Punishment as communication

Punishment is traditionally modeled as a “constructed incentive”. Specifically, the person who punishes is constructing a set of incentives that will motivate the person being punished to change her behavior. This model spans moral psychology, behavioral economics, artificial intelligence and comparative psychology. Yet, while it captures something important about the logic of punishment, it also misses important contours of how humans punish and how they learn from punishment. We argue that, in addition to understanding punishment as a form of incentive, we must also understand it as a form of communication. According to the constructed incentive model, a person learns from punishment the way she learns from other negative events: by trying to avoid the state or action that elicited it. In contrast, according to the communication model, a person can learn from punishment by attempting to infer the punisher’s communicative intent. Meanwhile, the punisher can choose punishments that are maximally informative given the learner’s likely inferential stance. This coordinated communicative reasoning enables faster and more accurate learning during punishment and provides a better fit to empirical data on how people actually punish and learn from punishment.