This post is also available in:
עברית (Hebrew)
Artificial intelligence is making significant strides in psychological profiling, with new research from the University of Barcelona demonstrating how machine learning models can detect and interpret personality traits from text. In a recent study published in PLOS One, researchers explored how AI can not only identify key psychological characteristics but also reveal how these decisions are made—an important step toward transparent, ethical deployment.
Using two leading natural language processing models, BERT and RoBERTa, the team analyzed how AI interprets written language to classify personality traits. These models were trained on datasets containing thousands of text samples previously labeled according to two popular psychological systems: the Big Five personality traits, and the Myers-Briggs Type Indicator (MBTI).
Crucially, the study applied explainable AI techniques—specifically, a method known as “integrated gradients”—to trace the models’ decision-making. This allowed researchers to identify exactly which words or phrases influenced predictions. For example, even emotionally charged words like “hate” might point to empathy rather than hostility, depending on context, according to TechXplore.
The researchers explain that this deep insight into AI reasoning helps avoid misinterpretation and ensures that automated analysis aligns with psychological theory, rather than relying on data artifacts. It also paves the way for more accurate and responsible applications in fields like mental health, human resources, and education.
While the Big Five framework proved more robust for machine-based assessment, the widely used MBTI model showed weaknesses, with models often picking up on superficial patterns rather than genuine personality indicators.
Beyond diagnostics, the implications are far-reaching. AI-driven personality analysis could enable large-scale, low-intrusion profiling based on everyday language—be it social media posts, job applications, or chatbot interactions. It could also support clinicians in monitoring patient progress or flagging psychological shifts through changes in verbal expression.
Looking ahead, researchers aim to expand their models to additional languages, cultural contexts, and psychological constructs like emotions or attitudes. They’re also integrating voice and non-verbal cues for a richer, multimodal understanding of human behavior.
