Deep learning has shown that many systems we thought were effectively unsimulatable on classical computers can still be modeled accurately, even when we cannot explicitly derive the relevant rules from their microscopic physics. Protein folding is a good example. In principle, a