Paper 2605.22109
Perception or Prejudice: Can MLLMs Go Beyond First Impressions of Personality?
- Published
- May 2026
- Research lab
- Independent
- Citations
- 0
- GitHub
- 11 stars
01 In brief
Summary
This paper introduces Grounded Personality Reasoning (GPR) and the MM-OCEAN benchmark to evaluate whether Multimodal Large Language Models (MLLMs) perceive personality through behavioral evidence or merely prejudge via superficial patterns.
The benchmark includes 1,104 videos and 5,320 cue-grounding MCQs, built via a multi-agent human-collaborative pipeline.
A three-tier evaluation (rating, reasoning, grounding) and four failure-mode metrics (Prejudice Rate, Confabulation Rate, Integration-failure Rate, Holistic-Grounding Rate) are used to assess 27 MLLMs.
Results reveal a 'Prejudice Gap': 51% of correct ratings are not grounded in retrieved cues, and Holistic-Grounding Rate ranges only 0–33.5%.
The closed-source frontier shows ~15% ungrounded correct ratings, while the open-source frontier shows ~47%.
The gap between closed and open models is small for rating and reasoning but large for cue retrieval (-26.6%).
The paper identifies two failure archetypes: 'Confident Raters' (good rating, poor grounding) and 'Cautious Reasoners' (good grounding, poor rating).
The authors argue that traditional rating-only evaluations overestimate competence and emphasize the need for fine-grained spatiotemporal grounding in post-training.
02 From the paper
Abstract
Multimodal Large Language Models (MLLMs) are increasingly deployed in human-facing roles where personality perception is critical, yet existing benchmarks evaluate this capability solely on numerical Big Five score prediction, leaving open whether models truly perceive personality through behavioral understanding or merely prejudge through superficial pattern matching. We address this gap with three contributions. (i) A new task: we formalize Grounded Personality Reasoning (GPR), which requires MLLMs to anchor each Big Five rating in observable evidence through a chain of rating, reasoning, and grounding. (ii) A new dataset: we release MM-OCEAN (1,104 videos, 5,320 MCQs), produced by a multi-agent pipeline with human verification, with timestamped behavioral observations, evidence-grounded trait analyses, and seven categories of cue-grounding MCQs. (iii) Benchmark and analysis: we design a three-tier evaluation (rating, reasoning, grounding) plus four sample-level failure-mode metrics: Prejudice Rate (PR), Confabulation Rate (CR), Integration-failure Rate (IR), and Holistic-grounding Rate (HR), and benchmark 27 MLLMs (13 closed, 14 open). The analysis uncovers a striking Prejudice Gap: across the field, 51% of correct ratings are not grounded in retrieved cues, and the Holistic-Grounding Rate spans only 0-33.5%. These findings expose a disconnect between getting the right score and reasoning for the right reason, charting a roadmap for grounded social cognition in MLLMs.