Regarding the purpose of AI as it pertains to computer projects -
idea: the AI doesn't process the projects itself - it assists heavily in writing complicated code and then computers execute that code.
Could or does AI execute its own code that it has generated?
I don't know. My first guess is no. My second is rarely.
Chess will never be solved, here's why
Sort:
Feb 17, 2025
0
#14402
DiogenesDue wrote:
OctopusOnSteroids wrote:
Dubrovnik-1950 wrote:
It is absolutely true.
"Mathematical rules can be abstracted from chess positions."
Yes, that is how chess engines have always worked. And work today.
That does not solve chess.
Oh yes if its done it does solve chess. If a single position (like fortress or zugzwang) can be solved without brute force search then its entirely possible any position can. We dont know whats doable.
And by the way engines have been horrible at detecting fortresses in the past, dont know how much better they are now.
My challenge to all the "algorithmic" solvers (to use Optimissed's misusage of the term) is to start proving the premise. Pretend chess is geometry and start building theorems from the ground up. If you can get to one that's completely unknown/new and provable, then you have something.
Well the challenge is as impossible as many other complex problems that AI has helped us solve... It's way too much complexity for it to be worth it for any human to start building and proving theorems. Some simple zugzwang theorem perhaps but thats not what you want is it?Thats what we need AI for that is much more advanced than what it is today.
Feb 17, 2025
0
#14404
DiogenesDue wrote:
OctopusOnSteroids wrote:
Dubrovnik-1950 wrote:
It is absolutely true.
"Mathematical rules can be abstracted from chess positions."
Yes, that is how chess engines have always worked. And work today.
That does not solve chess.
Oh yes if its done it does solve chess. If a single position (like fortress or zugzwang) can be solved without brute force search then its entirely possible any position can. We dont know whats doable.
And by the way engines have been horrible at detecting fortresses in the past, dont know how much better they are now.
My challenge to all the "algorithmic" solvers (to use Optimissed's misusage of the term) is to start proving the premise. Pretend chess is geometry and start building theorems from the ground up. If you can get to one that's completely unknown/new and provable, then you have something.
Which doesn't say it's impossible.
The counter argument is like saying the only way to solve Fermat's last theorem is by calculating a^n+b^n for all combinations of nonzero integers a, b and n>2 and checking if the result is c^n for some integer c. There are obviously far too many integers for this to be practicable even on the fastest current computers, so Fermat's last theorem will never be solved.
MARattigan wrote:
Which doesn't say it's impossible.
The counter argument is like saying the only way to solve Fermat's last theorem is by calculating a^n+b^n for all combinations of nonzero integers a, b and n>2 and checking if the result is c^n for some integer c. There are obviously far too many integers for this to be pracicable even on the fastest current computers, so Fermat's last theorem will never be solved.
No, I am simply saying it's somewhat pointless to discuss a premise where not a single step has been taken towards proving it is viable.
Feb 17, 2025
0
#14406
MARattigan wrote:
DiogenesDue wrote:
OctopusOnSteroids wrote:
Dubrovnik-1950 wrote:
It is absolutely true.
"Mathematical rules can be abstracted from chess positions."
Yes, that is how chess engines have always worked. And work today.
That does not solve chess.
Oh yes if its done it does solve chess. If a single position (like fortress or zugzwang) can be solved without brute force search then its entirely possible any position can. We dont know whats doable.
And by the way engines have been horrible at detecting fortresses in the past, dont know how much better they are now.
My challenge to all the "algorithmic" solvers (to use Optimissed's misusage of the term) is to start proving the premise. Pretend chess is geometry and start building theorems from the ground up. If you can get to one that's completely unknown/new and provable, then you have something.
Which doesn't say it's impossible.
The counter argument is like saying the only way to solve Fermat's last theorem is by calculating a^n+b^n for all combinations of nonzero integers a, b and n>2 and checking if the result is c^n for some integer c. There are obviously far too many integers for this to be pracicable even on the fastest current computers, so Fermat's last theorem will never be solved.
I like that one.
