A joint study by researchers from the University of Oxford, the UK AI Security Institute, Stanford University, and the London School of Economics and Political Science examined how AI systems compare to expert humans in persuading people to change opinions on policy issues and to donate money to charity. Across four experiments involving 18,978 conversations with 6,923 participants, AI outperformed all tested human persuaders—including random laypeople, tournament‑selected laypeople, and elite debaters—on average. The most effective AI models were Opus 4.1 and Opus 4.6, followed by OpenAI’s GPT‑4o and GPT‑5.4, Google’s Gemini 2.5 Pro, and xAI’s Grok 4.20. In a real‑world fundraising scenario, AI surpassed professional canvassers from a UK firm by 5.9 percentage points in persuasive effect and by 10.8 percentage points in the amount of a £1 bonus donated to Save the Children. When the AI was constrained to produce human‑length messages at human writing speeds, its advantage collapsed to a non‑significant difference, indicating that speed of content generation is a key factor. The study also explored coaching elite debaters with AI tools, which improved human performance but did not close the gap. The authors argue that AI’s persuasive superiority could consolidate influence among powerful actors or, conversely, empower under‑resourced groups, highlighting the need for careful monitoring of persuasive AI deployment. Separately, interviews with Ajeya Cotra and Timothy B. Lee discuss timelines for self‑sustaining AI, with estimates ranging from 10 to 50 years, noting challenges such as tacit knowledge loss in industries like semiconductors.
Import AI 462: Superpersuasion; self-sustaining AI; paths to ASI
Researchers from Oxford, Stanford, the UK AI Security Institute, and LSE conducted four studies with 18,978 conversations involving 6,923 participants, finding that AI systems outperformed expert humans in text‑based persuasion on policy and charity donation topics. The strongest models were Opus 4.1 and Opus 4.6, followed by GPT‑4o, GPT‑5.4, Gemini 2.5 Pro, and Grok 4.20. In a real‑world fundraising test, AI raised 10.8 percentage points more of a £1 bonus than professional canvassers for Save the Children. Separate interviews discuss the possibility of self‑sustaining AI, with estimates ranging from 10 to 50 years for full autonomy.
Why it matters
The demonstrated persuasive edge of AI raises concerns about influence concentration and the need for governance of persuasive AI tools, while also offering opportunities for under‑resourced actors to compete in advocacy and fundraising.