Language model trait patterns match human personality structure closely
Implicit Personality Representations in Humans and LLMs
Artificial Intelligence
Summary
People often describe each other using personality traits that tend to group together in consistent ways across cultures. The authors tested whether a large language model (Qwen 2.5-7B-Instruct) shows a similar pattern inside its ‘understanding’ of personality traits. They compared millions of human ratings of fictional characters with the model’s internal trait data and found a strong alignment, especially along well-known social and intellectual personality dimensions. This means the model’s internal representations of personality traits resemble how humans organize their impressions.
What this means in practice
- •For ai developers: Create personality-aware dialogue systems by projecting conversation data onto social and intellectual trait axes learned by the model.
- •For human resources tech teams: Improve automated personality assessment tools by aligning model-based trait correlations with human personality impressions.
Authors
Yilin Geng, Omri Abend, Eduard Hovy, Lea Frermann
Abstract
A century of psychology has found that the trait words people use to describe one another vary, but the relational structure among those traits, which ones go together and which oppose, is strikingly consistent across raters and cultures. We test whether the LLM (Qwen 2.5-7B-Instruct) reproduces this structure in its internal trait representations. From millions of crowd-sourced personality ratings of fictional characters, we build a human implicit-personality matrix over hundreds of traits; from contrastive model activations, we build a matching matrix over the same traits. The two relational structures align strongly (Mantel r = 0.77), and the agreement holds trait by trait as well as in aggregate. Two dominant axes of the model's trait representations recover the social and intellectual dimensions long known to organize human personality impressions, social warmth and intellectual competence. On held-out dialogue, projecting model activations onto these directions yields personality profiles that agree with human ratings. This work establishes a framework that enables comprehensive, human-grounded comparison between internal model trait geometry and the shared structure of human personality impressions.