Punishment Enhances Reward Learning through Striatal Prediction Errors

A gazelle cautiously approaches a watering hole, eyeing its surroundings for predators. It is driven by two simultaneous needs: to drink water and to avoid danger. Its survival depends on balancing these opposing forces: approach motivation, which drives reward seeking, and avoidance motivation, which drives evasion of punishment. Reinforcement learning is the process by which animals learn to make decisions through trial and error using rewards and punishments as feedback. Over time, animals learn to select actions that maximize rewards and minimize punishments. Behaviors followed by rewarding outcomes become more likely to be repeated due to approach motivation, whereas behaviors followed by negative outcomes become less likely to occur through avoidance motivation. These dynamics play out continually in day-to-day life as all animals, including humans, learn from these trade-offs. Although rewards and punishments frequently co-occur in real-world environments, little research has investigated how receiving a mix of rewards and punishments influences learning processes. Reinforcement learning paradigms often study them separately, creating siloed research domains that mirror the classic “carrot and stick” metaphor in which rewards and punishments are treated as independent tools for shaping behavior. However, behavioral evidence suggests that they are not discrete learning processes; for example, conditioned suppression studies show that cues signaling a potential punishment can disrupt approach behavior directed toward rewards (Allcoat et al., 2015). This finding demonstrates that reward and punishment systems dynamically interact rather than operate in isolation. One possibility is that experiences of reward and punishment influence each other, because reinforcement learning shares a neural substrate: dopamine. Dopaminergic neurons calculate prediction errors: the difference between what was expected and what actually occurred. Better-than-expected outcomes produce positive prediction errors through increased firing of dopaminergic neurons, whereas worse-than-expected outcomes produce negative prediction errors through reduced dopamine release. Studies in … Correspondence should be addressed to Robin B. Blades at robinblades{at}gmail.com.