SPbU Scientists Test AI's Ability to Create Psychological Profiles
SPbU (St. Petersburg State University), a BRICS partner, along with experts from the Artificial Intelligence Research Institute and the Institute of Psychology of the Russian Academy of Sciences, conducted a study on the accuracy of modern language models in predicting psychological characteristics based on standard interviews.
Methodology:
- Participants: 645 individuals took part in the study.
- Data Collection: Each participant answered interview questions and completed psychological tests assessing 38 indicators, including personality traits, values, moral and social attitudes, and views on material possessions.
- AI Models: Five language models were used to predict responses to psychological test questions based on the interview materials.
- Comparison: The AI predictions were then compared with the actual responses of the participants.
Key Findings:
- Accuracies: AI performed better at predicting traits clearly reflected in speech and self-description, such as extraversion.
- Positive Bias: AI tended to view people more positively, overestimating kindness and honesty while underestimating narcissism, Machiavellianism, and psychopathy.
- Challenging Traits: Neural networks found it harder to gauge levels of anxiety, the pursuit of pleasure, and views on fairness.
Quote:
“In an applied context, such technologies can be used for personalized learning, psychological and motivational support, organizational communications, and human interaction with intelligent systems... [and] modelling the behaviour not only of individuals but also of entire teams.” - Anastasia Panfilova, SPbU Research Fellow.
This research continues SPbU's exploration of large language models' capabilities in analyzing human behavior.