From Esoteric to Transcendental Chess Programming?

[smatovic]

Heyho,

I do not follow SF development, but I get here and there a breadcrumb, for
example:

"is LVA as in MVV-LVA useless ?"

http://talkchess.com/forum3/viewtopic.php?t=70918

"...Lazy SMP feeds on chaos..."

http://talkchess.com/forum3/viewtopic.p ... 84#p824068


So I ponder if we left the paradigm of esoteric chess programming, one has to
get into the techniques, understand them, implement them, improve them to
transcendental chess programming, "it tested better"?

If we consider that chess engines run on Turing-Machines, we could conclude
that everything what happens in the chess engine is traceable by using pen n
paper, obv. this is not the case anymore? And I am not talking about NNs here,
just the classic approach. Hence the question, did we enter such a kind of
development and when?

--
Srdja

[maksimKorzh]

Heyho,

I do not follow SF development, but I get here and there a breadcrumb, for
example:

"is LVA as in MVV-LVA useless ?"

http://talkchess.com/forum3/viewtopic.php?t=70918

"...Lazy SMP feeds on chaos..."

http://talkchess.com/forum3/viewtopic.p ... 84#p824068


So I ponder if we left the paradigm of esoteric chess programming, one has to
get into the techniques, understand them, implement them, improve them to
transcendental chess programming, "it tested better"?

If we consider that chess engines run on Turing-Machines, we could conclude
that everything what happens in the chess engine is traceable by using pen n
paper, obv. this is not the case anymore? And I am not talking about NNs here,
just the classic approach. Hence the question, did we enter such a kind of
development and when?

--
Srdja

I guess this can't be unified for all engines.
SF is a community engine with a complicated testing framework and the way they approach is based on these circumstances behind.
For engines maintained by single authors such an extended test-driven approach is not the case due to the limited resources - not everyone
would invest money into testing like Andrew Grant does.
I thinks what you call esoteric vs transcendental chess programming is the matter of resources being involved.
The "new" era starts for an engine as soon as people start to invest money into it's testing.
So IMO it's all the matter of development scale and goals.

[hgm]

I used to call this 'Voodoo development'.

[Daniel Shawul]

Heyho,

I do not follow SF development, but I get here and there a breadcrumb, for
example:

"is LVA as in MVV-LVA useless ?"

http://talkchess.com/forum3/viewtopic.php?t=70918

"...Lazy SMP feeds on chaos..."

http://talkchess.com/forum3/viewtopic.p ... 84#p824068


So I ponder if we left the paradigm of esoteric chess programming, one has to
get into the techniques, understand them, implement them, improve them to
transcendental chess programming, "it tested better"?

If we consider that chess engines run on Turing-Machines, we could conclude
that everything what happens in the chess engine is traceable by using pen n
paper, obv. this is not the case anymore? And I am not talking about NNs here,
just the classic approach. Hence the question, did we enter such a kind of
development and when?

--
Srdja

NN is more of a black box as one doesn’t have any idea how the NN decided to evaluate one move better than the other. Classic stockfish has shown an effective testing framework and tuning methodology is fundamental, which btw was helpful even after Stockfish went NNUE too. Lc0 still lacks that framework and rely on testers to pick nets for example. Some complain recent NN/nnue Evans being “button press” solutions, but in reality this “problem” started when extensive testing was needed to verify if an idea is +1 Elo. You basically need a group of developers to generate ideas and test them on cluster of computers.

That a NN is a blackbox doesn’t matter to me as long as the methodology to train a strong net is understood.

[smatovic]

Interesting, I take this as confirmation that already without NNs in chess 'we'
entered a kind of black-box level (transcendental chess programming) with our
test-driven development methods, thanks.

--
Srdja

[JohnWoe]

NN are just massive PSQT. The "magic mushroom era" started way earlier testing all kind of +1 Elo crap with massive HW power. Looking at the top programs search functions. The same stuff, same order even same comments as in SF search. That's why I'm not reading any top programs sources. So boring. You learn nothing. Entropy is gone when products are alike.

[smatovic]

Hehe...

"trascendetal chess programming"
"new era"
"Voodoo development"
"black box"
"magic mushroom era"

any other suggestions?

j.k.

It seems we are missing here the right kind of terminology for something people are well aware of, or alike?

--
Srdja

[Henk]

Maybe Idiotery started with magic bitboards. Just call it perfect hashing or something like that instead of superstitious nonsense.

I also see neural networks as resignation: We can't solve the problem so let's use neural networks as a last resort.

[op12no2]

While a NN is a black box I think new knowledge can emerge from it, if there is new knowledge to be had. For example if LC0 or something like it becomes much stronger than all heuristic based engines because it has been freed from its human heuristic evaluation roots - and it seems to have a totally original style - one could conceivably discover new knowledge by hypothesising new heuristics based on observations of LC0 game play by experts and trying them out Texel-style (for example). Or even figure out an algorithms that works through a series of structure formations and similarly try them in the same way. New knowledge could emerge.

One thing is for sure. If these new structures/heuristics exist, they are going to seem super-weird - otherwise they would have been discovered already organically.

[smatovic]

...
One thing is for sure. If these new structures/heuristics exist, they are going to seem super-weird - otherwise they would have been discovered already organically.

I think I agree...

http://talkchess.com/forum3/viewtopic.p ... 57#p869633

My intend with this post was not to judge about NNs or any kind of black box development, I am interested in this from an viewpoint from the concept of the Technological Singularity, or alike. If we assume such a thing like a TS take off, and we look back, when and how did this happen? Did the TS take off via NNs, or already earlier? Where was the breaking line of systems which exceed human understanding? Something like this. Hence the question about black-box development before NNs.

--
Srdja