Chess will never be solved, here's why

Haven't read it. Should I?? I don't as a rule read back to find what I may have missed.

Let me remind you that all was peaceful and respectful here in this thread, until you lot decided on a trolling expedition. So I vamoosed, only to return and find Elroch repeatedly insulting tygxc. I reported it and nothing happened. In fact, the trolling got worse, with the profile called Mega insulting tygxc with every post.

If you are sincere about improving the general behaviour here, don't blame others for the behaviour of yourself and your friends. After reporting bad behaviour on and off for three months and nothing happened, I finally renewed my old correspondence with erik and told him what I think the situation is. I hadn't talked to him for over 8 years.

Link to that post:
https://www.chess.com/forum/view/general/chess-will-never-be-solved-heres-why?page=742#comment-104581067An issue is that a particular person (not I) will go all out to bait other posters and draw criticism and then go all-out on his trolling and personal attacks - feeling he can then counter-report.
And thereby deter criticism of his posts and get a double standard.
For ten years now.
Should we care? No.
But 'being passive' doesn't follow from that though.

it's not learning, it's updating its playstyle to yours, better said adapting its moves, which means it is learning. a move that could have been a blunder can now be a brilliant move, so it is learning.

I find it less interesting than perhaps you do. I'll tell you why, if I may. Often, when we reach firm conclusions regarding methodology, it's the result of a very fundamental bias, regarding the dichotomy, which exists at a fundamental level in our cognitive apparatus, between the inductive and deductive approaches. However, we should know that deduction can only be carried out if a set of "givens" is coherent enough that a logical structure can be constructed from them. The "givens" don't have to be true .... just valid. A lack of truth in the premises affects the truth of the conclusion to some degree at least but it's still a valid conclusion. It just might not be right.

Somewhere along the way, mathematicians have invented axioms which are assumed correct and act as the fundamental givens when a set of premises is investigated and assessed. Unfortunately, whilst some of these axioms are basic and obviously reasonable, in the world of mathematics there's been a tendency to invent artificial axioms which supplement the basic ones and which are use to give incontrovertible finality to the pronouncements of some mathematicians, even if those pronouncements are suspect of even wrong. It's as if the mathematicians have forgotten that philosophers exist to add a counterbalance and if necessary to criticise mathematical thinking when it gets a bit too carried away with its own glory.

To put that into context, there was a discussion regarding the odds against 114 pairs of errors being even. I know that I showed Elroch's judgement to be biassed towards accepting as true that which he wanted to accept as true, without applying to it the criticism and scepticism which he applies to arguments with which he disagrees. He hasn't responded to my refutation, so far as I know. Dula standards aren't a good way to approach these discussions. Wanting to win rather than being open to likely truths isn't good, either.

In my opinion, tygxc's argument about error pairs wasn't refuted. A refutation would have depended on an understanding of error distribution in typical, real games. But that understanding doesn't exist, since it cannot be achieved until chess is solved. This discussion is about "trying to solve chess" and the refutation of tygxc's argument depended on assumptive thinking .... that chess is solved and we understand the error distribution.

So far, the best evidence we have is the 114 games. I'm not arguing that the evidence is sufficiently compelling for us to be deductively sure that chess is a draw: just that it's the best evidence we have, so far. I have pointed out that it is necessary to look at much longer games using different strategies before we can be completely sure and even then, the certainty is more of an inductive nature. But then, most of the foundations of deductive reasoning are inductive too.

You misunderstand how its analysis works. It wouldn't give my moves a higher score than its own if its own were better so it isn't about playstyle. These engines are fast but the reason we can beat a bot that's rated at 2000 is that we find a deeper line than it looks at, which if it was set to look one move further, it would see. We don't win by short term tactics but by longer term projects where we promote a pawn.

So it isn't a big thing at all but Dio didn't bother to compliment me on good play which he certainly couldn't replicate. He just tried to find an error in my thinking and it turns out that he was the one who made the error. He always does actually.

So the engine doesn't "adopt my playing style". It simply holds my move in its cache so it can find it next time. Like I said, it isn't true AI because true AI doesn't exist yet.

I'll respond to this part as for the rest I'm either lacking the context or agree on the general stuff.

