The most human-like engine?

[Fritz 0]

Sorry if this question has already been asked here, but which is, in your opinion, the most human-like chess engine?

[dkappe]

Sorry if this question has already been asked here, but which is, in your opinion, the most human-like chess engine?

If you mean “plays like a human being, warts and all,” I’d pick the small leela-style nets like Maia, or the small Bad Gyal nets that inspired them at low nodes.

[Fritz 0]

Sorry if this question has already been asked here, but which is, in your opinion, the most human-like chess engine?

If you mean “plays like a human being, warts and all,” I’d pick the small leela-style nets like Maia, or the small Bad Gyal nets that inspired them at low nodes.

Yes, I mean exactly that. Actually, I'm mostly interested in which engines play like a human when lowered down to human strength, say, between 1800 and 2800 Elo. Thanks for your input.

[Frank Quisinsky]

Strengths and weaknesses of strong chess players are different.
Arguments turn out differently.

Examples:
Most of World Champions are strong in endgames. I think most of 16 World Champions are World Champions because the endgame strength was better as from the challengers.

Kasparow or Carlsen are allrounders in my opinion.
Karpow is more a strategy expert and Tal more the attacker and so one.

Genereally I would like to say:

- humans give quickly draw if the game is draw. What the programs from Ukraine do since a long time, Igel or Booot. And this is absolutely correct!

- humans have today no chance in endgames or transition into endgame vs. strongest chess engines. The endgame-strength is "hyperhuman, superhumanly" for strongest players in the World. More _human-like_ is if the strength in endgames to be compare to the the strength in mid-games and not "hyperhuman". So, not the best of TOP-41 engines in endgames are playing in endgames more human-like.

- humans have the strenght with a long-plan in mid-games + after openings. Engines like to be strong after openings with a great king-safty and aggressiveness are on the hightest "human-like-level".

So what is human-like if I am looking on available engines and strengths and weaknesses engines have.

- Hakkapelitta
- Spark
- Fizbo
- Velvet
- Wasp
- Hiarcs (often)

Other example:
Computer chess Elo from Engines like to lose to many games to fast isn't right!
The game started with the opening and not with the endgame!

List of strong programs, have problemes with to many fast lost games:
All on the list are stronger in endgames!

- Black Marlin
- Zahak
- Fritz 18 (Gingko)
- Rebel versions basesd on Fruit / Senpai
- generally all engines based on Fruit / Senpai
- Rybka
- CPU Lc0 ... of course not the GPU Lc0
- Combusken
- Shredder
- Halogen
- Winter

and different others.

Example Shredder 13: If 3125 Elo in computer chess = 2975 Elo in human chess because the weaknesses in the beginning of games (to many fast lost games) is fact.

Vs. "weaker" after openings Carlsen chance is much higher to win!
Generally a stronger human can make clearly more points because chess started not with an endgame position.

Thats my opinion!

Last year a good known trainer of Grandmaster try to explain me why programs stronger in endgames are more human-like!
But to compare 3500 Elo on one core with 2850 Elo max. strengths for humans isn't easy.

I wrote him ... if computer chess engines are on max. level from 2850 (today 3500 Elo) it can be an argument that programs strong in endgames are more "Human-like". Not in times today, time is over! Each engines in TOP-50 with weaknesses in endgames are on min. 3100 Elo endgame-level. His answere ... I will not believe that Engines on one Core with 4.5Ghz are on 3500 Elo.

Best
Frank

[dkappe]

Frank,

I think Stefan Pohl demonstrated that draw length isn’t nearly as relevant as the material situation. But this all seems pretty beside the point. An engine like Igel is pretty inhuman. We’re not really looking at the nature and frequency of its mistakes and blunders.

[Frank Quisinsky]

Hi Dietrich,

best chess players in the World are playing blunders.
This is more an argument for "Human-like".

I am not saying Igel is playing a "human-like-style".

I like the tools by Stefan Pohl, but I make such stats not since a year, more or less since 20 years and try to find out the strengths and weaknesses from engines. In the time I am working for GM Jörg Hickl and his chess-magazine the selection about playing-styles of engines are most populare. I had many dicussions with very strong grandmasters and correspondence chess players about it. At this time Junior produced a lot great fast won games and Junior and Spark are the big topic.

A very interesting time:
Most of stronger correspondence chess players have the same opinion that Rybka is playing in the beginning of the games chess with to many blunders. The chess computer community are thinking ... oh, Rybka is on 1 and an engine on 1 is perfect for all playing-phases. No, no ... very interesting that the correspondence players have the same opinion that Rybka is interesting for the very late mid-games - endgames only.

Back to Igel.
Igel lost many games with blunders, yesterday I saw here a game vs. Wasp 6.00 NN (mate after 27).
But Igel can win many games very fast ... have here also strengths.

Stefan can not see that with his fast time controls.
The differents from 40 in 4 to 40 in 8 is gigantic.
After all I know Stefan tested with much faster time controls on 2Ghz only.

For me it was very interesting to compare engines with 40 in 20 I do with my FCP-Tourney-KI.
So, I can see and can learn a lot more.

Best
Frank

[dkappe]

Frank,

I know it’s tempting to shoehorn your passion project into every discussion of engine style, but AB engines are out of the running as far as “human-like play” goes. It’s like arguing about which car engine is the most human-like. So we can table your extensive statistics, unless you have some on lc0, classic ara and their relative performance.

MCTS with its incomplete exploration of the game tree at least can be made to simulate the error rate and style of a human being.

[istolacio]

It is difficult for an engine to behave like a human. In general, they base their decisions on the depth of calculation. Therefore, I am more interested in knowing which chess engine is the best to use as a tutor.
Which one has more chess knowledge? Stockfish is the strongest but he is useless as a teacher. I want to know your opinion about the most suitable program positionally, not tactically.

[dkappe]

It is difficult for an engine to behave like a human. In general, they base their decisions on the depth of calculation. Therefore, I am more interested in knowing which chess engine is the best to use as a tutor.
Which one has more chess knowledge? Stockfish is the strongest but he is useless as a teacher. I want to know your opinion about the most suitable program positionally, not tactically.

I’m tempted to say something snarky about the best tutor being the one with the shortest average draw length, but that would be unkind.

I’d suggest instead you open a new thread, as it’s not really on topic for the original post.

[CornfedForever]

It is difficult for an engine to behave like a human. In general, they base their decisions on the depth of calculation. Therefore, I am more interested in knowing which chess engine is the best to use as a tutor.
Which one has more chess knowledge? Stockfish is the strongest but he is useless as a teacher. I want to know your opinion about the most suitable program positionally, not tactically.

You really need to look to a GUI that takes an engines output to manipulate thru some algorithms for that...no 'engine alone'.