Future of Computer Chess!!-!?-?!

ive been fooling around with the engines, and it dawned on me today that there should be a new symbol/statistic/indicative in a position (assessment of a position), and especially in Equal (=) positions that tells you the probability of error that your opponent has given the amount of candidate moves.

Not all Equal positions are the same!?

heavy, im sure the greats empoly this type of logistic.

??

Yes, this makes sense.

Interesting -- basically a complexity measure for otherwise equal positions then?

I love how charlesgalofre typed "Equal (=)". That is so cute!

If you think of the position score +/- as the mean, this new metric is a bit like the variance ... maybe ?

I've thought about this alot in the past... it is especially applicable when you look at engines reaching the tablebases.  Computers will often avoid taking a sharp but drawn tablebase path in a position it scores as +=, hoping to find a winning path by further evaluations of different lines, even if a human could identify that tablebase path as a near impossible ending to solve.  If playing against a human, you want to evaluate the "sharpness" of a position... even if the position is technically a forced draw.  Then a computer could lead its opponent down the "sharpest" drawn path.

Rather than just the variance in the candidate scores, I think you could compare the difference between a candidate moves rank in the initial heuristic sort and the rank final search order based on its score... arguing that moves initially sorted low but scoring the best are the least intuitive, and thus hard for people (and computers) to discover.  That would help define a sharp position in my mind.

this sounds massively insightful, i need to spend time playing with that. thank you for posting that insight.

something else that i found curious is that it would be instructive if there was a volatility of advantage indicator. it makes studying through opening books more colorful and imaginative for the brain.

That volatility depends a lot on the knowledge and on the strengths of your opponent. If you do not manage to surprise him and avoid his preparation, or if you play into his strengths, volatility will be small or not in your favour.

But isn't it already the case that when a chess engine has evaluated the current position in a game to be equal, give or take 0.5 points, hasn't the computer not only calculated the remaining pieces on the board but also the lack of clear winning chances for either side irregardless of how the final position was reached?

The future of computer chess should involve animal brain cells interfacing with computers. I saw a video on YouTube where scientists had trained a culture of rat brain cells to successfully operate a fighter jet simulator program based on the feedback the cells got from the software program.

Just a thought.

Isn't that the method of Convekta products already?

Here's what you could do. Looking at the current position, the computer evaluates as many ply deep as it normally does--maybe 20 or 25. It then looks at combinations 5-7 ply deep (since this is about as deep as humans tend to look), or maybe a little deeper to balance for "positional instinct," and figures out how many combinations are acceptable at that depth that lead to unacceptable outcomes several ply deeper. For a high-level player, choosing one of these moves is a much more plausible form of error. Looking at all the moves immediately available in a position would make players look more accurate than they really are, since any position's going to have a lot of moves that are obvious blunders at a depth of 1 ply and thus aren't really worth considering.

tbischel already suggested (almost) the same exact method in post# 8:

I wonder if this algorithm is already in place for games without time controls.