MM-SCALE: Grounded Multimodal Moral Reasoning via Scalar Judgment and Listwise Alignment
This work addresses the challenge of aligning VLMs with human moral preferences in multimodal scenarios, offering an incremental improvement over prior binary methods.
The paper tackled the problem of Vision-Language Models struggling with moral reasoning in ambiguous contexts by introducing MM-SCALE, a dataset with 5-point scalar ratings and modality grounding, resulting in VLMs achieving higher ranking fidelity and more stable safety calibration compared to binary-trained models.
Vision-Language Models (VLMs) continue to struggle to make morally salient judgments in multimodal and socially ambiguous contexts. Prior works typically rely on binary or pairwise supervision, which often fail to capture the continuous and pluralistic nature of human moral reasoning. We present MM-SCALE (Multimodal Moral Scale), a large-scale dataset for aligning VLMs with human moral preferences through 5-point scalar ratings and explicit modality grounding. Each image-scenario pair is annotated with moral acceptability scores and grounded reasoning labels by humans using an interface we tailored for data collection, enabling listwise preference optimization over ranked scenario sets. By moving from discrete to scalar supervision, our framework provides richer alignment signals and finer calibration of multimodal moral reasoning. Experiments show that VLMs fine-tuned on MM-SCALE achieve higher ranking fidelity and more stable safety calibration than those trained with binary signals.