ChatGPT predicts human personality patterns via language training

A study from the Hebrew University of Jerusalem has found that ChatGPT may be able to predict how groups of people respond to personality tests. The research suggests that large language models (LLMs) might acquire an implicit understanding of human personality as a byproduct of being trained on vast amounts of human language.

Researchers utilized GPT-4 to generate personality questionnaires from diverse sources, including the clinically validated DSM-5 and a non-scientifically validated astrology textbook. The AI was able to predict population-level patterns in responses before the human data was collected. Dr. Rotem Monsa, the study's lead author, noted that while the models were not explicitly taught psychology, these traits are "already embedded in the language that LLMs learn from."

This capability could potentially assist psychologists in developing assessments more rapidly. However, the study clarified that ChatGPT does not understand individuals or possess the ability to diagnose mental illness; rather, it captures meaningful patterns in human behavior reflected through communication styles.