Feb 17, 2025
0
#14407
DiogenesDue wrote:
MARattigan wrote:
Which doesn't say it's impossible.
The counter argument is like saying the only way to solve Fermat's last theorem is by calculating a^n+b^n for all combinations of nonzero integers a, b and n>2 and checking if the result is c^n for some integer c. There are obviously far too many integers for this to be pracicable even on the fastest current computers, so Fermat's last theorem will never be solved.
No, I am simply saying it's somewhat pointless to discuss a premise were not a single step has been taken towards proving it is viable.
But if everyone takes the same view, not a single step will ever be taken towards proving it is viable.
Feb 17, 2025
0
#14408
Dubrovnik-1950 wrote:
playerafar wrote:
Regarding the purpose of AI as it pertains to computer projects -
idea: the AI doesn't process the projects itself - it assists heavily in writing complicated code and then computers execute that code.
Could or does AI execute its own code that it has generated?
I don't know. My first guess is no. My second is rarely.
The answer is yes, Stockfish evaluation functions is 100% generated by the AI. The only thing it was taught was the rules of chess.
That is why Alphazero, and Leela Chess Zero. Are named Zero. Everything was machined learned, and generated by the self teaching AI. (Stockfish uses Lc0 AI learned Evaluation)
"Mathematical rules can be abstracted from chess positions."
But these mathematical rules, either generated by a AI, or code written by a human. Can not solve chess.
That is why all chess engines either being AI generated, or code written by a human. Still have to do a search of the game tree of chess.
Because any evaluation code is just an approximation of the true evaluation. And a search gets you closer to the true evaluation of the position up to the effective search horizon.
If we had a code that could do what these people suggest. No search of the game tree would ever be needed.
But here is the problem. Chess is a 100% tactical game. And can not be solved with just some mathematical rules. That are just an approximation, or also know as a guess. Of the real true evaluation of the position.
And there are only 3 true evaluations of any chess position. White wins, Black wins, or draw with perfect play.
Does that 'yes' answer mean that Stockfish is 'AI'?
AI in the general sense has been around for a long time.
One could say any operating computer and its software is 'AI'.
So Stockfish software inside a computer is 'AI' in that sense.
But what about AI in the more recent sense?
Things like chatgpt and copilot?
Stockfish is in that group?
You can have a conversation with Stockfish like you would with those AI's?
You can verbally tell Stockfish to set up programs?
I guess I'll see the reply next time.
Feb 17, 2025
0
#14409
perfect play with regard to win or draw can't be properly defined because chess isn't solved.
Nor is there evidence that chess is a draw.
Indicators are not evidence in such a math-related and well defined context like proof of a draw. It would be like claiming that since a 2500 beat a 2600 then he should be a 2700. Its an invalid extrapolation.
Does the claimer of a draw not know that?
upon return - prediction is he'll have started talking about credentials or switching majors or whatever. Some kind of personalization or other mistake.
Feb 17, 2025
0
#14410
playerafar wrote:
perfect play with regard to win or draw can't be properly defined because chess isn't solved. ...
Of course it can be defined. You just don't know what actions fit the definition.
You can define a prime number without knowing whether particular numbers fit the definition.
what the !@#$%^are you people talking about - please speak clearly on point in a forum debate about whatever it is you are talking about-- hate and trolls. and troll haters seem to go hand and hand here now
They are speaking about the forum topic. If you wish to comment, please write about the subject itself and not how you don't understand it.
It appears a schoolbus of children has crashed into the thread...
Feb 17, 2025
0
#14415
MARattigan wrote:
playerafar wrote:
perfect play with regard to win or draw can't be properly defined because chess isn't solved. ...
Of course it can be defined. You just don't know what actions fit the definition.
You can define a prime number without knowing whether particular numbers fit the definition.
I said 'properly' defined. Not just 'defined'.
Example: 'perfect play' versus 'play that doesn't lose'.
'defined' isn't digital A or B.
Another term is 'better defined'.
You can define a car as 'something that moves' ...
or as a ground vehicle with wheels and propulsion.
-------------
Martin - try it out.
Try 'perfect play' versus 'moves that don't lose'.
