Resampling reduces bias amplification in experimental social networks

Area of Science:

• Social Psychology
• Computational Social Science
• Behavioral Economics

Background:

• Large-scale social networks are hypothesized to increase societal polarization by amplifying individual biases.
• The complex nature of these digital platforms obscures the specific mechanisms driving bias amplification and hinders the development of effective mitigation strategies.

Purpose of the Study:

• To investigate the causal impact of social network transmission on motivational bias amplification in a controlled setting.
• To develop and evaluate a computational strategy for mitigating bias amplification within social networks.

Main Methods:

• A large-scale behavioral experiment was conducted using a simple artificial decision-making task under controlled laboratory conditions.
• Participants were assigned to either social network or asocial conditions across 40 independently evolving populations.
• A content-selection algorithm adjustment, inspired by Bayesian statistics, was designed to promote representative sampling of perspectives.

Main Results:

• Social network participation led to significantly increased rates of biased decision-making compared to asocial participants.
• The proposed algorithmic adjustment effectively reduced bias amplification in two large experiments.
• The mitigation strategy successfully maintained the benefits of information sharing while reducing bias.

Conclusions:

• Social network structures demonstrably amplify motivational biases in decision-making tasks.
• A computationally derived strategy can effectively counteract bias amplification in social networks.
• This approach offers a promising method for enhancing the health of online information ecosystems.