Open Problems in Mechanistic Intepretability of Biological AIs

I run a research program at BiodynAI aimed at advancing mechanistic interpetability of bioligical foundation models. I mean models trained on things such as DNA sequences, proteins, gene-expression profiles, cell images, tissue samples, spatial measurements, and other biological data. Many of them learn through some version of “hide part of the data and predict it back”, although the exact objective varies.There are 3 cases for that:Bio AIs are "model organisms for mechinterp in the wild": they ...