Value control of punishment

Abstract

Punishment raises a basic question for human cognition: do agents suppress actions because they represent the negative consequences those actions produce? Everyday deterrence, from parental reprimands to legal sanctions, assumes that people regulate behavior by learning what their actions cause and by evaluating those consequences negatively. However, in spite of this widely-held assumption, experimental evidence that punished instrumental behavior is controlled by the current value of the punisher is currently lacking. To test this idea, we applied an outcome revaluation procedure to instrumental punishment in rats. Rats first learned to press two levers for food. One lever was then additionally punished by response-contingent footshock, whereas the other remained unpunished. The aversiveness of the shock was subsequently reduced by pairing it with food in the absence of the levers. To test if the punishers value was part of the animals representation of the consequences of their actions, responding was tested in extinction, with no shock delivered. We found that revaluation of the shock selectively abolished the suppression of the punished response. Thus, punishment was not a fixed behavior controlled by the prior occurrence of shock, but depended on the shocks current value. Our results show that instrumental punishment can be controlled by a value-sensitive representation of the aversive consequence produced by the action and is, in this sense, structured in the same manner as canonical goal-directed reward seeking.