are still one or two more innovations before the "game over" indicator lights up.

1. Kellin Pelrine, a middling Go player, was able to beat a top Go program when he identified an astonishing weakness: it didn't understand a simple basic concept in the game - a group of Stones! Link.

2. A God algorithm would end it more finally than it is at the moment

3. A good, quick algorithm combined with a strong drive to create complexity. Complexity probably hasn't been investigated as much as it could have been as a strategy - but it would be my choice: it creates situations the engines probably aren't familiar with and, more importantly, it absolutely throttles search depth by massively increasing the branching factor. "Selective search" goes out the window.

[Jouni]

Maybe engine can be perfect without solving all puzzles? It simple wins/draws all games before these positions happen!

[swami]

Even in CCRL page, Stockfish versions did lose few games. There's still grounds to cover.

Also, I have noted following observations:
During the live game of World championship, the top engine diverges from its choice after a new move (happened in game) is introduced and it begins to choose the new move and evaluates it favourably.

Time taken by the top engine need to reduce with less hardware speed going forward. There's a scope for improvement on that front.

Puzzle is also equally important. Failure to solve the puzzle by treating it as a draw, when win is clearly possible, is not acceptable.

[towforce]

Puzzle is also equally important. Failure to solve the puzzle by treating it as a draw, when win is clearly possible, is not acceptable.

Agreed. It will be disappointing if the best ever chess engine turns out to be unable to solve puzzle positions quickly: this would mean that the underlying secrets of chess will never be uncovered.

The history of mathematics is that secrets do get uncovered eventually.

[towforce]

The poll is now closed, and the majority of voters believe that things will trundle along as they are right now into perpetuity. This strikes me as pessimistic: of course we're going to unlock chess's big secrets - and quite a bit sooner than most people are expecting. Were you ready for the proof of Fermat's last theorem in 1993?

But if it is to be incremental progress, as NNs get bigger and training methods improve, then one day we'll wake up and find we have a chess program that never loses. Its NN might contain the secret pattern that solves the game - but because it will be buried deep inside a bloody huge set of weights, we will never know!