Palminteri et al. (2016) investigate how the computational mechanisms underlying reinforcement learning develop across adolescence and adulthood.
The study found that adolescents and adults use different computational strategies when learning from rewards, punishments, and feedback. Adolescents relied primarily on simple reward-based reinforcement learning, whereas adults incorporated more sophisticated processes, including learning from counterfactual outcomes and contextualising information across different situations. Adolescents showed a stronger tendency to learn from rewards than from punishments and gained less benefit from information about unchosen alternatives. Overall, the findings suggest that developmental changes in reinforcement learning contribute to age-related differences in decision-making and may help explain characteristic features of adolescent behaviour.