If you’re a complex machine whose action could be perfectly predicted (with full knowledge of everything you ever experienced) it’s still reasonable to punish you for breaking rules - the risk of punishment goes into your programming as part of the (deterministic) calculation of what action to take