Hassabis's conjecture is that any pattern that can be generated or found in nature can be efficiently discovered and modeled by a classical learning algorithm, provided there is sufficient data and to a certain level of resolution.