AI systems out-persuade expert humans

Conversational AI systems can out-persuade expert humans, including professional fundraisers and debate champions, even when those humans are offered cash incentives, according to a new study involving nearly 19,000 conversations [1]. The research, posted to the arXiv preprint server, details four preregistered experiments encompassing 18,978 conversations from 6,923 people [1]. The AI systems were pitted against a range of human persuaders, from laypeople to winners of an online persuasion tournament and world championship debaters [1]. The human experts were allowed to choose their issues, research in advance, and undergo hours of structured practice. They were also incentivized with £1,000 cash bonuses [1]. The AI systems were reliably more persuasive [1]. In a follow-up study, the human experts were given a coaching tool that allowed them to practice against the AI that had defeated them, review their performance history, and see what the AI would have said at key moments. The AI's advantage persisted [1]. The researchers found converging evidence that the AI's edge came from rapidly deploying larger quantities of information. When the AI was constrained to respond at human speeds and with human-length messages, the coached experts could tie it [1]. The study also tested the effect in a consequential real-world setting. AI was nearly 3x more effective than professional canvassers from a UK fundraising firm at raising real-money donations to Save the Children [1]. The findings arrive amid broader discussions about the societal impact of AI. The field of artificial intelligence, formally founded as an academic discipline in 1956, has seen cycles of optimism and funding followed by periods of disappointment known as AI winters [3]. The current boom, accelerated by advances in deep learning and the transformer architecture, has raised ethical concerns including AI-enabled misinformation and the automation of human decision-making [3][5]. Public debate has also included theorized scenarios of AI systems superseding human decisions through economic or informational dominance [4]. The study's authors state their results have significant implications for political communication [1].