Confirmatory reinforcement learning changes with age during adolescence.

Area of Science:

• Developmental science
• Cognitive neuroscience
• Computational biology

Background:

• Human development involves significant changes in learning processes.
• Individuals exhibit a confirmation bias, learning more from expected outcomes than unexpected ones.
• Adolescence is a critical period for exploration and learning, necessitating an understanding of developmental changes in learning biases.

Purpose of the Study:

• To investigate whether the confirmation bias in learning is stable across human development.
• To examine if this bias changes during adolescence, a period characterized by beneficial exploration.
• To understand the computational mechanisms underlying age-related differences in learning.

Main Methods:

• A reinforcement learning (RL) task was administered to 77 participants aged 11-32 years.
• Participants learned to maximize rewards by sampling novel options with varying reward/punishment probabilities.
• Mixed-effect models and computational modeling (confirmation model) were used to analyze age-related learning patterns.

Main Results:

• Accuracy in RL tasks increased with age, particularly in stable environments.
• An age-related asymmetry emerged: participants were more likely to repeat a rewarded choice than to switch after a punished one.
• Computational models indicated that age differences were primarily due to decreased noise/exploration, not an increased confirmation bias magnitude.

Conclusions:

• Learning from rewards and punishments improves with age during adolescence, with greater gains in stable environments.
• An increasing tendency to confirm choices (confirmation bias) develops with age, driven by reduced exploration.
• These findings offer insights into developmental changes in learning and can inform age-tailored educational strategies.