AI can outperform humans in predicting correlations between personality items

Area of Science:

• Artificial Intelligence
• Psychology
• Computational Social Science

Background:

• Assessing human personality is crucial for various fields.
• Traditional methods rely on self-report questionnaires and expert interpretation.
• The emergence of AI, including large language models (LLMs), offers new avenues for personality analysis.

Purpose of the Study:

• To evaluate the performance of specialized AI (PersonalityMap) and general LLMs (GPT-4o, Claude 3 Opus) in predicting correlations between personality questionnaire items.
• To compare AI performance against laypeople and academic experts.
• To investigate the "wisdom of the crowds" effect by comparing group median predictions.

Main Methods:

• AI models predicted correlations between personality questionnaire items.
• Performance was compared against median predictions from laypeople and academic experts.
• The study analyzed both individual AI predictions and aggregated "wisdom of the crowds" estimates.

Main Results:

• All AI models significantly outperformed most laypeople and academic experts in individual predictions.
• Aggregated "wisdom of the crowds" estimates improved prediction accuracy.
• Specialized AI (PersonalityMap) and academic expert medians surpassed general LLMs and laypeople medians.

Conclusions:

• Advanced LLMs demonstrate superior predictive capabilities over most individuals in personality research.
• Specialized AI models can achieve expert group-level performance in domain-specific tasks.
• Both AI advancements and human expertise remain vital for the future of personality research.