In 1928, Dirac correctly foresaw that the main future goal of physics is finding approximate methods for solving Schrödinger's equation. The current status of first-principles quantum mechanical methods applied to the modelling of interactions between molecules will be presented, with particular emphasis on symmetry-adapted perturbation theory. However, to model condensed matter, including biosystems, one has to fit ab initio data by simple functions of interatomic separations called force fields. A machine-learning program called autoPES performs such fits purely automatically, opening new avenues for investigating condensed phases. Applications for predicting crystal structures will be presented.