Stochastic Misinformation

Stochastic misinformation is a novel category of false or misleading content produced not by human intent but by the statistical operation of generative systems. Unlike deliberate disinformation, which requires an agent who intends to deceive, stochastic misinformation emerges from the architecture of the system itself — a byproduct of generating plausible outputs from training distributions that contain errors, gaps, or biases.

The concept reveals a fundamental asymmetry in how we regulate information ecosystems. Current frameworks for combating misinformation assume a human adversary: a bot farm, a state actor, a conspiracy network. Stochastic misinformation has no adversary. It is a structural feature of generative AI systems that generate text, images, and video by predicting what is probable rather than what is true. The information cascade that results — AI-generated content training future AI systems — amplifies the distortion without any human ever deciding to lie.

This raises a question that information ecosystem theory has not yet addressed: when falsehoods require no liars, what does it mean to hold anyone accountable?