Scholarlink

<p>Visual language models (VLMs) show remarkable performance in visual reasoning tasks, successfully tackling college-level challenges that require a high-level understanding of images. However, some recent reports of VLMs struggling to reason about elemental visual concepts such as orientation, position, continuity and occlusion suggest a potential gulf between human and VLM vision. Currently, few assessments enable a direct comparison between human and VLM performance, which limits our ability to measure alignment between the two systems. Here we use the toolkit of neuropsychology to systematically evaluate the capabilities of three state-of-the-art VLMs across low, mid and high visual domains. Using 51 tests drawn from 6 clinical and experimental psychology batteries, we characterize the visual abilities of leading VLMs relative to normative performance in healthy adults. While the models excel in straightforward object recognition tasks, we find widespread deficits in low- and mid-level visual abilities that would be considered clinically significant in humans. These selective deficits, profiled through validated test batteries, suggest that an artificial system can achieve complex object recognition without developing foundational visual concepts that in humans require no explicit training.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Visual language models show widespread visual deficits on neuropsychological tests

  • 期刊论文

  • 发表日期：

    2026年2月6日

查看全文
  • Gene Tangtartharakul,
  • Katherine R. Storrs

摘要

Visual language models (VLMs) show remarkable performance in visual reasoning tasks, successfully tackling college-level challenges that require a high-level understanding of images. However, some recent reports of VLMs struggling to reason about elemental visual concepts such as orientation, position, continuity and occlusion suggest a potential gulf between human and VLM vision. Currently, few assessments enable a direct comparison between human and VLM performance, which limits our ability to measure alignment between the two systems. Here we use the toolkit of neuropsychology to systematically evaluate the capabilities of three state-of-the-art VLMs across low, mid and high visual domains. Using 51 tests drawn from 6 clinical and experimental psychology batteries, we characterize the visual abilities of leading VLMs relative to normative performance in healthy adults. While the models excel in straightforward object recognition tasks, we find widespread deficits in low- and mid-level visual abilities that would be considered clinically significant in humans. These selective deficits, profiled through validated test batteries, suggest that an artificial system can achieve complex object recognition without developing foundational visual concepts that in humans require no explicit training.