Brief introduction about me: I don't know how to code chess engines though I use them a bunch. Not sure what that makes me, probably lazy. Yay!
I figured I might as well take a shot at making a post here as I could get a 'yes' for an answer, at the very least maybe get a response from people experienced with the ins and outs of engine development whether or not the thought process behind my idea could/would work as intended.
The idea I have is based on the following:
Engines calculate a move at various depths. After every depth, the evaluation may change, either slightly or wildly.
My assumption is that, for moves where this eval score fluctuates more radically between depths, the resulting position(s) are more likely to be or become complicated/tactical/imbalanced, for the engine appears to struggle more to correctly analyze this position.
Think of it as if there are two possibilities:
A) - A move, which at just about every depth has an evaluation score of right around 0.0.
B) - A move, for which the evaluations at every depth average out at 0.00, although the eval scores ranged between -0.5 and +0.5.
The former is likely to be a rather bland, drawing endgame position, whereas the latter is more likely to be a position such as White is up 5 points of material but facing a mating attack by Black which isn't easily stopped.
With the recent new 'boom' of some aggressive chess engines, like Patricia and Rebel Extreme to name a few (shoutout to Stefan Pohl for his excellent EAS work), I figured this idea might be worth exploring, if it hasn't been explored already (in which case feel free to point this out to me).
As an example, you could set a bonus to be added towards an engine's evaluation score for a move, based on the standard deviation of the last x depths that were checked. The higher the standard deviation, the harder it appears to be for the engine to evaluate the position correctly, so the more likely the position is to get more complicated, so the bigger the bonus should be. Or you could do the reverse and get a very drawish engine.
If there are any developers in here who would like to try out something along these lines, or let me know if there are engines that are already doing something similar, or want to chime in to the discussion to discuss the potential of this idea, please go ahead and engage! Thanks in advance.
P.S. For your information, I did attempt briefly to add a 'bonus formula' like this to the code of Stockfish, but as I pointed out at the beginning of this post, I have no knowledge on this at all, so I took a shot at ChatGPT, and unsurprisingly, that lead me nowhere.
