Chess will never be solved, here's why

I'm not. ...

@MARattigan
"Just post your own relevant views as you have time. There is no obligatory minimum frequency of posts you must comply with in the forum."
Lol. No kidding?
I was referring to direct conversations with AI.
Getting logical evidence from AI while presenting your own logical evidence to it.
AI does sometimes disagree with whoever posting to it - but does so very diplomatically.
You can qualify why you disagree - present it.
The AI will gush approval - but you can ignore that.
Point: it is responsive to logical evidence.
For example if it says a certain mathematician used another mathematician's work - you can sometimes point out that the second mathematician wasn't born till after the death of the first!
Logic in action!

So long as the source for your dates is not an AI.

Talking of AI, I have had some interesting experiences related to chess. I asked AIs to find sets of good exemplar games including PGNs in chosen opening variations . At first I believed them (typical example would be referring to the 1992 Fischer Spassky match game 3 and then giving the full score of the game),

As well as some red flags in the quality of play, in one of them I found a plain illegal PGN. One player left their king in check.

This is when I found that there was a lot of hallucination going on. The latest AIs are quite good (but not perfect) at generating plausible looking (but not very high quality) games, with occasional plain illegality, but they are often not real games at all, despite their descriptions. They even lie about games from matches as prominent as that Fischer-Spassky one.

Oh well...

Yes. I should have said, "... not another AI".

How would you rate their current intelligence compared with your own?

I'm actually very impressed by AIs. I have a distaste, not for AIs, rather misuse of the same by humans (notably assertions that AI routines in Chess programs guarantee theoretically accurate predictions of game results and that the output from LLMs are authoritative sources for any topic).

Certainly a quote is different from an assertion, but assertions which are uncritically based on AI output are worse than quotes of the same because the true source is not apparent.

O worries about things like that. Plus he doesn't know how to discuss intelligence properly.
Are you catching his bug Martin?
I think not.
Were you and 'Dubro' at odds?
Was he touting Grok AI heavily and posting massive quotes of Grok sessions?
Obviously one should keep the AI quotes short enough.
--------------------

@playerafar

Some might also find it very tedious.

The first introduction of mandatory game termination was in the 2017 FIDE handbook. Prior to that all draws had to be optionally claimed.

All versions prior to 2017 allowed for infinite games. Basic rules chess still does because the 5R/75M rules are excluded (and, though not required for my assertion, the 3R/50M rules were dropped) from the basic rules in the same edition.

The only generally accepted versions of chess where unlimited games are disallowed is FIDE competition rules chess post 2017 and versions of chess directly based on that game.

You guys no nothing about computer chess.

But give them their due; they no how to spel, bless 'em.

This is another bad post by both of you.

"Oh yes if its done it does solve chess. If a single position (like fortress or zugzwang)"

Zugzwang was a problem because chess computers moved away from a full width search, a type 1 search.. And would prune out the relevant lines. So the type 2 chess engine would not detect the Zugzwang. That did not solve chess. It helped solve some of the pruning issues with type 2 chess engines. In the SEARCH of the chess game tree.

Moving from type 1 to type 2 helps pruning issues because type 2 prunes selectively whereas type 1 prunes anything it hasn't managed to get to in the time available.

A zugzwang is a feature of a diagram not a position, it is not necessary to detect it once you arrive at it, though it can be helpful in organising your moves in advance to have correctly evaluated the diagram with both sides to play. Programs with a type 2 search are generally much better at that than those with a type 1 search because they can look further ahead to the positions with zugzwang diagrams and beyond to better evaluations in the same amount of time, but that's just true in general. 

Neither type of program has helped solve chess. Nor is that what type 2 programs are written for (or, realistically, type 1 - they may be designed to solve shortish mates in reasonable time). 

If the starting position is not a short win or there are (desperately unlikely) two short draws from it then neither kind of program can solve anything that might be called "chess" in any time that could be regarded as reasonable in the absence of purely science fiction improvements in computing speed. 

And this was the same type of issue with fortress positions. The type 2 chess engines could search very deeply by pruning the chess game tree, but did not know what a fortress position was, and would prune out the relevant lines again in the SEARCH of the chess game tree.

Sometimes true sometimes not, but type 2 engines will generally be less prone to pruning out relevant lines than type 1 in a given time, because they're more selective in the lines they prune.

This helped the type 2 chess engines like Stockfish, but did not solve the problem with the search completely. As there are many types of fortress positions. And the Neural Nets have learned to recognize some of these positions.

