Future progress of chess software in chess

[Uri Blass]

1)Chess machines are equal to human+machines in playing correspondence chess
2)Chess machine are equal to union of humans+chess machines in developing better chess engines.
3)Chess machines are equal to union of humans+chess machines in teaching humans to play chess better in OTB chess.

I think that we are very close to 1.
When do you expect 2 and 3 to happen so we need no humans to make engines or to teach chess?

[maksimKorzh]

1)Chess machines are equal to human+machines in playing correspondence chess
2)Chess machine are equal to union of humans+chess machines in developing better chess engines.
3)Chess machines are equal to union of humans+chess machines in teaching humans to play chess better in OTB chess.

I think that we are very close to 1.
When do you expect 2 and 3 to happen so we need no humans to make engines or to teach chess?

Most likely when the processing power would be enough to train NNs to replace "hard coded knowledge" in a domain, do you agree?

[JohnW]

Didn't we already reach #2 when Alpha Zero taught itself to play?

[Alayan]

2 and 3 are absurdly hard. They require humans to become absolutely worthless compared to automated methods at making better engines and at teaching humans. It won't happen in our lifetimes and may never happen at all.

1 is just about playing strong chess moves, we're close to it (already there from the start position which is a dead draw for top correspondence players)

[Dann Corbit]

1)Chess machines are equal to human+machines in playing correspondence chess
2)Chess machine are equal to union of humans+chess machines in developing better chess engines.
3)Chess machines are equal to union of humans+chess machines in teaching humans to play chess better in OTB chess.

I think that we are very close to 1.
When do you expect 2 and 3 to happen so we need no humans to make engines or to teach chess?

I think that the problem with #2 and #3 is that nobody is working on it.
As for #1, I do not know if anyone has measured that, but I also guess it depends on the player.
Some players will surely contribute to the match in a positive way, some in a neutral way (as described in your #1) and some in a negative way.
So the alternative question:
1)Chess machines are equal to (the world's best human correspondence player)+machines in playing correspondence chess
is really a much more difficult question in my view.

[Uri Blass]

Didn't we already reach #2 when Alpha Zero taught itself to play?

Not exactly because using ideas of humans together with using alphazero's method is still better than using only alphazero's method.


Stockfish still test patches written by humans in order to get better and I think same for lc0.

[Dann Corbit]

Didn't we already reach #2 when Alpha Zero taught itself to play?

The human wrote the software that allowed the self-learning.
The human directed the program what games to play and how many.
The computer contributed nothing except brute force.
The software not only did not teach itself to play, it does not even know if it won or lost a game.
It is kind of like the Deeper Blue verses Kasparov match.
Deeper Blue has no idea whatsoever that it won the match.
On the other hand, Alpha Zero and LC0 are almost half way to thinking. Much closer than Deeper Blue.
In a sense, they can think, when we tell them what to think.

[Alayan]

AlphaZero's code is written by humans. There was human input everywhere: general design and concept, net format, search algorithm, training game parameters, and so on. Having some part that can improve through automated training is not the same thing at all as humans being useless in the process and contributing nothing at all that would yield more strength than if they didn't intervene.

I think that the problem with #2 and #3 is that nobody is working on it.

You could put a million AI engineers on the issue, and they wouldn't solve these problems any time soon. These problems are much harder than what's currently the bleeding edge in AI sophistication.

[MikeB]

1)Chess machines are equal to human+machines in playing correspondence chess
2)Chess machine are equal to union of humans+chess machines in developing better chess engines.
3)Chess machines are equal to union of humans+chess machines in teaching humans to play chess better in OTB chess.

I think that we are very close to 1.
When do you expect 2 and 3 to happen so we need no humans to make engines or to teach chess?

We need to get R2D2 involved.

[Milos]

1)Chess machines are equal to human+machines in playing correspondence chess
2)Chess machine are equal to union of humans+chess machines in developing better chess engines.
3)Chess machines are equal to union of humans+chess machines in teaching humans to play chess better in OTB chess.

I think that we are very close to 1.
When do you expect 2 and 3 to happen so we need no humans to make engines or to teach chess?

I think that the problem with #2 and #3 is that nobody is working on it.
As for #1, I do not know if anyone has measured that, but I also guess it depends on the player.
Some players will surely contribute to the match in a positive way, some in a neutral way (as described in your #1) and some in a negative way.
So the alternative question:
1)Chess machines are equal to (the world's best human correspondence player)+machines in playing correspondence chess
is really a much more difficult question in my view.

That's an interesting and completely wrong point of view.
The reality is that humans are useless in correspondence chess on top level because no amount of human intervention can score a win against SF on decent hardware on ICCF TC. Ergo, human contribution to engine performance is exactly 0 (ZERO).