Researchers collected and analyzed over 62,000 conversations with AI models to predict personality traits.
The AI model was able to identify user traits with up to 61% accuracy, particularly successful in predicting agreeableness and emotional stability.
AI was more successful when it had access to longer conversation history, suggesting higher likelihood of personality identification the more users interact with AI systems.
Researchers warn of significant large-scale risks if malicious actors begin exploiting personality data for manipulative campaigns.
The study used the ‘Big Five’ personality traits model: agreeableness, conscientiousness, emotional stability, extraversion, and openness.
Researchers hope the findings can be used to develop tools that reduce the risk of oversharing personal data with artificial intelligence.