Predictive auxiliary objectives in deep RL mimic learning in the brain