Stanford researchers at the Polarization and Social Change Lab conducted a controlled study in which thousands of participants read persuasive arguments on controversial issues (assault weapon bans, carbon tax, paid parental leave, smoking bans, gun control, child tax credit) and rated their support before and after. AI-generated messages via GPT-3 proved at least as persuasive as human-written ones, with effect sizes small on an individual basis (a few points on a 0-100 scale) but potentially significant at population scale on polarized topics. Participants rated AI arguments as more factual and logical and less reliant on emotional storytelling. The study was framed by its authors as a cautionary demonstration of AI's capacity to influence democratic discourse, with the implication that regulatory attention to AI use in political contexts is warranted. For financial institutions, this signals emerging regulatory and reputational risk around AI-enabled mass persuasion campaigns and potential demand for content-authentication and disclosure infrastructure.
213 other tracked developments touching the same claims.