Neural prediction errors reveal a risk-sensitive reinforcement-learning process in the human brain.

Area of Science:

• Neuroscience
• Cognitive Science
• Decision Science

Background:

• Humans and animals learn from experience, adapting to risk and variance in outcomes.
• Traditional reinforcement learning models primarily focus on mean reward, neglecting risk (variance).
• Neural mechanisms underlying risk sensitivity in learning are not fully understood.

Purpose of the Study:

• To investigate if neural correlates of human reinforcement learning are sensitive to experienced risk.
• To examine the role of the nucleus accumbens in processing risk during learning.
• To explore the relationship between neural measures of risk evaluation and behavioral risk aversion.

Main Methods:

• Functional magnetic resonance imaging (fMRI) was used to measure brain activity.
• Analysis focused on blood oxygen-oxygen level-dependent (BOLD) signals in the nucleus accumbens.
• Learned values of cues predicting rewards with equal mean but different variance were derived.

Main Results:

• BOLD signals in the nucleus accumbens closely corresponded to reward prediction errors.
• Learned values were modulated by experienced risk, indicating sensitivity to variance.
• A strong coupling was observed between neural evaluations of risky options and behavioral risk aversion.

Conclusions:

• Risk sensitivity is an integral component of human reinforcement learning.
• Neural mechanisms in the nucleus accumbens play a key role in integrating risk into learning.
• Findings have implications for economic models of choice and neuroscientific models of affective learning.