Personality inferred from LinkedIn photos predicts MBA graduates' job matches and pay as strongly as race and education, and workers whose photo-inferred traits fit occupational demands earn higher wages. While the scalable 'Photo Big 5' enables new large-scale research on noncognitive skills, it also risks facilitating covert screening and statistical discrimination.

Main Finding

Using AI to infer Big Five personality traits from facial images (“Photo Big 5”) of 96,000 MBA graduates linked to LinkedIn microdata, the authors show that these image-extracted personality measures predict labor-market outcomes — school rank, job matching, compensation, job transitions, and career advancement. The Photo Big 5 has predictive power comparable to race, attractiveness, and educational background, is only weakly correlated with cognitive test scores, and helps explain systematic occupational sorting and wage gains when worker traits align with occupational demands. The method’s scalability creates new research opportunities but raises serious ethical and policy concerns about screening and discrimination.

Key Points

• Dataset: 96,000 MBA graduates with LinkedIn profiles and facial images.
• Measurement: AI/computer-vision models are used to estimate the Big Five personality traits from photos (the “Photo Big 5”).
• Predictive power: Photo Big 5 predicts school rank, job matching quality, pay, job transitions, and career advancement.
• Comparative strength: Its predictive contribution is similar in magnitude to race, attractiveness, and educational background.
• Distinct from cognition: Photo Big 5 is only weakly correlated with cognitive measures (test scores), indicating it captures noncognitive traits.
• Sorting and returns: Individuals sort into occupations that value their inferred traits and receive higher wages when their traits align with occupational demands.
• Scalability vs. ethics: While scalable measurement enables new empirical work on personality in labor economics, it risks enabling statistical discrimination and threatens individual autonomy if adopted in industry screening.

Data & Methods

• Data sources: High-coverage LinkedIn microdata linked to facial images for a large sample (96k) of MBA graduates.
• Measurement approach: AI/computer-vision models extract estimates of the Big Five personality dimensions from profile photos (termed Photo Big 5).
• Outcomes analyzed: Educational attainment/rank, job matches (occupation and firm fit), compensation/wages, job transition dynamics, and career progression.
• Comparative analyses: Predictive comparisons against other observables (race, attractiveness, education) and correlations with cognitive test scores.
• Mechanism tests: Evidence presented that occupational sorting and trait–occupation alignment explain wage differentials (workers earn more when traits match job demands).

Implications for AI Economics

• Measurement opportunity: Photo-based personality measures provide a scalable proxy for noncognitive skills, enabling large-sample research on personality’s role in labor-market dynamics where survey-based measures are infeasible.
• New research angles: Study of personality-driven sorting, complementarities between traits and job tasks, the role of noncognitive skills in wage determination, and heterogeneity in career mobility using rich administrative or platform-linked data.
• Methodological cautions: Photo-derived traits may contain measurement error, cultural and sample bias, and domain-specific validity limits; economists should assess robustness, cross-population validity, and potential confounding before causal interpretation.
• Labor-market impacts: Widespread adoption by firms could change hiring processes, amplify demand for certain noncognitive traits, alter occupational composition, and affect wage inequality through differential returns to inferred personality.
• Ethical and policy concerns: Risk of statistical discrimination, privacy violations, reduction in individual autonomy, and reinforcement of social bias. Calls for regulation, transparency, auditability, and safeguards (e.g., restrictions on automated screening, accuracy/fairness testing, informed consent).
• Research & policy priorities: Validation studies across populations, causal identification of trait returns, assessment of long-run welfare effects of automated screening, development of fairness-aware tools, and evaluation of legal frameworks to govern use of inferred psychological attributes.