Exploring variability in risk taking with large language models.

Area of Science:

• Cognitive Psychology
• Decision Science
• Artificial Intelligence

Background:

• Individual differences in risk-taking are typically studied using psychometric methods analyzing survey data.
• Existing methods correlate behaviors and individuals but lack a deeper explanation of underlying preferences.

Purpose of the Study:

• To identify sources of individual differences in risk-taking behavior.
• To determine how these differences vary across different decision domains and situations.
• To provide a decision-theoretic foundation for understanding psychometric findings.

Main Methods:

• Utilized large language models (LLMs) to quantify everyday risky behaviors by their descriptive attributes or reasons.
• Employed decision models to link these attributes and reasons to individual participant responses.
• Analyzed correlations between behaviors and individuals based on elicited reasons and individual weighting of reasons.

Main Results:

• LLM-based decision models successfully explained correlations between different behaviors by identifying shared reasons.
• These models also explained correlations between individuals by revealing how they weigh specific reasons.
• Demonstrated accurate out-of-sample prediction for numerous everyday behaviors.

Conclusions:

• The proposed approach offers a decision-theoretic framework for psychometric findings in risk-taking research.
• LLMs can generate quantitative representations of decisions, enabling prediction and interpretation of behavioral heterogeneity.
• This methodology has significant theoretical and practical implications for studying individual differences in behavior.