Survey Proposes Framework for LLMs as Human Psychology Measurement Tools

A new survey from researchers—including Yudong Li, Xiaoyi Chen, and others—systematically reviews how Large Language Models can perceive and measure complex human psychological constructs like personality, emotions, and cognitive states. Published on arXiv and accepted by IEEE Transactions on Cognitive and Developmental Systems (TCDS), the paper proposes a comprehensive analytical framework built on three dimensions: Theoretical Plausibility (examining whether LLMs have emergent cognitive properties that enable measurement), Measurement Methodology (categorizing approaches into active conversational assessment, passive natural language analysis, and multimodal fusion), and Application Effectiveness (reviewing real-world results in personality trait assessment and mental health evaluation). This survey stands apart from prior work by focusing on the psychometric properties of LLMs as measurement instruments, rather than on general applications or the 'psychology' of the models themselves.

The researchers critically analyze existing paradigms and discuss limitations, including bias, reproducibility, and the gap between statistical patterns and true psychological understanding. They find that while LLMs show surprising accuracy in some personality assessments and mental health screenings, significant challenges remain—particularly around validity, standardization, and ethical deployment. The framework aims to guide future development of AI-driven psychological assessment tools, potentially enabling scalable, low-cost mental health monitoring and personalized interventions. The paper also notes the need for multimodal integration (e.g., combining text with speech or facial expressions) to improve accuracy. For tech professionals, this work signals a growing convergence of AI and psychology, with implications for HR tech, digital therapeutics, and user experience analytics.