Chess will never be solved, here's why

Have you run any races in the last 40 years? Want to boast about your speed?

Are you aware that IQ (irrelevant as it is after childhood) falls with high ages?

@Optimissed is a child in almost all respects.

Diog, I was going by that rating system description. Assuming every 400-500 points is 10x better than the previous player, and the current engines that can calculate billions of positions per second, the "perfect game solving computer" with the 10^200+ possible games, would have to be magnitudes higher in rating. It wasn't just a descriptively high number.

Totally missed my further clarification. I didn't say larger and larger tablebases would be the efficient route to a solve, nor that opening lines can be universally solved 30 moves deep. I said it might be a combination of the two, rather than all of one vs all of the other being completed.

Yes, your point is made, but a mistake to add a perfect estimate because we have no idea what it might be.

I once had a thread about the question of "could a 4000 just as easily beat a 3000 the same way a 3000 could easily beat a 2000". The surprising answer is that yes, because if they couldn't, the rating of 4,000 wouldn't be reachable in the first place. It's counterintuitive to realize these numbers don't just exist in a vacuum. A 4,000 would have to be over 100x "better" than a 3,000 to achieve that rating in the first place. But here's where it gets funky. To increase ones rating one has to keep winning, but if at a certain rating both players know the perfect moves despite one being 1000 points higher than the other, the 4000 won't win, and drop back down from draws. These engine ratings are compared to human performance, not their own, with the exception of TCEC which actually matches engines of varying "ability". There's no way to know what would happen matching supercomputer stockfish to super-supercomputer stockfish, it may just all be draws at that point regardless of one being 1,000x more powerful.

In other words EE said both and is essentially wrong on both - especially the second one.
When talking about solving from the 'front' of the game then the collission is with the Shannon number or numbers related to it - instead of with the number formed by John Tromp which is what the tablebases are colliding with.
The John Tromp number is already more than prohibitive enough.
But for people thinking that chess would be solved from the front ... then you're not just talking about trillions of trillions of trillions of years as with the JTromp number ... you're talking about the JTromp number cubed.
Or more.
----------------------------------
Just now - had a conversation with AI about the Shannon number.
This next is not 'AI generated' (wouldn't matter if it was because there's no rule against it) - its rather 'AI assisted' which is something else.
After a lot of adjustments I got the AI to say this:
"The Shannon number is an estimated lower bound on the number of possible chess games, introduced by Claude Shannon in 1950. Shannon apparently arbitrated a maximum game length of 40 full moves (80 plies) and apparently also arbitrated an average of 30 moves per position.
And then Shannon came up with 10^120 as a lower bound on the gigantic number of possible chess games."
---------------------------------------
Point 1) if anyone thinks that's inaccurate they could say - with reasons.
Point 2) the nature of the Shannon number including the fact that its so very very much bigger than the JTromp number is why 'solving from the front' is so bogus and nonsensical. The task is already daunting and would take trillions of trillions of years with todays technology. But trying to solve from the opening position would cube the difficulty. In other words multiply it by itself twice.

'at that point' many trillions of years from now.
Chess isn't solved.
Result: the 4000 beats the 3000.
5000 will beat 4000 ... and so on.
But it doesn't take a 1000 point difference.
Even a 200 point difference is big.
The stronger computer doesn't need to totally crush the weaker one to establish the point. The point is whether wins are present or not.
---------------------
Idea - pair two computers together that are the same strength.
Have them play each other.
Is the incidence of draws higher than the incidence of draws between two GMs of the same strength?
I think you'd find it is.
But it could be looked into.
Two 2500 computers playing each other (some might not see that each computer might as well be playing itself ?) ...
-------------------------
relevant: are the computers well under top strength computers because they've been 'handicapped'? Or is their processing speed lower?
also relevant: the very top computers would not have similiar programming to each other? Similiar enough? Result: very high incidence of draws - depending on the time controls.
Try giving them one minute each for the whole game and see what happens.
Or ten seconds each. They can move very fast. Could probably play a 100 move game in that time.

You can't set a move limit on estimating the number of games. It might be possible to have a chess game with that many moves if the 50 move rule isn't taken into account. The number can be narrowed down somewhat doing forward analysis by omitting obviously flawed games, such as this one:

Think of how many absurd variations of nonsense like this exists. All of which normal chess programs could weed out. The crazier the game, the more forced mates or mates in 5 or less are going to appear, all of which can be detected easily, and can be pruned accordingly (of course trillions of exceptions are going to exist too though lol)

from EE just now another mistake by him:
"You can't set a move limit on estimating the number of games. "
EE not grasping the notion of 'lower bound' and the purpose of arbitration.
He doesn't get it that Shannon's work is famous because he's establishing that the number of possible games is More than that.
EE missed the point completely.
EE began with an invalid premise (and strawman) so I skipped reading the rest of his post.
Better posters begin with 'Valid' ...

It still doesn't work that way.

To reach X rating, you need to have a pool of other players that are within several hundred ratings points of the rating you are trying to reach. If Carlsen started playing within a fixed pool of 500 rated players for 10 years and never ever left it, his effective rating ceiling would be 1,200 or so, and he would be gaining a smaller and smaller fraction of 1 rating point for wins thereafter. His playing skill matters little at the highest end of the pool, it just maintains his position without extending it, thus his difficulties in trying to reach 2900. Once his opponent pool dropped back under 2800, 2900 became infeasible for him to achieve. His rating is completely relative to (and defined by) his pool of opponents.

Systems that fudge this by, say, allowing 1 full rating point regardless of rating difference are not valid Elo or Glicko ratings. The diminishing returns are crucial to avoid inflation/deflation by creating ratings points out of thin air (ala chess.com ratings and thehir artificial boosts, etc.).

Yes and I clarified that in my other post, a 5000 rating may be impossible for that reason.

Shannon's premise is inaccurate as literally high level grandmaster games have lasted 150-200+ moves in the past before. Without even the nonsense random move games, top level games have naturally lasted way longer than 40, not sure where he got that number from. And it definitely couldn't have been referring to the number of "realistic" chess games as then it would be far too high. "Nonsense" games can likely go on for 10^50+ moves without the 50 move rule. It would be interesting to calculate that, the longest possible game without even 2-fold repetition. 64 squares, 32 pieces, pawns only inching 1 square at a time. Note that the same moves and positions can be different games due to reversed order. D4, d5, e4, e5 is different from e4, e5, d4, d5, which is also different from d4 e5 e4 d5..etc.

AI will solve chess in the near future, that's for sure

Stockfish's rating is about 3650 now, so effectively, the rating ceiling would be about 4350 right now, but you would have to beat Stockfish every single time over hundreds of games. One draw would plummet the rating.

The various pools will continue to attenuate over time. It's not a fully closed system, so new players introduce ratings points that get thinly spread across the whole ratings pool. Players that stop playing do not cause points to be removed, so, 5,000 could be theoretically possible in a number of decades or next century. A million? Not too likely unless you artificially produced it in a way that any observer would agree was "cheating" the ratings system, like playing endless bullet games with a server farm of obsolete engines vs. one top engine, and then "reintroducing" new engine "players" to inject rating points into the system.

Nope.

Yes it will, but stay entitled to your opinion, I will stay entitled to mine.

You are definitely entitled to your opinion, it's just not going to pan out for you.

It's not going to pan out for you, too. I believe in the "impossible" and you seem like you don't.

The same argument sans graph was made to Tygxc many, many times. It probably would not have helped anyway given his need to exonerate Sveshnikov posthumously...but it's nice to see someone ran the data and graphed it.