Czy AI może zidentyfikować dominujące cechy osobowości osoby na podstawie 30-sekundowej próbki tekstu z dokładnością dorównującą wyszkolonym psychologom ?

Modele dużych języków analizują wzorce językowe, aby wnioskować o typach Myersa-Briggsa lub Wielkiej Piątki. Badania wykazują silną korelację z samoopisem cech oraz ocenami obserwatorów. Dokładność poprawia się wraz ze wzrostem długości tekstu.

Background

Large language models analyze language patterns to infer Myers-Briggs or Big Five traits. Studies show strong correlation with self-reported traits and observer ratings. Accuracy improves when text length increases.

--- Current AI systems can infer broad personality traits such as the Big Five from brief text samples, and in some studies they match or exceed the accuracy of human experts when predicting traits like neuroticism, conscientiousness, or extraversion on samples as short as a few sentences. Techniques typically combine large language models fine-tuned on personality-annotated corpora with psycholinguistic features like LIWC categories, achieving around 0.3–0.4 correlation with ground-truth scales—comparable to inter-rater reliability between trained psychologists. However, these models rely on self-report questionnaires for training labels, which may not capture unconscious or context-sensitive traits, and performance drops when the writing sample contains atypical vocabulary, sarcasm, or cultural references not well represented in the training data. Ethical and privacy concerns also limit real-world deployment without explicit consent and robust safeguards.

— Enriched May 12, 2026 · Source: Matz et al., “Deep learning reveals predictive models of human language for personality assessment,” PNAS Nexus, 2023