When will the chess programmers write an engine that plans ?

[mclane]

LC0 is the evidence for me that it is possible to build strategy over tactics and still play decent human chess.

Artificial neural networks have as little to do with human intelligence as using a compressor to press manure through a hole in the wall has to do with human digestion.

So far NN play a more intelligent chess then normal AB Programs. The engines made by humans play machine chess while the engines made by machines play human chess. A Paradoxon.


But it’s true, NN need a lot of GPU power although they also play very good chess if they have low NPS, as I have tried out many times.

The optimum would be an engine not using much tree but planning beside tree search by geometrically analysing the possibilities in the position and then trying to build a tree on the base of the plans.

[Ovyron]

On the other hand, if the OP wanted a "real" answer he would have given a more detailed and concise description of his needs along with a $1,000,000 prize to the person that can provide it first.

No prize money was required to make the community come up with projects like Stockfish or Leela, and they are run at a loss (people are paying electric bills to test them, with no return of investment.) This tells a lot about the nature of the "boohoo my engine doesn't play like a human" *air quotes* problem. *air quotes*

[jp]

Humans have transported their human chess thinking into the programs.
The evaluation and search ideas are, in fact, human thinking.
<snip> and just about everything else a GM does.
Sure, it is just calculation, but that is also true when a GM does it,

This is not what a GM or any human does. No human sits there adding up lots of different terms (multiplied by awkward coefficients like 0.38526) to get a number, etc. The closest a human gets to that is counting pieces and assigning values to them to see who's ahead on material.

Have you ever read correspondence champion Hans Berliner's book?
That is exactly what he did.

Over the board?? No human has ever memorised lots of long equations and then sat at the board calculating them with mental arithmetic to "play" chess. Of course, in correspondence chess, you can manually play like a computer program if you want. Not over the board.

Dann, do you own a copy of the book? Could you quote the relevant passages? (Berliner was delusional about his achievements, but it'd be amusing to see what he said.)

[Dann Corbit]

The book is called The System, and he gives numeric points to various features of chess.
He gave a valiant exposition of why d4 is best (not sure if he was right or wrong about it, but it is pretty good at any rate).
The whole book is about how to compute chess numerically.
If you look at his correspondence record, it is incredible.
And the computers at the time were pretty pitiful in comparison.
His machine (hitech) was a hardware based solution.
I do have the book, but it is at work, and I am not allowed to go there, as we are required to work from home. So I can't give you any specific details right now, Whenever the crisis is over, I can give you quotes.

[jp]

The whole book is about how to compute chess numerically.
If you look at his correspondence record, it is incredible.
And the computers at the time were pretty pitiful in comparison.

I think that's why his correspondence record was so good. The computers were pitiful, and he could just work through in effect a computer program by hand.

I do have the book, but it is at work... Whenever the crisis is over, I can give you quotes.

Thanks!

[jp]

While we wait, a position Berliner opined about:

[d]r1bq1rk1/pp2ppbp/2n3p1/2p5/2BPP3/2P1B3/P3NPPP/R2QK2R w KQ -

He insisted White must play Rc1 here.

Engines at low depth go for 0-0. It looks like 0-0 is by far the most popular GM move too, but I don't have access to a big database.

[Dann Corbit]

Stockfish at minimal depths likes them the same (both draw scores).
The position after Berliner's move has a slightly better outcome statistically.
I will let LC0 have a look.

