Valence biases in reinforcement learning shift across adolescence and modulate subsequent memory

Area of Science:

• Cognitive Neuroscience
• Developmental Psychology
• Neuroeconomics

Background:

• Individual differences in reinforcement learning exist, with some people overweighting positive outcomes and others overweighting negative outcomes (valence biases).
• These valence biases may extend to memory, influencing what information is retained from past experiences.
• Understanding how these biases develop and impact memory is crucial for comprehending decision-making and learning across the lifespan.

Purpose of the Study:

• To investigate age-related changes in valence asymmetries during reinforcement learning across adolescence.
• To determine if individual learning biases predict the content of subsequent episodic memory.
• To explore the interplay between value computation, learning, and memory consolidation.

Main Methods:

• Participants aged 8-27 learned the values of different 'point machines' through trial and error.
• Memory for trial-unique images, presented alongside choice outcomes, was assessed after the learning phase.
• Statistical analyses examined the relationship between individual learning biases and memory recall.

Main Results:

• Adolescents (compared to children and adults) demonstrated an over-weighting of worse-than-expected outcomes during learning.
• Individual valence biases in learning significantly modulated incidental memory.
• Participants who prioritized worse- (or better-) than-expected outcomes during learning showed enhanced memory for images associated with those outcomes, a finding replicated in a separate dataset.

Conclusions:

• Reinforcement learning exhibits age-related shifts in subjective value computation, particularly during adolescence.
• Valence-asymmetric learning processes directly influence the prioritization and content of episodic memory.
• These findings reveal a critical link between learning biases and memory formation, impacting how experiences are encoded and recalled.