Egocentric Bias in Vision-Language Models
arXiv:2602.15892
The paper introduces FlipSet, a benchmark that tests vision‑language models on Level‑2 visual perspective taking by requiring them to mentally rotate 2D character strings, and finds that most models show strong egocentric bias and struggle to combine social reasoning with spatial transformations.
Abstract
Visual perspective taking--inferring how the world appears from another's viewpoint--is foundational to social cognition. We introduce FlipSet, a diagnostic benchmark for Level-2 visual perspective taking (L2 VPT) in vision-language models. The task requires simulating 180-degree rotations of 2D character strings from another agent's perspective, isolating spatial transformation from 3D scene complexity. Evaluating 103 VLMs reveals systematic egocentric bias: the vast majority perform below chance, with roughly three-quarters of errors reproducing the camera viewpoint. Control experiments expose a compositional deficit--models achieve high theory-of-mind accuracy and above-chance mental rotation in isolation, yet fail catastrophically when integration is required. This dissociation indicates that current VLMs lack the mechanisms needed to bind social awareness to spatial operations, suggesting fundamental limitations in model-based spatial reasoning. FlipSet provides a cognitively grounded testbed for diagnosing perspective-taking capabilities in multimodal systems.
Accepted at CogSci 2026 (Best Undergraduate Student Paper)