{
  "abstract": "Video foundation models now reach human-level accuracy on physical-reasoning benchmarks, yet such tasks require predicting unobserved physical outcomes. Do these models perform human-like forward simulation, or do they exploit statistical regularities in visible scenes? Accuracy alone cannot distinguish these strategies. We introduce a distributional evaluation framework that treats model seeds and human raters as populations, enabling comparison of consensus, uncertainty, and strategy. On the Physion benchmark, we evaluate three ViT-L architectures (V-JEPA2, VideoMAEv2, DINOv2). V-JEPA2 narrows the accuracy gap to ~1 percentage point (73.2% vs. 74.2%), yet model-human disagreement reaches 26.4%, far exceeding human-human disagreement (4.8%), with substantially lower agreement (kappa ~ 0.48 vs. 0.91). The divergence follows forward-simulation demands: models outperform humans on geometric reasoning (linking, +11.8 pp) but underperform on gravitational dynamics (rolling, -11.8 pp) and causal chains (dominoes, -10.5 pp). Strategy fingerprinting confirms all three architectures share non-human strategies while none aligns with humans. Attribution analysis suggests that unobservable outcome features, rather than visible scene properties, predict this divergence, consistent with models relying more on scene-level statistical regularities than on explicit forward simulation, a systematic divergence that accuracy alone cannot reveal. Code is available at https://github.com/fanhong-li/model-human-divergence.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": "Journal reference: Computer Vision -- ECCV 2026, LNCS vol. 17036, pp. 74-90, Springer, 2026",
  "arxiv": "2609.22788",
  "author_total": 6,
  "authors": [
    {
      "name": "Fanhong Li"
    },
    {
      "name": "Shurui Zheng"
    },
    {
      "name": "Zi Yin"
    },
    {
      "name": "Junbo Cui"
    },
    {
      "name": "Lei Ji"
    },
    {
      "name": "Jia Liu"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/8p65rsa3hjh7v.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2609.22788",
  "fulltext_url": null,
  "id": "8p65rsa3hjh7v",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2609.22788",
  "language": null,
  "licence": "CC-BY-4.0",
  "licence_label": "CC BY 4.0",
  "licence_source": "article",
  "licence_url": "http://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.CV"
    ],
    [
      "arxiv",
      "stat.AP"
    ]
  ],
  "openalex": null,
  "pdf_url": "https://arxiv.org/pdf/2609.22788",
  "pmcid": null,
  "pmid": null,
  "published": "2026-09-19",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/8p65rsa3hjh7v.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:23:33Z",
      "native_id": "2609.22788",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2609.22788"
    }
  ],
  "subjects": [
    {
      "field": "mathematics",
      "origin": "arxiv:stat.AP",
      "subfield": "statistics"
    }
  ],
  "title": "Human-Level Accuracy, Non-Human Strategies: Revealing Model-Human Divergence in Video Physical Reasoning",
  "title_text": "Human-Level Accuracy, Non-Human Strategies: Revealing Model-Human Divergence in Video Physical Reasoning",
  "type": "preprint",
  "updated_at": "2026-10-30T12:23:33Z",
  "url": "https://research.draconitic.com/paper/8p65rsa3hjh7v",
  "venue": "arXiv",
  "versions": [
    {
      "doi": "10.1007/978-3-032-37132-4_5",
      "pid": null,
      "relation": "published-as",
      "title": null,
      "venue": null
    }
  ]
}