[d]r1bq1rk1/pp2ppbp/2n3p1/2p5/2BPP3/2P1B3/P3NPPP/2RQK2R b K - acd 40; acs 282; bm cxd4; c3 "cxd4"; cce -37; ce 0; pm cxd4; pv cxd4 cxd4 Qa5+ Rc3 Bg4 f3 Rad8 Qd2 Be6 Bxe6 fxe6 Rd3 Qa6 O-O Rd7 Rc1 Rfd8 a3 Ne5 Rdc3 Nc6; white_wins 202; black_wins 154; draws 221;

[d]r1bq1rk1/pp2ppbp/2n3p1/2p5/2BPP3/2P1B3/P3NPPP/R2QK2R w KQ - acd 40; acs 344; bm O-O; c0 "Rc1 and Rb1 {while rare} have performed well also"; c3 "O-O"; cce 22; ce 0; id "QuietTest.45069"; pm O-O {2889} Rc1 {213} h4 {21} Rb1 {11} dxc5 {4} Qd2 {4} d5 {1} f3 {1}; pv O-O Na5 Bd3 b6 Rc1 e6 dxc5 Qc7 cxb6 axb6 Nd4 Nc4 Bxc4 Qxc4 Qb3 Qd3 Qb1 Qxb1 Rxb1 Rxa2 Rxb6 Ba6 Rc1 Bd3 f3 Rfa8 Nc6 h5 Nb4 Re2 Bf4 Bc4 Rc6 Bb3 Rc7 h4 Rb1 Bc2 Rc1; white_wins 23221; black_wins 18828; draws 53727; Opening D87d Grünfeld: Classical Exchange, 8...c5 9.Be3 Nc6;

[bob]

The book is called The System, and he gives numeric points to various features of chess.
He gave a valiant exposition of why d4 is best (not sure if he was right or wrong about it, but it is pretty good at any rate).
The whole book is about how to compute chess numerically.
If you look at his correspondence record, it is incredible.
And the computers at the time were pretty pitiful in comparison.
His machine (hitech) was a hardware based solution.
I do have the book, but it is at work, and I am not allowed to go there, as we are required to work from home. So I can't give you any specific details right now, Whenever the crisis is over, I can give you quotes.

There was a book I used back in the 70's called "point count chess". It did exactly what you are saying. The author was extremely detailed in assigning points/fractions of a point for various positional structures. I think Horowitz was an author, although he probably had help. Used to have the book, no longer however.

[bob]

While we wait, a position Berliner opined about:

[d]r1bq1rk1/pp2ppbp/2n3p1/2p5/2BPP3/2P1B3/P3NPPP/R2QK2R w KQ -

He insisted White must play Rc1 here.

Engines at low depth go for 0-0. It looks like 0-0 is by far the most popular GM move too, but I don't have access to a big database.

This is interesting. I tried it on Crafty. O-O pops out as the favored book move. If I disable the book, then Rc1 is chosen until the 8 second mark on my MacBook air, where it then also switches to O-O.

Hans was no doubt a strong chess player. But he believed he was stronger than he really was, it seems. We played him a blitz match at some conference, using Cray Blitz and the real chess clock (which means my typing time was on CB's time). CB won every single game easily. I used to think "If I could buy him for what I think he is worth, and sell him for what he thinks he is worth, I could retire immediately..."

He was certainly amazing at correspondence speeds. At least in the tournament he won.

[Ras]

So far NN play a more intelligent chess then normal AB Programs.

That's a gross misunderstanding based on three factors:
1) You understand AB a lot better than NN.
2) You misinterpret what the node count of LC0 even means.
3) Wishful thinking.

If NN had anything to do with the (still weak!) human way of thinking, then you wouldn't need a super high-end graphic card with all its raw computing power to run LC0. And if AB is blindly throwing cr*p against the wall and see what sticks, then NN with MCTS is rolling dice to decide which cr*p to throw against the wall and see what sticks.

Even back in the days of the Mephisto units, the legendary Glasgow which actually tried to emulate human thinking was rated at only 1776 Elo despite a 12 MHz 68k processor. The MM5 with a more traditional engine came in at around 2000 Elo although it was based on the much slower 5 MHz 6502 processor. In fact, the Glasgow was the weakest chess computer ever with a 12 MHz 68k. The strongest was probably the Berlin at 2190 Elo, again with a traditional engine.

Conclusion: if you emulate the weak human way of understanding chess with a computer, all you'll get is a computer that plays weakly. On today's hardware, even that might be better than humans, but not better than traditional engines - which is why there's no interest.