Implicitly learning to reason in first-order logic