We present a simulation strategy for the real-time dynamics of quantum fields, inspired by reinforcement learning. It builds on the complex Langevin approach, which it amends with system specific prior information, a necessary prerequisite to overcome this exceptionally severe sign problem. The optimization process underlying our machine learning approach is made possible by deploying inherently stable solvers of the complex Langevin stochastic process and a novel optimality criterion derived from insight into so-called boundary terms. This conceptual and technical progress allows us to both significantly extend the range of real-time simulations in 1+1d scalar field theory beyond the state-of-the-art and to avoid discretization artifacts that plagued previous real-time field theory simulations. Limitations of and promising future directions are discussed.