LLM-Powered Social Digital Twins: A Framework for Simulating Population Behavioral Response to Policy Interventions

TL;DR

This paper introduces Social Digital Twins, a general framework that uses LLM-powered cognitive engines to drive multi-dimensional behavioral outputs for synthetic agents. A calibration layer grounds these outputs in real observational data, enabling accurate population-level predictions and counterfactual policy analysis across domains. In a COVID-19 mobility case study, the calibrated digital twin outperformed a gradient-boosting baseline by 20.7% on macro-average RMSE and demonstrated plausible, monotonic responses to policy variations. The approach offers domain-agnostic policy simulation with interpretable agent reasoning and rigorous validation, suggesting broad applicability from transportation to environmental and economic interventions.

Abstract

Predicting how populations respond to policy interventions is a fundamental challenge in computational social science and public policy. Traditional approaches rely on aggregate statistical models that capture historical correlations but lack mechanistic interpretability and struggle with novel policy scenarios. We present a general framework for constructing Social Digital Twins - virtual population replicas where Large Language Models (LLMs) serve as cognitive engines for individual agents. Each agent, characterized by demographic and psychographic attributes, receives policy signals and outputs multi-dimensional behavioral probability vectors. A calibration layer maps aggregated agent responses to observable population-level metrics, enabling validation against real-world data and deployment for counterfactual policy analysis. We instantiate this framework in the domain of pandemic response, using COVID-19 as a case study with rich observational data. On a held-out test period, our calibrated digital twin achieves a 20.7% improvement in macro-averaged prediction error over gradient boosting baselines across six behavioral categories. Counterfactual experiments demonstrate monotonic and bounded responses to policy variations, establishing behavioral plausibility. The framework is domain-agnostic: the same architecture applies to transportation policy, economic interventions, environmental regulations, or any setting where policy affects population behavior. We discuss implications for policy simulation, limitations of the approach, and directions for extending LLM-based digital twins beyond pandemic response.

Figures (1)

• Figure 1: Social Digital Twin Framework. The architecture is domain-agnostic: policy signals, agent attributes, and observable outputs are instantiated per-domain, while the LLM engine and calibration methodology are general.