Policymakers in domains such as emergency management, public health, and urban planning often have to make decisions under deep uncertainty. Outcomes depend on how large populations interpret information, coordinate, and adopt new behaviors, which is difficult to predict with traditional models. A new arXiv paper presents WhatIf, a system designed for interactive exploration of large-scale LLM agent simulations to support policy reasoning in these settings.
The paper positions WhatIf as a way to let decision-makers explore simulation results interactively rather than treating them as a black box. By simulating how populations might respond to different policies, the tool aims to give policymakers a clearer view of potential outcomes before committing to a course of action.
WhatIf is notable for combining large-scale agent-based simulation with an interactive interface, making it more practical for real-world policy analysis. The full paper is available on arXiv and describes the system's design and intended use cases.