For us to trust it on certain subjects, researchers in the growing field of interpretability might need to learn how to open the black box of its brain.
