Reproduce “AI systems out-persuade expert humans” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kobi Hackenburg, Caroline Wagner, Luke Hewitt, Ben M. Tappin, Ed Saunders, Hannah Rose Kirk, Helen Margetts, Christopher Summerfield
Frontier AI systems consistently demonstrated greater conversational persuasive power than highly skilled human experts, including professional debaters and canvassers, across attitudinal and real-money donation tasks. This advantage is largely attributed to AI's superior information throughput, enabling it to out-persuade humans by up to 10.8 percentage points in real-money donations.
Key insights
- AI's persuasive advantage over human experts is primarily driven by its superior information throughput, allowing it to rapidly deliver a higher volume of fact-checkable claims.
- Targeted coaching and practice, including exposure to the AI that had defeated them, did not enable elite human debaters to significantly close the persuasion gap.
- The findings suggest a societal shift towards an era of sophisticated AI-driven advocacy, with implications for the concentration of influence among those with AI access and the potential for both well-informed citizens or widespread disinformation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.