aiqbench analysis
This research identifies a critical flaw in multimodal model evaluation where the order of conflicting evidence sources alters judgment outcomes, revealing noncommutative behavior in vision and speech-text conflicts. By isolating position as a variable, the study demonstrates that placing perceptual inputs later significantly increases reliance on their content compared to earlier studies with fixed or mixed orders. These findings are essential for refining bias measurement protocols and ensuring robustness in multimodal large language models.
Same evidence, different judgments: Evidence noncommutative in vision/speech-text conflicts
arXiv:2609.26986v1 Announce Type: new Abstract: For multimodal large language models, when images or speech conflict with accompanying text, measured text reliance can entangle modality preference with evidence position. Earlier studies of text bias often used a fixed evidence order or moved task instructions with the evidence, leaving the contribution of order unclear. In this paper, we use a paired comparison that keeps the instructions and evidence content fixed and swaps only the positions of the two sources to quantify this potential influence. Across vision and speech models, placing an image or recording after conflicting text consistently shifts answers toward its content. We also revisit previous studies and analyze why their experimental settings can lead to misleading conclusions. These findings reveal cross-modal evidence noncommutativity: the same evidence can lead to different judgments when its order changes, and placing perceptual evidence later can increase the model's reliance on its content.
Abstract:For multimodal large language models, when images or speech conflict with accompanying text, measured text reliance can entangle modality preference with evidence position. Earlier studies of text bias often used a fixed evidence order or moved task instructions with the evidence, leaving the contribution of order unclear. In this paper, we use a paired comparison that keeps the instructions and evidence content fixed and swaps only the positions of the two sources to quantify this potential influence. Across vision and speech models, placing an image or recording after conflicting text consistently shifts answers toward its content. We also revisit previous studies and analyze why their experimental settings can lead to misleading conclusions. These findings reveal cross-modal evidence noncommutativity: the same evidence can lead to different judgments when its order changes, and placing perceptual evidence later can increase the model's reliance on its content.
Artificial Intelligence (cs.AI)
arXiv:2609.26986 [cs.AI]
(or arXiv:2609.26986v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.26986 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 237 words
