A groundbreaking study led by researchers at the University of Barcelona has harnessed artificial intelligence (AI) to reveal how everyday language can be used to detect personality traits, while also making key inroads into understanding how such AI models make their decisions. Using advanced machine learning techniques and a transparent, explainable AI approach known as “integrated gradients,” the research demystifies the inner workings of AI personality assessments. Their findings, recently published in PLOS ONE, could transform how personality is measured and ethically deployed across fields ranging from clinical psychology to education and human resources (source).
For Thai readers, this research arrives at a time when digital transformation and the integration of AI into daily life are accelerating across Southeast Asia. In contexts from student counseling at Thai universities to hiring practices in multinational companies based in Bangkok, the ability to understand individuals through language—be it Thai, English, or other regional dialects—has significant implications. The possibility of conducting personality assessments from natural writing or digital communication, in a transparent and scientifically sound manner, brings both opportunities and ethical considerations as Thailand charts its course in the AI era.
The study analyzed how two advanced language models—BERT and RoBERTa, both famed for their natural language processing capabilities—examined texts to predict personality traits. The researchers focused on two major psychological frameworks: the “Big Five” personality model (covering openness, conscientiousness, extraversion, agreeableness, and emotional stability) and the Myers-Briggs Type Indicator (MBTI), a typological assessment favored by many HR departments and pop psychology resources worldwide. Hundreds of texts, preclassified by indicators within both frameworks, were processed by the AI models, which then highlighted which specific words or linguistic patterns swayed their predictions, thanks to the integrated gradients technique (source).
Integrated gradients allowed the research team to “open the black box” of AI algorithms. This transparency addresses a major criticism of deep learning models—their mysterious inner workings. By pinpointing which words or phrases led to particular personality trait assessments, the models are held to established psychological theory, ensuring their conclusions are not based on statistical quirks but on identifiable, interpretable data. The researchers pointed out an example where the word “hate”—generally a negative marker—could, in context (as in “I hate to see others suffer”), reflect empathy or concern rather than negativity. This nuance would often be lost if algorithms “read” words in isolation. As noted by the study’s lead investigators from the Faculty of Psychology and the Institute of Neurosciences at the University of Barcelona, “Explainability techniques allow us to ‘open the black box’ of algorithms, which ensures that predictions are based on psychologically relevant signals and not on artefacts in the data.”
A key finding was that AI could more reliably detect Big Five traits than MBTI types. The Big Five model, long favored in academic psychology, was found to be more stable and robust in linguistic analysis. In contrast, the MBTI framework, while popular in other fields, showed structural weaknesses and was prone to misleading results due to its less empirically grounded categorization system. As the researchers emphasize, “Despite being widely used in computer science and some applied fields of psychology, the MBTI model has serious limitations for automatic personality assessment, as our results indicate that the models tend to rely more on artefacts than on real patterns” (source).
So, how might these discoveries be relevant for Thailand? Personality assessments are increasingly being integrated into Thai educational and workplace settings, especially in large organizations with multinational links or in high-stakes admission procedures at top universities. If automatic language-based assessment tools—powered by AI and explainable models—become mainstream, they could revolutionize clinical intake questionnaires, employee screening, student counseling, or even the personalization of language-learning apps.
Yet, the incorporation of AI also raises questions. As language and personality are shaped by culture, how will English-trained models interpret Thai written or spoken language? The research team from the University of Barcelona is aware of this limitation, noting the importance of validating these models in various languages and cultural contexts. Further, they advocate integrating multimodal data—including voice or behavioral cues—and collaborating with clinicians and human resources professionals for practical, real-world use, emphasizing the ongoing nature of this work (source).
From a historical and cultural standpoint, Thai society places high value on group harmony, social politeness, and emotional control—traits that may manifest differently in communication compared to Western cultures. The Buddhist-influenced norms of kreng jai (consideration for others) and jai yen (cool-heartedness) often encourage understatement and indirectness in Thai communication. This raises fascinating challenges for AI models trained primarily on Western data: Can these systems be effectively adapted to “decode” personality from Thai texts, or will they misinterpret politeness markers as passivity or lack of assertiveness? For Thailand, developing or fine-tuning AI models on local samples and in local languages will be essential for accurate and ethically sound implementation.
Looking ahead, the researchers envision a future in which personality assessment is multimodal—combining traditional questionnaires, natural language analysis, digital behavior, and other sources for a 360-degree view of individual differences. They caution, however, that AI models will not replace traditional personality tests in the short term, but can supplement them, especially in situations where data collection by conventional means is difficult or when analyzing large volumes of available text. The team is also exploring the use of similar techniques to assess emotional states and attitudes, not just fixed personality traits.
For those in Thai education, healthcare, business, or technology policy, this study underscores the importance of choosing psychometrically validated frameworks like the Big Five over popular, but less reliable, tools like the MBTI. It also emphasizes the ethical imperative of transparency in AI: decision-makers must ensure that algorithms influencing academic placements, hiring, or health interventions are interpretable, fair, and calibrated to local norms.
Practical recommendations for Thai readers are thus twofold. First, institutions considering AI-based personality assessments should prioritize transparency, ensuring that models used are explainable and validated for Thai populations—potentially by partnering with local psychologists and data scientists. Second, ordinary Thais using AI-powered language tools (such as chatbots or resume advisors) should be aware of both their power and their limitations, especially when interacting in Thai rather than English. Teachers and HR professionals can look for emerging Thai-language research or open-source tools that align with best practices established internationally.
Ultimately, the University of Barcelona study marks a major advance in aligning AI personality assessments with psychological science rather than algorithmic guesswork. As Thailand shapes its digital future, bridging international research with local expertise will be the key to ethical, effective deployment—whether in the classroom, the clinic, or the workplace.
Sources: