[Paper Review] Human collective intelligence as distributed Bayesian inference

P. M. Krafft, Julia Zheng+6 more|arXiv (Cornell University)|Aug 5, 2016

Opinion Dynamics and Social Influence25 references21 citations

TL;DR

The paper proposes that human collective intelligence emerges through distributed Bayesian inference, where individuals use popularity as a boundedly rational prior to guide decisions, leading to rational group-level belief formation. In a dataset of 50,000 social traders, this mechanism approximates Thompson sampling, demonstrating that collective rationality arises from individual heuristics.

ABSTRACT

Collective intelligence is believed to underly the remarkable success of human society. The formation of accurate shared beliefs is one of the key components of human collective intelligence. How are accurate shared beliefs formed in groups of fallible individuals? Answering this question requires a multiscale analysis. We must understand both the individual decision mechanisms people use, and the properties and dynamics of those mechanisms in the aggregate. As of yet, mathematical tools for such an approach have been lacking. To address this gap, we introduce a new analytical framework: We propose that groups arrive at accurate shared beliefs via distributed Bayesian inference. Distributed inference occurs through information processing at the individual level, and yields rational belief formation at the group level. We instantiate this framework in a new model of human social decision-making, which we validate using a dataset we collected of over 50,000 users of an online social trading platform where investors mimic each others' trades using real money in foreign exchange and other asset markets. We find that in this setting people use a decision mechanism in which popularity is treated as a prior distribution for which decisions are best to make. This mechanism is boundedly rational at the individual level, but we prove that in the aggregate implements a type of approximate "Thompson sampling"---a well-known and highly effective single-agent Bayesian machine learning algorithm for sequential decision-making. The perspective of distributed Bayesian inference therefore reveals how collective rationality emerges from the boundedly rational decision mechanisms people use.

Motivation & Objective

• To develop a formal framework explaining how collective rationality emerges from boundedly rational individual decisions.
• To model human social decision-making as a distributed Bayesian inference process where individuals use social popularity as a prior.
• To validate the model using a large-scale dataset from an online social trading platform with real financial decisions.
• To test whether the observed social influence patterns align with the proposed social sampling mechanism over alternative models.
• To assess whether the interaction between popularity and performance supports the hypothesis of approximate Bayesian inference in groups.

Proposed method

• Proposes a social sampling mechanism where individuals first select options based on popularity (treated as a prior), then commit based on objective evidence.
• Models individual decision-making as approximate Bayesian updating, with popularity encoding historical evidence.
• Formalizes the mechanism as a form of approximate Thompson sampling at the group level, where popularity approximates the posterior distribution.
• Empirically tests the model using a dataset of over 50,000 users on eToro, tracking mimicker counts and performance (30-day ROI) over time.
• Uses regression analysis with interaction terms between previous popularity and performance to test for multiplicative effects.
• Controls for confounding factors such as position bias by analyzing low-popularity users and examining interface design constraints.

Experimental results

Research questions

• RQ1 How do individuals form beliefs in social decision-making contexts when relying on both social cues and objective evidence?
• RQ2 Can a boundedly rational individual decision mechanism lead to rational group-level inference?
• RQ3 Does the interaction between popularity and performance in social platforms reflect an underlying Bayesian inference process?
• RQ4 Is the observed social influence pattern better explained by the proposed social sampling model than alternative models?
• RQ5 To what extent can observational data support causal claims about the mechanism of collective belief formation?

Key findings

• Individuals treat popularity as a prior distribution, using it to guide which options to consider, consistent with a Bayesian interpretation.
• The interaction between previous popularity and performance significantly predicts future popularity changes, with a p-value < 0.05 in regression models.
• The effect is strongest among users with low popularity (e.g., 0 or 1 mimicker), indicating that the mechanism operates even at low visibility levels.
• The observed multiplicative interaction between popularity and performance is unlikely to be explained by position bias due to interface constraints and lack of simultaneous sorting.
• User self-reports from eToro support the model, with advice emphasizing popularity as a starting point and performance as a secondary check.
• The collective behavior approximates Thompson sampling, a well-known Bayesian algorithm for sequential decision-making, indicating rational group-level inference emerges from individual heuristics.