Causal Evidence for Attention Head Imbalance in Modality Conflict Hallucination
Quick Take
Study reveals attention head imbalance in MLLMs leads to modality-conflict hallucinations.
Key Points
- Identifies hallucination-driving and resisting attention heads.
- Driving heads are more distributed, resisting heads are localized.
- MACI intervention reduces hallucinations effectively during generation.
📖 Reader Mode
~2 min readAbstract:Modality-conflict hallucination occurs when multimodal large language models (MLLMs) prioritize erroneous textual premises over contradictory visual evidence. To understand why visual evidence fails to prevail during generation, we take a mechanistic perspective and examine which internal components drive or resist this failure. We perform head-level causal analysis using path patching across five open-source MLLMs and identify two groups of attention heads with opposing causal roles: hallucination-driving heads and hallucination-resisting heads. We find a consistent asymmetry: driving effects are more broadly distributed and carry greater aggregate weight, whereas resisting effects concentrate in a small number of high-importance heads. Ablation experiments further confirm that these groups exert opposing effects during generation: distributed driving influence and localized resistance together form an imbalanced routing structure that biases generation toward the erroneous premise. Motivated by this finding, we propose MACI (Modality-conflict-Aware Causal Intervention), a conditional intervention that suppresses causally identified hallucination-driving heads only when conflict is detected. Across five MLLMs, MACI achieves the largest hallucination reduction among compared inference-time baselines on the MMMC benchmark with a favorable hallucination-accuracy trade-off, and transfers zero-shot to the SCI-SemanticConflict test.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.19250 [cs.AI] |
| (or arXiv:2605.19250v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2605.19250 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jinrui Jiang [view email]
[v1]
Tue, 19 May 2026 01:47:53 UTC (311 KB)
— Originally published at arxiv.org
More from arXiv cs.AI
See more →From Prompts to Protocols: An AI Agent for Laboratory Automation
An AI agent integrates large language models for automating laboratory protocols, enhancing efficiency and accuracy.