Chinese researchers just built "Light Society," a simulation that runs 1 billion AI characters at once, each one modeled after a real person's demographic profile pulled from a massive global survey. These AI agents negotiate, form opinions, argue, trust or distrust each other, and make irrational decisions just like real people do. They'll turn down a good deal out of spite. They get persuaded by popular voices in their network. They change their minds slowly, then all at once when enough of their peers flip. The real story isn't the "are they conscious?" speculation floating around. It's that this is a sandbox for testing how to move public opinion at a planetary scale. You can dial up certain influencers, change how the message is framed, and watch a billion simulated people react. Built by a state-funded research institute in Beijing using real global demographic data. Draw your own conclusions. https://arxiv.org/pdf/2506.12078