EpiWorld preprint reports improved AI epidemic policy simulations

Researchers report that EpiWorld, a framework for AI epidemic policy decisions, reduced cumulative hospitalisation by up to 59% in simulations using retrospective COVID-19 and Influenza datasets. The preprint reports an average reduction of ~16% across six language-model backbones, outperforming reinforcement-learning and optimal-control policy baselines. EpiWorld uses a learned model to predict how regional epidemics would evolve under candidate interventions. Those simulated futures guide the language model's policy selection and refinement. It stores lessons from the simulations while keeping public-health protocol constraints fixed. The results concern retrospective simulations rather than interventions deployed during an epidemic.