Confirmation bias in human reinforcement learning: Evidence from counterfactual feedback processing.

Area of Science:

• Cognitive Science
• Neuroscience
• Decision Making

Background:

• Factual learning, or learning from outcomes, shows a bias towards positive prediction errors.
• The influence of prediction error valence on counterfactual learning (learning from forgone outcomes) remains unexplored.

Purpose of the Study:

• To investigate whether prediction error valence affects counterfactual learning.
• To compare valence-induced biases in factual versus counterfactual learning.

Main Methods:

• Two experiments were conducted using reinforcement learning tasks.
• A computational model analyzed participant performance to assess the impact of prediction error valence.
• Factual learning involved partial feedback; counterfactual learning involved complete feedback.

Main Results:

• Replicated previous findings: factual learning showed a bias for positive prediction errors.
• Counterfactual learning exhibited an opposite bias, favoring negative prediction errors.
• A general tendency to prioritize information confirming current choices was observed.

Conclusions:

• Prediction error valence significantly influences both factual and counterfactual learning, but in opposing directions.
• Individuals exhibit a bias towards information that validates their existing choices, regardless of learning context.