Brain-deep AI interface for user-responsive reinforcement learning