AI Now Builds Fake Societies to Test What Really Causes What

Scientists have a long-standing problem: it's hard to know what actually causes what in society. If a neighborhood gets safer after a new program, was it the program, or something else entirely? You can't rewind real life and run it again to check. So a team of researchers tried a new trick — building pretend societies inside a computer, filling them with AI characters who behave like people, and seeing what happens when you change one rule at a time.

Their method is called RePair, described in a new paper. Here's the simple version: instead of writing complicated math to describe how people behave, you just write the rule in ordinary language — something like 'people tend to trust those who look like them' or 'rumors spread faster than facts.' The AI characters then act on those rules. Because you can run the same fake world twice — once with the rule and once without — you get a cleaner sense of whether that rule actually caused the outcome you saw. The team tested this across four simulated worlds built from established social science research.

Three things came out of it. First, plain-English rules did produce measurable group-level effects — the pretend societies really did change. Second, as they added more scenarios, the comparisons between rules became more stable, less like noise. Third, the researchers could trace individual characters' actions and conversations to explain why a group result happened, not just that it did. That last part matters: it turns a black box into something a human can actually inspect and argue with.

So what's the catch? Simulated people are not real people. If your rule is wrong or your fake world is too simple, you get confident-sounding answers that don't hold up outside the computer. The authors say as much — this is a feasibility study, not a crystal ball. Still, it points at a future where cities, companies, and governments test social policies on AI stand-ins first, the way engineers crash-test cars before putting them on the road.