Researchers build a way to generate realistic fake people for simulations — matching both real-world statistics and individual behavior

A research team created a method to build populations of simulated agents that match real-world demographic patterns while behaving like actual people. Instead of either copying from existing datasets (which fails when the topic is new) or using language models to generate agents (which ignore real statistical distributions), this approach does both at once, reducing errors in population accuracy by 38% and improving how believable individual agents behave.

Agent-based simulations run everything from epidemiology models to economic forecasting to policy testing. Until now, populating these simulations required either hand-tuning agents to match real behavior (slow and expensive) or accepting that your simulated people don't actually match the statistics of the real population they represent. This means simulations can now be spun up faster for new topics without sacrificing either statistical accuracy or behavioral realism. The practical effect: policy modelers, urban planners, and disease modelers can test scenarios on populations that actually look like the places they're modeling.