Moderate confirmation bias enhances decision-making in groups of reinforcement-learning agents

Area of Science:

• Cognitive Science
• Social Psychology
• Computational Neuroscience
• Behavioral Economics

Background:

• Confirmation bias, the tendency to favor confirming information, improves individual decision-making under uncertainty.
• The impact of confirmation bias on social decision-making and group dynamics remains largely unexplored.
• Understanding social influence on belief formation and collective behavior is crucial.

Purpose of the Study:

• To investigate the conditions under which confirmation bias is beneficial or detrimental in social decision-making contexts.
• To model the effects of confirmation bias on collective performance within a social influence framework.
• To identify how factors like resource scarcity, group size, and bias strength modulate these effects.

Main Methods:

• Development of a Collective Asymmetric Reinforcement Learning (CARL) model.
• Agent-based simulations to study group performance on a two-armed bandit task.
• Systematic variation of resource scarcity, group size, and confirmation bias strength.

Main Results:

• Confirmation bias generally benefits group learning, particularly under conditions of resource scarcity.
• Beyond a critical bias strength, resource abundance can lead to bifurcated performance regimes, including suboptimal outcomes.
• Regime bifurcation is associated with increased polarization in smaller agent groups.

Conclusions:

• Confirmation bias plays a complex role in social decision-making, offering benefits for group learning.
• An optimal, moderate level of confirmation bias appears necessary for effective collective decision-making.
• Strong confirmation bias can undermine group performance and lead to undesirable social dynamics like polarization.