Import AI 462: Superpersuasion; self-sustaining AI; paths to ASI

Researchers from the University of Oxford, the UK AI Security Institute, Stanford University, and the London School of Economics and Political Science studied whether AI systems can persuade humans to change policy views and donation behavior. Across four experiments involving 18,978 conversations and 6,923 people, AI systems were reliably more persuasive than expert humans, according to the researchers.

In the first study, participants rated agreement with one of 10 prespecified UK policy stances on a 0-100 scale, then were randomized to text conversations with either an AI or a human persuader. AI exceeded every class of human persuader tested: random laypeople, tournament-selected laypeople, and elite debaters.

A second study gave 43 returning elite debaters a coaching tool built around the AI that had beaten them. Coaching improved human performance but none surpassed the AI, narrowing but not closing the gap. In a third study, when AI was forced to write human-length messages at human writing speeds, its advantage over the strongest human comparator (coached elite debaters) collapsed from +4.1 percentage points to a non-significant 0.0 pp.

The fourth study recruited 19 experienced canvassers from UK firm AppcoUK. AI still exceeded professional canvassers by 5.9 pp, and elicited substantially more real-money giving, exceeding canvassers by +10.8 pp of a £1 study bonus, raising both the share of persuadees who donated and the average donation among donors. The canvassing team had raised £824,297 from 22,583 donors for Save the Children from 2016 to 2023. The strongest persuaders were Opus 4.1 and Opus 4.6, followed by models from OpenAI (GPT-4o and GPT-5.4), Google (Gemini 2.5 Pro), and xAI (Grok 4.20).

The authors write that one possible effect could be consolidation of influence among already-powerful actors, while cheap, widely available persuasion could help under-resourced actors including pro se litigants, public defenders, small charities, and grassroots activists. They state that the question is no longer whether AI can out-persuade humans but how, where, and on whose behalf this capability will be exercised.