Wenqian Xing

Wenqian Xing

Pairwise human comparisons are a primary interface through which modern AI systems learn human preferences. RLHF and related alignment pipelines typically model such comparisons with Bradley--Terry log-odds, where choice probabilities are governed by latent reward differences. This paper examines what this assumption misses through a reduced-form model motivated by rational inattention, in which each label is generated by a low-capacity evaluation channel. The model separates two forms of ambiguity that standard reward modeling tends to conflate: a comparison may be difficult because the two candidates are genuinely close in value, or because the relevant distinction is hard to detect under limited attention. We show that limited attention can fundamentally distort what pairwise comparisons reveal. In particular, passive comparison data cannot generally distinguish reward, attention, and default tendencies, and heterogeneous attention can make standard Bradley--Terry reward modeling recover misleading rankings. Our analysis shows that learning is governed not by the raw number of labels, but by the amount of attended information each label carries. A case study on human votes over language-model pairs from Chatbot Arena exhibits the predicted signature, a cyclic component of the comparison data that exceeds sampling noise and that no scalar reward can represent; a second case study on perceptual comparisons shows that response times and gaze carry gap information that the labels do not. This perspective suggests that human feedback should be treated not as direct revealed preference, but as an attention-limited measurement process: a weak preference signal may reflect hidden evaluation difficulty rather than genuine indifference.

Wenqian Xing, JungHo Lee, Chong Liu et al.

Black-box optimization (BBO) has become increasingly relevant for tackling complex decision-making problems, especially in public policy domains such as police redistricting. However, its broader application in public policymaking is hindered by the complexity of defining feasible regions and the high-dimensionality of decisions. This paper introduces a novel BBO framework, termed as the Conditional And Generative Black-box Optimization (CageBO). This approach leverages a conditional variational autoencoder to learn the distribution of feasible decisions, enabling a two-way mapping between the original decision space and a simplified, constraint-free latent space. The CageBO efficiently handles the implicit constraints often found in public policy applications, allowing for optimization in the latent space while evaluating objectives in the original space. We validate our method through a case study on large-scale police redistricting problems in Atlanta, Georgia. Our results reveal that our CageBO offers notable improvements in performance and efficiency compared to the baselines.