Researchers propose a new framework for debiased inference using multiple imperfect AI-generated data measurements without relying on gold-standard labels. Existing methods often combine error-prone AI-based measurements with costly and difficult-to-obtain gold-standard labels to reduce bias in downstream analyses. The proposed framework, DMM, assumes that the measurements are independent conditional on the latent true label and observed unit-level features, allowing for unknown misclassification rates to vary across annotation methods and units. Simulation results show that DMM yields valid inference and can improve efficiency when accurate, though imperfect, measurements are added. --- Why it matters: This matters because it provides a solution to a common problem in AI-generated data analysis: how to reduce bias when the gold-standard labels are not available or too expensive to obtain. The proposed framework has implications for researchers who rely on large language models and other AI-based measurements in their work. Source: https://arxiv.org/abs/2608.18294