summary: Researchers have shown that AI can detect personality traits from written texts and, importantly, understand how these models make decisions. By applying explanatory AI techniques such as integrated gradients, the team has identified how specific words and linguistic patterns contribute to predictions based on key psychological frameworks.
This study found that Big Five traits were more reliably detected than MBTI types, with the former being more appropriately tuned with verbal markers of behavior. These insights could pave the way for transparent, ethical personality assessments in psychology, HR, education and digital platforms.
Important facts:
- Explainable AI: Using integrated gradients, which words influenced personality predictions and opened the “black box” of AI decisions.
- Big Five vs MBTI: The Big Five model has proven to be more reliable and psychologically evidenced for AI-based personality analysis than MBTI.
- Practical: Findings could enhance clinical assessment, education personalization, HR processes, and adaptive AI assistants.
sauce: University of Barcelona
Researchers at the University of Barcelona show how artificial intelligence (AI) models can detect personality traits from written texts, and for the first time they were able to analyze in detail how these systems make decisions.
These results are published in the journal PLOS 1opens up new perspectives to understand how personality manifests in natural language, and also how we can build more transparent and reliable auto-detection tools.
This paper has been signed by three UB experts. David Seteros and David Garraldo Pujol, researchers and directors of the Individual Difference Lab Research Group (IDLAB) at the Faculty of Psychology and the Institute of Neuroscience (Ubunero), and Daniel Ortiz Martinez, researchers in Massatic and computer science, respectively.
Open the “black box” of the algorithm
In this study, we analyzed how two advanced AI models, Bert and Roberta, process textual data. Detect personality traits according to two main psychological frameworks: Big Five Personality Characteristics System (Experience, Responsibility, Extraversion, Emotional Stability, Emotional Stability), and Myers Briggs Type Indicator (MBTI), people are categorized along the instruments along that level of intext, along with people along that level of people. And touches on thoughts.
“In psychology, there are general models of personality and other unvalidated models that we use to understand and measure individual differences in behavior, emotions, and thought,” the researchers explain these two psychological frameworks.
The text analyzed in this study was obtained from two databases fed to questionnaires for both models (Big Five and MBTI). This was previously classified according to the presence of different personality traits and the types of indicators that constitute them.
Researchers then observed AI models using explanatory AI techniques to see which language patterns influence the identification of personality traits of these works.
“The explanability technique allows the algorithm to “open a black box.” This ensures that predictions are based on psychologically relevant signals, not on artifacts of the data.”
Specifically, they used the technique called Integrated gradientallowing you to accurately identify which words or phrases contribute to the prediction of a particular personality trait.
“This method allowed us to visualize and quantify the importance of different linguistic elements in model prediction,” they say. For example, they observed such words dislikewhich was traditionally associated with negative traits, but may actually appear in a context that reflects kindness (“I hate seeing others suffer”).
\ “We may draw false conclusions without understanding how the model interprets these words in context.”
This approach guarantees the scientific validity of the performance of AI models. “We provide a solid basis for continuous improvement by ensuring that the model is based on linguistic patterns truly related to psychological constructs intended to verify and measure whether it is consistent with established psychological theories,” he added.
MBTI Model Limitations
This study also highlighted the limitations of the MBTI model compared to the Big Five One, which provides stronger evidence for both automated personality analysis and classical psychometric analysis.
“Despite being widely used in several applications in computer science and psychology, MBTI models suffer from severe limitations on automated personality assessment, as the results show that models tend to rely on artifacts rather than actual patterns,” they point out.
Automatic personality detection application
The use of automated personality detection techniques using AI models can have a significant impact on the field of personality psychology.
“In these methods, psychologists identify linguistic patterns associated with various personality traits that may be noticed in traditional ways, which could lead to more natural and uninterrupted assessment methods.
In the clinical field, the authors point out that “initial evaluation and follow-up of patients can be supported by focusing on “focusing on language changes or verbal expressions as indicators of key psychological components of treatment.”
They also point out that they can play an important role in other fields: in personnel choice, education personalization, and social studies, it facilitates the analysis of large volumes of textual data as it helps to create more natural and adapted interactions.
“It is important to emphasize that all such applications are based on scientifically sound models and that the explanationability techniques we explore must be incorporated to ensure ethical and transparent use,” he adds.
Despite the possibilities, researchers believe that these models do not replace traditional personality tests in the short term, but complement them and provide a deeper perspective.
“We see evolution towards a multimodal approach to a more complete situation of personality, combining traditional assessment with natural language analysis, digital behavior, and other data sources,” they note.
Researchers say this integrative approach is built on the strengths of each methodology and provides a “richer and subtle view of human personality.”
In this sense, AI models add that “it is especially useful in contexts where traditional data collection is difficult, or when large amounts of information need to be efficiently analyzed.”
Verification of research in other contexts
The next step in this study involves extending the analysis to other text types, platforms, languages and cultures to see whether identified patterns are consistent in different contexts. Researchers want to apply the application of these techniques to other psychological constructs beyond personality, such as emotional states and attitudes.
Researchers are also working to integrate multimodal data into these analyses – combining text with other forms of representation, such as speech and nonverbal behavior, using technologies such as automatic audio transcription (whisper.ai), and applications in real-world contexts.
The team said, “We want to work with clinicians and HR experts to assess the effectiveness of these tools in a real-world setting and ensure that they have a positive and ethical impact.
About this AI and personality research news
author: Rosa Martinez
sauce: University of Barcelona
contact: Rosa Martinez – University of Barcelona
image: This image is credited to Neuroscience News
Original research: Open access.
David Saeteros et al. PLOS 1
Abstract
Text speaks loudly: Insight into personality from natural language processing
In recent years, advances in natural language processing (NLP) have enabled new approaches to personality assessment.
This article presents interdisciplinary research that utilizes explanatory AI techniques, in particular integrated gradients, to scrutinize the decision-making process in character assessment of NLP models and test their alignment with established personality theories.
Use essays and MBTI datasets to compare the effectiveness of typology (MBTI) and dimension (Big Five) models.
Our methodology applies the log ODDS ratio to beneficial Dirichlet Prior (IDP) and fine-tuned transformer-based models (Bert and Roberta) to classify personality traits from textual data.
Our results show moderate to high accuracy in personality predictions, and the NLP model effectively identifies textual personality signals in line with previous research.
Our findings revealed theoretical collaborative patterns of language use related to different personality traits and highlighted important biases in the MBTI data set, resulting in no robust results.
This study highlights the potential of NLPs to enhance personality psychology and the need for more interdisciplinary research to fully realize the capabilities of these transparent technologies.
