Inflated Engine Evaluations

Engined don't count the sheer material but take also positional considerations into account. We all know that.

For 40 years ago we have had something like 1 pawn is something close to +1 depending on the positional factors. The last couple of years this evaluations have skyrocketed. Why`?

Engines calculate very deep and therefore assess the winning properties very high, there is almost hardly any connection to the material involved if there is a winning position.

Maybe a total switch to winning probabilities is the better way? Like AlphaZero does? A paradigm shift.

Practical example of a Blitz game of mine. Black is pawn down in the CK Advance, maybe a bit cramped but has certainly compensation.

The evaluation is minus 8 though...

Black to move:

lichess.org/editor/r4rk1/1pq1b1pp/4pp2/pbPpP3/1P1Q1PP1/2P5/P3NR1P/RN4K1

As engine looks depper, the rating will go polarized it is normal.

I like evaluations. I don't think engine evaluations have become inflated, they just have become better. Engines are much more intelligent than they were back only say 5-10 years ago. They understand long term advantages much better as well as several other positional considerations. An engine 5-10 years ago is up a pawn and calculates and evaluates it at +1. Today the engine can calculate much more efficiently (not really examining more moves but examining deeper into critical lines because they are selectively evaluating rather than attempting to use brute force). The engine can see that there is no counterplay and the evaluation reflects that. I don't think this makes the evaluation "inflated," I think it makes it more right.

Myself I am definitely too materialistic. I think almost all humans are. But how engines play is very beautiful to me because there's a level of dynamism and lack of materialism that is incredible to see. Recent evaluations are now tied far less to material concerns and are now almost completely dependant on positional and tactical (concrete) concerns. I don't want the engine to tell me I am + 5 if I am up a rook. I can count by myself, I don't really require a machine to do that for me. The great thing about evaluations is they are just far more objective assessments of the position than almost all human evaluations.

Like for example: "Black is pawn down in the CK Advance, maybe a bit cramped but has certainly compensation."
If the evaluation is - 8 black is winning. It doesn't matter what the material is. It doesn't matter if you think black has some compensation. Because the reality of the situation is white is lost regardless of what your opinion is. Of course this only is true with perfect play, but if a human is telling me that someone has compensation in a worse position, and a computer is telling me it is -8, the human is always wrong. Doesn't matter who it is really.

I think it is really good that evaluations are shifting away from materialism and towards more positional concerns. I think this will teach people new to the game (and more experienced) fairly important lessons. I don't see a reason for it to be changed. I think it is amazing that we live in a time when we have such incredibly powerful tools to show us some of the deepest ideas chess has to offer.

The engine evaluation I get is around -1,6, which feels right. White is a pawn down and has some positional issues. The top line shows how black can swap pieces and pin the rest down with an attack. Engines can only give the evaluation based on best play, not human play and I think the attacking potential down the f file is underestimated at a first glance.

@KurokosEchec said in #5:
> The engine evaluation I get is around -1,6, which feels right. White is a pawn down and has some positional issues. The top line shows how black can swap pieces and pin the rest down with an attack. Engines can only give the evaluation based on best play, not human play and I think the attacking potential down the f file is underestimated at a first glance.

Black to move.

A position can only be won or a draw. So were approaching the three possible outcome: -infinite, 0.00, +infinite

If this is ok then ok.

Btw, the whole game. 3+2, I had the Black pieces.

Last question: then why not shifting to winning probabilities? CB can for example, AlphaZero did so from scratch.

Better than sticking to numbers like +1, +10, +100. The position is won, either way.

@Sarg0n said in #1:
> Engined don't count the sheer material but take also positional considerations into account. We all know that.
>
> For 40 years ago we have had something like 1 pawn is something close to +1 depending on the positional factors. The last couple of years this evaluations have skyrocketed. Why`?
>
> Engines calculate very deep and therefore assess the winning properties very high, there is almost hardly any connection to the material involved if there is a winning position.
>
> Maybe a total switch to winning probabilities is the better way? Like AlphaZero does? A paradigm shift.
---
I think the human brain is more powerful than the engine. Even the 3000 performance engine. I send a sample of the game. Even the engine of analysts sometimes suggests that the motion they make is weaker than the motion that humans find.

@Sarg0n said in #6:
> Black to move.
>
> A position can only be won or a draw. So were approaching the three possible outcome: -infinite, 0.00, +infinite
>
> If this is ok then ok.
Oh OK, the analysis board was set with white to move.
I still stand by my statement that the analysis is accurate. Black's going to win a pawn and gain an overwhelming attack with white pieces not even having time to develop. Yes if the position was static black's edge would be small, but he's going to gain two points and pin down all of white's pieces. A quick look at the top engine line shows you'll most likely lose a knight (plus a pawn) leaving you a knight and two pawns down. That's an easily won position. Obviously it demands some accurate moves from black, but engines can only evaluate best play, it's not suggesting that a pawn up position is an easy win it's saying tactics mean that the position will gain a lot of material.