None of these techniques solved chess or could solve chess. Again it just helped in the search of the chess game tree. So the type 2 chess engines had a chance to spot these types of issues. And not prune them out of the search of the chess game tree.

Since chess is a 100% tactical game, the only way to solve chess. Is a full width search of the whole game tree of chess.

Which is obviously complete and utter garbage.

The most usual meaning of "game tree" is the tree recursively generated by adding nodes corresponding with legal moves and starting with a node representing the initial chess position. If there are no iM or jR rules limiting the length of a game (as is the case in some versions of chess) the game tree has infinite depth and a full width search of the whole tree would not complete in finite time. If you're implying such a version of chess has no solution in principle, you're in the wrong thread.

It's total garbage even if iM or jR rules limiting the length of a game are included in your term "chess".

A single forced mate (i.e. just one move for the winning player at each of his nodes together with all opponent responses and with all leaf positions mate for the winning player) is enough to solve chess. The whole of the rest of the tree can be thrown away if you can find such. If chess is a draw then you don't need to look any further along any line in searching for a drawing strategy if you find a single forced mate at any node. The node and all nodes in continuations that have not yet been considered can also be thrown away. Further if you have a jR rule the continuation tree from any node where a single repetition occurs can be thrown away in its entirety.

If you consider the chess like game where the following position replaces the starting position:

SF17 finds a forced mate as described above after considering 1,869K nodes and thus solves that game. It's assuming 50/75 and 3/R rules, so the number of nodes in its game tree is finite, but difficult to calculate. It is however vast and the number of nodes actually searched is a very miniscule fraction of the total.

I do rather better than SF17. I can solve that game without searching any of the game tree at all.

As for whether a type 1 or type 2 search is more likely to solve chess ever, the answer is unknown, but heavily in favour of type 2. Unless the starting position is a very short mate, quantum computing becomes a practical reality and it follows a trajectory similar to that of conventional computing for the last 50 years, a type 1 search probably won't. Something like Stockfish using a type 2 search (however that could be implemented on a quantum computer) could, conceivably, under the same assumptions, but perhaps with a mate a few ply longer. It's not known whether the starting position is a win and we have no estimate of how long such a win might take should it exist (but it would probably be over 10 moves because nobody as far as I know has come up with a mate in 10 or less position that SF17 can't solve in an hour on a fast machine). 

Here is example of these new feature using the type 1 full width search in new chess engines, along with the type 2 search to analyze chess positions.

Advanced Chess Analyzer

Advanced analysis options, highly recommended for CC play

Full depth threads

Integer, Default: 0, Min: 0, Max: 512 The number of settled threads to use for a full depth brute force search. If the number is greater than threads number, all threads are for full depth brute force search.

This approach is now catching some of the errors made by the AI type 2 search of the chess game tree. In programs like Stockfish, and other type 2 search only chess engines.

Can you give examples of the "errors".

If your Advanced Chess Analyzer will do a full width search at consecutive depths (possibly that's not an option), see how many nodes it searches to solve the above game and how the time on your threadripper compares to 4s on my noddy desktop.

Or for a more accurate comparison here are 5 random White to mate in 38 positions in KNNvKP, each of which SF17 will probably solve in under 5m on your threadripper (solve means any fully analysed mate - not necessarily M38). Try them with a full width consecutive depth search (again if that's an option) and with SF17 and see which finds a fully analysed mate faster in each case. In case either fails in say 20m (or however long you're willing to wait) check how many "errors" each finds in the other. Which does better?

(Remember to disable any tablebase paths.)

8/2p5/8/2N5/6k1/8/8/K3N3 b - - 0 1

8/2p5/8/2N5/8/8/8/1N1K1k2 b - - 0 1

2k5/8/2p5/6N1/8/4N3/8/1K6 b - - 0 1

k7/8/3p4/8/1N6/1N6/8/4K3 b - - 0 1

8/4p3/8/6N1/8/8/N7/3K3k b - - 0 1

Even the number of possible unique moves is finite, regardless of whether positions are repeated or not. The number of possible moves, possible positions, possible games, max length of games, are all finite. Now if the rules back then required consecutive repetitions or either moves and/or positions, that's a different story. Then you can just alternate 2 different repetitions endlessly. So not sure what this in between thing is that permit an unlimited game. Even if you combined all these aspects, and said the game only ends when a position, move, move order, and side to move were all exactly repeated multiple times, that's still finite.