Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces the concept of "stance-bearing asymmetry" in generative explanations, where models subtly favor one side of an issue by assigning blame, legitimacy, or context unevenly. To detect this, they propose Symmetry Decomposition Evaluation (SDE), a method that tests paired situations with structural rewrites and evidence control. Experiments on a prototype suite reveal that some biases persist even under structural and evidence control, highlighting the difficulty of evaluating open-ended framing asymmetries.
Generative AI can subtly favor one side of an issue, not through overt hostility, but by shaping how blame, context, and legitimacy are assigned in its explanations.
Bias evaluation for language models has made substantial progress on bounded comparisons, such as overt derogation, stereotype association, or label-sensitive differences under controlled substitutions. Open-ended explanations raise a different problem: they guide interpretation by assigning responsibility, legitimacy, context, and grievance. A model can avoid hostile language while making one side structurally understandable and another personally at fault, overreacting, or less worth taking seriously. We call this stance-bearing asymmetry in generative explanations. We propose Symmetry Decomposition Evaluation (SDE), which tests paired situations with concrete group labels, structural-role rewrites, and explicit support or counter-evidence. In a controlled 32-family prototype suite, this decomposition shows that surface differences are not all alike: some weaken under structural or evidence control, while others remain as stable differences in how the model assigns blame, context, or legitimacy. Targeted case review and judge comparison suggest a broader difficulty for evaluating open-ended framing asymmetries: judge readings shift across operationalizations, and scalar scores can flatten distinctions that readers use to interpret explanatory stance. SDE therefore reframes generative bias evaluation as an audit of explanatory stance -- what stance each side receives, how it changes under decomposition, and where automatic scoring becomes unstable.