the striatum and frontal cortex likely support different aspects of
RL, where in the striatum prediction, error signals mainly support
bottom-up, model-free learning and in the frontal state prediction,
errors are thought to guide goal-directed, model-based learning