Researchers at the University of Barcelona have demonstrated that artificial intelligence (AI) can identify personality traits from written texts. The results, published in the journal PLOS ONEopen new perspectives for understanding how personality manifests in natural language and for developing more transparent and reliable automatic detection tools.
How AI detects personality traits in texts
The study examines how two AI models, BERT and RoBERTa, detect personality traits from text, according to two psychological frameworks: the Big Five and the MBTI. The researchers used databases of texts already labeled according to these traits, then applied explainable AI methods to understand which linguistic elements influence the predictions. Using the integrated gradients technique, they were able to precisely identify the words and expressions that guide the detection of a trait. This analysis shows, for example, that a word like “hate” can contribute to a positive signal depending on the context (“I hate seeing others suffer”). The authors emphasize that these tools make it possible to “lift the veil” to avoid erroneous interpretations and to verify that predictions are based on psychologically relevant signals, in agreement with established theories.
The limits of the MBTI model
The study also highlighted the limitations of the MBTI model compared to the Big Five model, which presents a more solid basis for both automated personality analysis and classical psychometric assessment. “Although widely used in computer science and some applied areas of psychology, the MBTI model has serious limitations for automatic personality assessment, as our results show that the models tend to rely more on artifacts than on true motives.they note.
Automatic personality detection
Automatic personality detection by AI opens new possibilities in psychology. It makes it possible to identify linguistic patterns linked to personality traits that escape traditional methods, offering more natural assessments adapted to large populations. In the clinic, these techniques can support the assessment and monitoring of patients through the analysis of language development. They also find applications in recruitment, education, social research and in the development of more personalized virtual assistants. Researchers emphasize the need to use scientifically reliable and explainable models to ensure ethical and transparent application. These tools will not soon replace traditional tests, but will complement them in a combined approach combining questionnaires, natural language and other data, for a richer understanding of personality.
Validating research in other contexts
The next steps of the study aim to test the analysis on other types of texts, platforms, languages and cultures, and to extend these methods to other psychological dimensions such as emotions or attitudes. Researchers also plan to integrate multimodal data (text, voice, non-verbal behaviors) using tools such as automatic audio transcription. Finally, they want to collaborate with clinicians and human resources professionals to evaluate the effectiveness and ethical impact of these tools in real-world situations.