LeelaZero's billion parameters for evaluating everything about a position (trivially including the material and anything you might include in "all other factors" (plus a million times more) provides it with enormously more testable understanding about this but does not provide it with certainty. A passable human player like yourself being certain about this is an example of your poorer judgement versus an AI that is over 1000 points stronger.
Just look at this nonsense. I've already explained to him that non-certainty is built into a machine like Leela. It can't do otherwise. Doesn't take a blind bit of notice.
Likewise Bayesian reasoning is the provably only fully consistent way of quantifying belief by reasoning from the specific to the general (inductive reasoning), and no amount of evidence can ever reduce a finite amount of uncertainty to zero uncertainty by Bayesian inference.
This is a (meta)fact about knowledge about the real world (such as all science) and also applies to questions that are in principle possible to decide by exhaustive analysis but presently impractical to do so. (I hope it is obvious that solving chess falls into that category).
The reason that AIs like Leela are designed to quantify uncertainty and not to ignore it is that that is appropriate.
@5170
"This method of solving chess relies on using the judgement of GMs or engines"
++ No, it does not rely on the judgement of engines, it relies on the ability of the engines to calculate until the 7-men endgame table base or a prior 3-fold repetition.
No, it does not rely on the judgement of GMs. The GMs reduce the computation to relevant width and depth. The proof of the Four Color Theorem did not involve coloring all maps, only a humanly determined relevant subset.
This is a misleading statement. The Four Colour Theorem involved proving that there was a way to colour ANY map with four colours. Not just a humanly determined relevant subset.
The proof fell into two parts. The first part was that if there was a map that could not be four coloured, either there was a smaller map that could not be four coloured or the map was in a specific finite set of maps (the original list had 1,834 maps and later version had 1,482). The second part of the proof was to show that every one of the maps in the finite set was four colourable. This was the part involving heavy computation. Doing so completed a reductio ad absurdum proof, which can be converted into a purely deductive proof.
The human agency was only in defining the structure of the proof. All of the steps were mechanisable deductive steps with no shortcuts.
"any engines used would have been surpassed by new developments"
++ The newer engines can complete the same task faster.
A newer Stockfish released during the 5 years of the task can be switched to.
A released 8-men table base can be used, but does not change much.
"casting doubt on the entire process." ++ No. Present computers are more powerful than those in 1976. That casts no doubt on the proof of the Four Color Theorem. Newer computers cast no doubt on the solutions of Losing Chess, Checkers, Connect Four, or Nine Men's Morris either.
"only a brute-force computation of all possibilities can be entirely reliable"
++ It is pointless to compute all possibilities of say 1 e4 e5 2 Ba6? until checkmate.
We know the outcome for sure: white loses. What would be the point of this computation?
You need to at least acknowledge that you think you know the outcome that white loses and that those with better understanding point out that your certainty is pragmatically reasonable as a chess player playing the odds, but inadequate for a proof.