Chess position representations in AI and humans - comparative study

It has been established that chess engines and humans process chess positions in a very different way, the former relying on search (as opposed to evaluation) much more than the latter, but recent neural engines are much closer to humans in that regard. I use custom heuristics and machine learning interpretability tools on a SOTA neural chess engine to estimate the difficulty of chess positions from a human perspective.

Tytuł polski: Porównanie ludzkiej i komputerowej reprezentacji szachowych pozycji

Link to meeting/Link do spotkania: https://meet.google.com/jbj-tdsr-aop