AI Predicts Personality Traits From Chat History With Up to 61% Accuracy

Artificial intelligence can predict a user’s personality traits from chat history with an accuracy of up to 61 percent, according to a pre-print study reported by Euronews. Researchers asked 668 ChatGPT users from the United States and the United Kingdom to share copies of their chat histories.

The team collected and analyzed more than 62,000 chats, sorting them by topic before training an AI model to estimate the likelihood that each user exhibited one of the Big Five personality traits: agreeableness, conscientiousness, emotional stability, extraversion and openness.

The same participants completed a standard psychological test that measured their actual personality traits. The fine-tuned AI model matched the test results with up to 61 percent accuracy. It performed best at identifying agreeableness and emotional stability but had more difficulty with conscientiousness.

The AI model produced more accurate predictions when given longer chat histories to examine. The study found that the more a person interacts with the AI system, the more identifiable their personality traits become from the accumulated data. While risks to any single individual remain limited, the researchers stated there are major risks at large scale if such personality data is used by bad actors.

They cited the possibility of large-scale manipulation campaigns that spread disinformation or political propaganda. The researchers hope their findings can support the development of tools that reduce the risk of oversharing personal data with AI systems.

One proposed tool would automatically remove identifying details from user conversations before they are stored or analyzed.