One distinct—and indeed, I argue defining—feature of social phenomena, as opposed to natural phenomena, is infinite population heterogeneity. In social science, therefore, causal inference is meaningful only for specific populations and is subject to variation across contexts and over time. This heterogeneity also implies that AI-generated data for social science should not be evaluated on the basis of individual-level predictive accuracy, as is common in the AI industry. Instead, I propose a general framework for assessing the validity of such data by returning to the foundational principles of survey research in the social sciences. Just as surveys based on representative samples yield statistics that approximate the corresponding statistical moments of the target population, AI-generated data should likewise be evaluated by their ability to reproduce key statistical moments observed in real populations—such as distributions, associations, and life-course pathways.