The hypothesis of Tygxc more or less can't be refuted, as you said, we are lacking the data. Some of it I agree with, some of it I don't as I've reasoned in my previous post.

The methods Tygxc uses such as his probability calculations and poisson distribution, he often applies wrong, which leads to some understandable false confidence. Now I wouldn't dump everything he says in the trash because of some miscalculation, because there's also evidence that supports some of the stuff he is saying.

Someone will think I'm even being too nice perhaps but atleast Tygxc keeps the conversation going which leads to interesting stuff sometimes.

tygxc's claims of 'proof' can be refuted and have been and are.
Engines of the same strength playing each other and continuing to draw each other does not prove those engines are playing perfect games.
Its as simple as that.
Refutation in two lines.
Arguing that they are playing perfect games because they keep drawing each other is a circular and invalid argument which also ignores the context of robotically programmed artificial intelligence.
Even resulting in engines not being able to recognize draws that human players can quickly see are obvious draws.
The engines have a 'horizon' they cannot see beyond.
They can only evalutate to that horizon.
---------------
Related to this:
I suggested an algorithm whereby the engines could quickly assign a win to positions with very lopsided material situations for one side that is on move.
For example - several pieces or pawns against a Long King.
WIthout having to 'brute force it out' every time.
Elroch indicated its impossible for a computer to think that way.
He appears to be right.
This kind of thing is related to the 114 draws simply being an aspect of artificial intelligence playing itself.
It suggests that the errors they make are 'deep enough' that they cannot see far enough to refute such play.
-------------
Also related to this:
a top ranked engine playing itself with five days per move.
Does it draw?
Does it make errors?
When it draws itself - it doesn't follow it made no errors.

It is true that engines change their evaluations and their preferred moves over time. It is true that sometimes this means they will change their preference to the first preference of a player who is, say 1500 points weaker, as @Optimissed points out. It's worth emphasising that more of the time, they do not. Nor does it mean that weaker players have better insight. It's better understood as an aspect of the random variation.

There's no 'magic' in the latest engines taking five days per move to draw each other.
They could get 10,000 consecutive draws against each other and it would only prove something about their limitations.
There is absolutely no proof they are playing 'perfect' games.
With artificial intelligence like that - how different is that from each of those engines playing itself?

I did a search just now 'what happens when the strongest chess computer plays itself?'
Right away the very first hit was ...
"With any engine, you will have a bunch of draws but you will also have wins and losses too. The "horizon effect" (not to mention the time constraints) will prevent the computer from always playing the best move. So it will make a move and "both sides" will evaluate it the same."
Is that correct?
It implies that the engine isn't perfect.
That indicates its games against itself aren't perfect either.
I intentionally did not go to the site that google entry came from.
Reason: evaluate the entry per se.
-----------------
there's a flaw in that internet quote.
I wonder if anybody spots it.
After a move is played - its evaluation may change because 'the same side' now has a different position in front of it that has advanced by one move - or ply.
Which would mean that 'both sides' do Not 'evaluate it the same' ...
----------------------
Now try out what would happen if a very strong engine is given a very long time to move ...
with each additional move of 'look-ahead' the possibilities multiply tremendously ...
so there's a paradox.
Does giving the computer extra time 'expand its horizon'?
Apparently. But the rate of lookahead is going to fall off exponentially - if not worse than that - the further ahead the computer gets to look.
The more time - the more lookahead - the more the computer is 'reaching' its horizon.
At which point it struggles.
As seen with the tablebase projects.
The computer's 'horizon' simply can't deal with eight pieces on board and the 'lookahead' problems that causes.
-----------------------------
It should not take long to figure out that engines getting five days per move that have positions in front of them with far more than seven pieces on board - are not 'solving'. They can't.
No Einstein stuff there.

@14899

"3. ICCF finalists with 5 days of avg time for turn are strong enough to always draw against future engines." ++ This is also what the late 3-times ICCF World Champion Dronov implied with 'It is necessary to radically reduce the time for thinking.'

"we know chess games are much deeper than todays players are able to process"
++ Chess is not as deep as people think. 10^38 positions is a huge number, but 10^38 = 3^80, i.e. 40 moves with average 3 non-transposing choices per move generate all of chess.

