Asymmetric Reinforcement Learning Explains Human Choice Patterns in Decision-making Under Risk.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Behavioral Economics

Background:

• Human decision-making under uncertainty is influenced by experience.
• Reinforcement learning (RL) is a proposed framework, but its application to risk remains debated.
• It's unclear if symmetric or asymmetric learning better explains behavior under risk.

Purpose of the Study:

• To investigate whether symmetric or asymmetric learning strategies better explain human choices.
• To examine learning strategies in novel decision-making tasks with contextual uncertainty and varied outcome distributions.

Main Methods:

• Developed and compared computational models of learning.
• Fitted candidate models to individual trial histories of human behavior.
• Analyzed predictions for choice and response time.

Main Results:

• A Risk Sensitive (RS) model with asymmetric learning rates provided the best fit to human behavior.
• Value signals derived from the RS model predicted both choices and response times.
• Asymmetric learning effectively captures behavior in decision-making under risk.

Conclusions:

• The Risk Sensitive (RS) model offers a concise and identifiable account of decision-making under risk.
• Asymmetric learning, weighting rewards and losses differently, is crucial for understanding human choices under uncertainty.
• Future research should explore the neural basis of asymmetric learning in decision-making.