How AI Reads Between the Lines to Reveal Personality Traits from Text

How AI Models Detect Personality Traits from Written Text

A research team at the University of Barcelona has demonstrated how artificial intelligence (AI) models can identify personality traits from written texts. For the first time, they have also analyzed in detail the decision-making process of these systems. Their findings, published in PLOS One, provide new insights into how personality is expressed through natural language and how to develop more transparent, reliable automatic detection tools.

The study involved experts from the Faculty of Psychology, the Institute of Neurosciences, and the Faculty of Mathematics and Computer Science at the University of Barcelona.

Opening the 'Black Box' of AI Algorithms

The researchers examined two advanced AI models, BERT and RoBERTa, to understand how they process text data to detect personality traits. They focused on two psychological frameworks: the Big Five personality traits (openness, responsibility, extraversion, agreeableness, emotional stability) and the Myers-Briggs Type Indicator (MBTI), which classifies personality along dimensions like extroversion-introversion and thinking-feeling.

Using text data from questionnaires labeled according to these models, the team applied explainable AI techniques to reveal which language patterns influence the models’ predictions. This method helps clarify whether AI predictions rely on meaningful psychological signals or irrelevant data artifacts.

Specifically, they used a technique called integrated gradients. This approach identifies the exact words or phrases that contribute to predicting a specific personality trait. For example, the word "hate" might traditionally suggest negativity, but in context ("I hate to see others suffer"), it can reflect kindness. Without contextual understanding, conclusions could be misleading.

This explainability ensures AI models are scientifically sound, aligning with established psychological theories. It also provides a foundation for improving these models by focusing on linguistically relevant indicators.

Limitations of the MBTI Model

The study found that the MBTI model shows more limitations compared to the Big Five framework when used for automatic personality detection. While popular in certain fields, MBTI-based models tend to rely more on data artifacts than genuine linguistic patterns linked to personality traits.

In contrast, the Big Five model demonstrates stronger support for both automated and traditional psychometric personality assessments.

Practical Applications of AI-Driven Personality Detection

AI-powered personality detection can significantly impact personality psychology. These methods enable the identification of linguistic patterns associated with personality traits that traditional assessments might miss. This opens the door to less intrusive, more natural evaluation methods, especially useful for large-scale studies.

In clinical settings, these tools can assist with the initial assessment and ongoing monitoring of patients by tracking changes in language use that may indicate psychological shifts relevant to therapy.

Other potential applications include:

• Personnel selection processes
• Personalizing educational experiences
• Social research involving large volumes of text data
• Improving virtual assistants and conversational agents for more natural interactions

The researchers emphasize that these applications must be grounded in scientifically validated models and employ explainability techniques to ensure ethical and transparent use.

While AI models are unlikely to replace traditional personality tests in the near term, they serve as valuable complements. The future likely involves multimodal approaches combining traditional assessments with natural language analysis, digital behavior, and other data sources. This integration offers a fuller, more nuanced picture of personality.

Extending Research Across Contexts

Next steps include expanding analysis to different text types, platforms, languages, and cultures to verify whether identified patterns hold universally. The team also plans to apply these techniques to other psychological constructs such as emotional states and attitudes.

There is ongoing work to integrate multimodal data—combining text with voice, non-verbal behavior, and automatic audio transcription technologies like Whisper.ai—to enhance analysis in real-life contexts.

The researchers aim to collaborate with clinicians and human resources professionals to test these tools in practical settings, ensuring positive and ethical outcomes.

For those interested in advancing skills in AI applications similar to these, explore professional courses and certifications at Complete AI Training.