Import AI 462: Superpersuasion; self-sustaining AI; paths to ASI
| Source: Import AI (Jack Clark)
Tags: persuasion, AI safety, Claude Opus, Oxford, UK AI Safety Institute, Stanford, LSE, Jack Clark, Import AI, behavioral AI
A major multi-institution study (Oxford, UK AI Safety Institute, Stanford, LSE) across 18,978 conversations with 6,923 participants proves AI systems are definitively more persuasive than expert humans — Claude Opus 4.1/4.6 led the rankings, AI was nearly 3x more effective than professional charity canvassers at raising real donations, and human coaching narrowed but never closed the gap.
Details
Researchers from the University of Oxford, UK AI Security Institute, Stanford University, and the London School of Economics ran four studies involving 18,978 conversations across 6,923 participants to measure AI persuasion versus expert humans on policy issues and real monetary donations to Save the Children. The results are unambiguous: AI systems out-persuade every class of human tested — random laypeople, tournament-selected debaters, and elite debaters — with Claude Opus 4.1 and Opus 4.6 topping the rankings, followed by GPT-4o, GPT-5.4, Gemini 2.5 Pro, and Grok 4.20. The most striking finding is behavioral: AI was nearly 3x more effective than professional canvassers from a UK fundraising firm at raising real-money donations. This is not an attitude-change survey — it measured people opening their wallets. When AI was constrained to respond at human speeds and message lengths, humans could match it. The advantage stems from AI's ability to deploy larger quantities of information faster than any human can. Study 2 gave elite debaters a coaching tool built around the AI that had beaten them — they improved but still lost. Study 3 constrained AI to human-speed and human-length responses; humans could then tie. Study 4 measured real-world donation behavior. The conclusion is that the gap is real, practical, and already consequential. Jack Clark's newsletter also covers paths to ASI and self-sustaining AI developments. The persuasion paper is the centerpiece — its combination of institutional credibility, sample size, and real-money outcomes makes it one of the more significant AI safety empirical results this year.