What I miss are alternative ideas how to program chess.Tord wrote: ↑Thu Oct 08, 2020 10:42 amThis is largely a myth. As explained by Ivan Sokolov in the preface to Sacrifice and Initiative in Chess:
When it comes to Shannon type A vs B: As has been pointed out earlier, all modern programs use type B, but they use shallow verification searches to verify that moves are safe to prune. Omitting verification searches would not make the engines play more intelligent or attractive chess, it would only cause them to make more tactical blunders (and this is the only way in which it would make the engines play more human-like).Ivan Sokolov wrote:Mikhail Tal's sacrifices have a reputation that there was a significant amount of bluff involved, and before I started working on this book and had a serious look at his sacrifices, I was inclined to concur to this general opinion.
But nothing could be further from the truth! Even if you give them enough time to run, computer engines are not able to refute 90% of Tal's sacrifices. There is always compensation even against the very best defense, and most of the time it is enough for at least a draw.
Of course, I do not know, and I will never find out, how much of those possibilities Tal actually saw and how much was his 'intuition' (please see in the chapter on 'Intuitive Sacrifices' my opinion on this subject). But I’m sure that he saw a lot! Tal was an attacking devil, a nine-headed monster, a true Houdini. Not the crap we buy for 80 euros and install on our computers – Tal was the real deal. He could hide an elephant!
While I agree with Thorsten that the engines of the past could be more entertaining and had more varied styles, this has nothing to do with type A vs type B searches. It is partly because of improved search depth and speed, and partly because human evaluation tuning has been largely replaced by automated tuning.
You can actually see something similar in elite human play: Elite chess has become much more about concrete calculation and less about clash of different styles and ideas, and there is less diversity of styles. If you go back to the age of Steinitz, Lasker, Nimzovich, Tarrasch, Spielmann and Rubinstein, you see much more variety and personality than in the current computer-assisted age.
We have now very strong chess engines.
To name komodo and stockfish.
But why is anybody interested in participating in the same race these 2 are in ?
If you do a verification search to be safe which branches to prune, you mainly destroy the whole idea that the engine itself needs to find out via knowledge which moves to follow and which moves not to follow.
Only knowledge can decide.
Similar it is about planning.
Instead of planning ideas in chess, the conventional chess engines create a huge chess tree.
And call this main line planning,
But this is not how chess works,
A chess player looks into the root position and sees weaknesses and creates ideas how to exploit the weaknesses by manoeuvre pieces etc.
Then he only generates those moves that follow his ideas.
All the other moves and areas on the board are not needed,
The chess player completely ignores them,
So I miss that today’s chess programmers , instead of trying out new ideas, repeat the common and used to ideas and by doing this
Repeat the development Komodo and stockfish were going,
Why ?
This way computerchess makes no progress,