I say 'proper definition' could mean the distinction is accurately defined.
And 'because chess isn't solved' is a reason I stated.
Its valid but there could be other reasons.
It 'can't be defined' as in not yet.
Another idea: the word 'perfect' could have variations in assigned meaning.
We already know there's been disagreement about 'perfect information'.
The meaning of 'Perfect play' could depend on who is asked.
In other words - its arbitrated. Arbitrary.
So its not properly defined.
----------------------
Feb 18, 2025
0
#14416
MARattigan wrote:
Looks like O has formed his own cabal.
There's also the idea of 'protection'.
I saw a lot of that in the clubs. Its rarer in the public forums but it happens.
Somebody demonstrates that he break the rules directly - with no consequences -
with the intention of baiting a reply in kind and then reporting that person or threatening to report them. As opposed to discussing what is reportable.
---------------------
O got muted for three months recently.
But an army of people shillling for him and reporting others for anything and wanting rules even against words like 'disinformation' can make it more difficult for the moderators. But 'protection' takes many forms.
And so do power plays like threatening to not talk to a person who talks to somebody else. Or telling somebody to not support somebody else.
You know somebody's gotten himself manipulated when he/she gives in to things like that. Especially when the power plays are made openly in public.
And you know somebody's 'protected' when he keeps insisting 'only a few can understand' but too many don't talk back to him.
After all - he might report them. Right? He has in the past.
And just now - another phony exit from him. 'threats to leave'.
We're to be scared? 
-------------------------------------
Forum subject:
Computers solving chess.
Stockfish is AI? AI chess engine?
AI is used to write programs.
Meaning of AI - spectral.
We can't talk about it if we don't have a degree in it?
When it appears in the news that AI is being used to assist developers to write software - they're not talking about Stockfish.
AIs like Chatgpt and Copilot aren't good at chess?
Not the point at all.
Stockfish 'learns' so its therefore writing its own code?
That doesn't mean it wrote its base code.
And Stockfish is fallible too.
Diagrams have been posted here about that.
Feb 18, 2025
0
#14417
I don't think Dubrovnik understood when I used the term 'computer projects'.
I was referring to the forum topic and the tablebases - for example.
And other computer projects - not software in chessplaying engines.
If Stockfish is being used in the tablebase projects - that would be different.
But many times here the idea has been brought up of 'improving the algorithm'.
I'm suggesting that that would be done by human programmers assisted by chat AI.
Is the software they're using to solve 8 piece positions 'self-learning'?
They probably use some of that.
But that doesn't mean it has to be stockfish. Or stockfish only.
Oh man
playerafar wrote:
Does that 'yes' answer mean that Stockfish is 'AI'?
AI in the general sense has been around for a long time.
One could say any operating computer and its software is 'AI'.
So Stockfish software inside a computer is 'AI' in that sense.
But what about AI in the more recent sense?
Things like chatgpt and copilot?
Stockfish is in that group?
You can have a conversation with Stockfish like you would with those AI's?
You can verbally tell Stockfish to set up programs?
I guess I'll see the reply next time.
Theres alot of information out there on different types of AI you can check out. ChatGPT is an LLM and Stockfish is powered with like a specialized AI, weak one at that. The search tree and evaluation function it uses is algorithm based. It might actively use a neural network as well during play... Atleast Leela uses neural network. They essentially predict the most likely best decision like move or word. Obviously LLM is totally based on the machine learning and neural networks it completely relies on the data its fed.
It is absolutely true.
"Mathematical rules can be abstracted from chess positions."
Yes, that is how chess engines have always worked. And work today.
That does not solve chess.
Oh yes if its done it does solve chess. If a single position (like fortress or zugzwang) can be solved without brute force search then its entirely possible any position can. We dont know whats doable.
And by the way engines have been horrible at detecting fortresses in the past, dont know how much better they are now.
My challenge to all the "algorithmic" solvers (to use Optimissed's misusage of the term) is to start proving the premise. Pretend chess is geometry and start building theorems from the ground up. If you can get to one that's completely unknown/new and provable, then you have something.