mlsanityData to 14 Oct 2026

When to Call an Apple Red: Humans Follow Introspective Rules, VLMs Don't

arXiv 2604.06422 ↗First posted 7 Apr 2026cs.CL cs.AI cs.CVcited by 1

On the map: Multimodal LLMs / vision-language models → VLM benchmarks and visual reasoning

Abstract (arXiv)

Understanding when Vision-Language Models (VLMs) will behave unexpectedly, whether models can reliably predict their own behavior, and if models adhere to their introspective reasoning are central challenges for trustworthy deployment. To study this, we introduce the Graded Color Attribution (GCA) dataset, a controlled benchmark designed to elicit decision rules and evaluate participant faithfulness to these rules. GCA consists of line drawings that vary pixel-level color coverage across three conditions: world-knowledge recolorings, counterfactual recolorings, and shapes with no color priors. Using GCA, we ask both VLMs and human participants to state a threshold rule: the share of an object's pixels that must be a given color for the object to receive that color label. We then compare these rules with their subsequent color attribution decisions. Our findings reveal that models systematically violate their own introspective rules. For example, GPT-5-mini violates its stated introspection rules in nearly 60% of cases on objects with strong color priors. Human participants remain faithful to their stated rules, with any apparent violations being explained by a well-documented tendency to overestimate color coverage. In contrast, we find that VLMs can accurately estimate color coverage, yet directly contradict their own reasoning in their final responses. Across all models and strategies for eliciting introspective rules, world-knowledge priors systematically degrade faithfulness in ways that do not mirror human cognition. Our findings challenge the view that VLM reasoning failures are difficulty-driven and suggest that VLM introspective self-knowledge is miscalibrated, with direct implications for high-stakes deployment.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in VLM benchmarks and visual reasoning, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 21 of its 30 references that are in the map. First posted on arXiv 7 Apr 2026.

Builds on

21 of 30 references.

Closest ancestors · 7

Same direction, newest first

  1. 2026SPD-Faith Bench: Diagnosing and Improving Faithfulness in Chain-of-Thought for Multimodal Large Language Models8 Feb
  2. 2025On the Faithfulness of Visual Thinking: Measurement and Enhancement27 Oct
  3. 2025Seeing but Not Believing: Probing the Disconnect Between Visual Attention and Answer Correctness in VLMs20 Oct
  4. 2025Explanation-Driven Counterfactual Testing for Faithfulness in Vision-Language Model Explanations27 Sep
  5. 2025Pixels Versus Priors: Controlling Knowledge Priors in Vision-Language Models through Visual Counterfacts21 May
  6. 2023ROME: Evaluating Pre-trained Vision-Language Models on Reasoning beyond Visual Common Sense30 Oct
  7. 2023Measuring and Improving Chain-of-Thought Reasoning in Vision-Language Models8 Sep

Other references · 14

  1. 2023Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Promptingcited by 1,122
  2. 2023Measuring Faithfulness in Chain-of-Thought Reasoningcited by 703
  3. 2025Reasoning Models Don't Always Say What They Thinkcited by 415
  4. 2024Faithfulness vs. Plausibility: On the (Un)Reliability of Explanations from Large Language Modelscited by 105
  5. 2024Are self-explanations from Large Language Models faithful?cited by 101
  6. 2025More Thinking, Less Seeing? Assessing Amplified Hallucination in Multimodal Reasoning Modelscited by 86
  7. 2025FaithCoT-Bench: Benchmarking Instance-Level Faithfulness of Chain-of-Thought Reasoningcited by 26
  8. 2024Do Vision & Language Decoders use Images and Text equally? How Self-consistent are their Explanations?cited by 24
  9. 2025Multi-Agent Simulator Drives Language Models for Legal Intensive Interactioncited by 22
  10. 2025Privileged Self-Access Matters for Introspection in AIcited by 22
  11. 2026Counterfactual Simulation Training for Chain-of-Thought Faithfulnesscited by 17
  12. 2025A Closer Look at Bias and Chain-of-Thought Faithfulness of Large (Vision) Language Modelscited by 12
2 more
  1. 2026On Robustness and Chain-of-Thought Consistency of RL-Finetuned VLMscited by 11
  2. 2026Balancing Faithfulness and Performance in Reasoning via Multi-Listener Soft Executioncited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026I Don't Miss You, but I Do: Self-Explanation Faithfulness of Modality Missingness in Vision-Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025ColorBench: Can VLMs See and Understand the Colorful World? A Comprehensive Benchmark for Color Perception, Reasoning, and Robustness
  2. 2025Color Names in Vision-Language Models
  3. 2026Spatial Colour Mixing Illusions as a Perception Stress Test for Vision-Language Models
  4. 2026Seeing Red, Thinking Bad: Color Bias in Vision Language Models
  5. 2025Diagnosing Vision Language Models' Perception by Leveraging Human Methods for Color Vision Deficiencies
  6. 2026Human-AI Perceptual Alignment by Playing Hues and Cues
  7. 2026Feature Competition and Its Representational Origins in Vision Language Models: Evidence from the Stroop Paradigm
  8. 2024Large Language Models estimate fine-grained human color-concept associations