"This factually leaves room for error" ++ An average ICCF WC Finals game ends in a draw in 40 moves. You only have 40 moves to go wrong.

@14905

"What we could examine is errors made by engines in tablebase positions."
++ I proposed a relevant position:

Black to play and draw.
It is a 7-men endgame table base position. As it is a draw, it could result from the initial position with optimal play from both sides. Humans in endgame books had this wrong and thought it a white win, so it is not trivial. As is is a rook ending, that is the most common type of endgame.

I’m on the iPad app right now, so it’s hard to quote posts, but I found myself wondering if tygxc was correct in suggesting that ICCF WC decisive games have been linearly decreasing over the years. As this sounds like a compelling trajectory …

So here’s what I found:

Decisive ICCF WC games per event (ignoring losses from a player who sadly passed away during the event)-

2011: 16
2013: 20
2015: 13
2017: 9
2019: 15
2020: 17
2022: not yet completed, with 22 games ongoing

I’m not sure what we can conclude from this data …

In any case, it’s already been pointed out that ICCF WC draws can still contain mistakes. Therefore, pointing at the number of draws per event doesn’t seem to prove much, as such draws have been shown to be unreliable measures of quality chess - especially when evaluated by future engines …

Well, the bot I was using was called Li, rated at 2000. Very strong tactically and prone to a piece sacrifice to get a lot of pressure. Apt to be a little too aggressive. Over the past couple of days I spent two hours working my way through all the Roman bots and then Li. It isn't what you would call strong and being able to play positionally, I found it was really good practice. You have to play aggressively in a very muted way, giving precedence to keeping your pieces well-placed and your formations flexible and strong. It fails to tactical-positional play.

In the game I posted, the bot's first preferences were identical to a high proportion of my chosen moves in that game and there were about five cases of my moves being better than the bot's, which isn't surprising because I'm a stronger player. I made three sub-par moves but I was never losing, as you can see from the game score. My opponent played far better than a player rated at about 1710 FIDE and then the rather quick time control, 80 mins plus 15 seconds, and my exchange sacrifice, which was winning, got the better of him. I was way behind on time for most of the game because I was really too tired to play and had problems with my glasses. This is my second rated game after 5 years of not playing rated otb. My last game, 5 years ago, was black against Brett Lund, FIDE 2350 and an old friend. He told me it was the hardest game he had to play all year although he did win from a losing position in the end, because I got into time trouble.

Anyway this was not about random variation, unless you're implying that the bot chooses a weaker move sometimes at random. It may be possible but then it would vary back again if you ran the game through again.

Just checked ... Brett's down to 2292 now. Around number 4000 in the world apparently. He's about 62 and in 11 years he'll find out what it's like to be 73.

Huh?

1. Why are you insulting him? Please be civil, like he is.

2. What do you define as “true AI”?

3. It doesn’t hold cache of your moves, what are you saying?

4. He doesn’t have to complement you, we’re talking about how chess will never be solved.

5. We can win against a 2000 bot by short term tactics as well, here:

Do you have the game? How did you know you were winning at first?

@14914

"Decisive ICCF WC games per event"
WC33: 22 ongoing + 10 GM Dronov deceased but in otherwise drawn positions + 57 draw agreed + 37 3-fold repetition + 10 7-men endgame table base draw claimed = 136 total
WC32: 4 decisive + 13 resigned by SIM Bock for personal reasons + 119 draws = 136 total
WC31: 15 decisive games + 111 draws = 136 total
WC30: 9 decisive games + 127 draws = 136 total
WC29: 12 decisive games + 124 draws = 136 total
WC28: 20 decisive games + 116 draws = 136 total
WC27: 16 decisive games + 120 draws = 136 total
WC26: 24 decisive games + 112 draws = 136 total
WC25: 32 decisive games + 16 cancelled + 88 draws = 136 total
WC24: 35 decisive games + 101 draws = 136 total

The number of decisive games goes down every year.
In previous years there were decisive games and thus also a few draws with a pair of error.
This year we have no decisive game, thus all 114 draws are perfect games with optimal play from both sides.

The thing is, we don’t know if the computers ACTUALLY played perfectly, we just know they are the SAME STRENGTH, since they drew.