按年被提到
20262026 · 5 次2026
AI 安全、伦理与社会影响›大模型与人类行为·arXiv:2606.16475 · cs.CY

AI systems out-persuade expert humans

AI 系统的说服力超过专家级人类
Kobi Hackenburg, Caroline Wagner, Luke Hewitt, Ben M. Tappin, Ed Saunders, Hannah Rose Kirk, Helen Margetts, Christopher Summerfield · 2026

四项预注册实验:对话式 AI 的说服力胜过专家级人类

开始阅读 ↗PDFalphaXiv 讨论
被提及 · 5 人首次 2026-06-23 · 最近 2026-09-19
5不同的人
4活跃月份
5讨论串
4被引(OpenAlex)

论文用四项预注册实验(6,923 人、18,978 次对话),把 AI 与外行、说服锦标赛获胜者、职业募捐员和世界辩论冠军相比,发现 AI 的说服力始终更强,即使人类专家自选议题、提前调研并获得 1,000 英镑奖金。在为 Save the Children 募捐的真实场景中,AI 的效果接近职业募捐员的 3 倍;作者认为其优势主要来自能更快地提供更多信息。

HN 为什么提它

输入里没有评论摘录,只能看出 5 人在 4 个月内提到它,具体的讨论角度无法确认。

实证研究难度 2 / 5约 35 分钟有时效性
读之前最好知道
1随机对照实验与预注册
2统计显著性与效应量
3大语言模型对话能力基础
4说服与态度改变研究

HN 书评

没有评论直接链到它;下面是它被提交的讨论串里回复最多的评论(不计入统计)
讨论串评论marojejian · 2026-06

“> We found converging evidence that AI's advantage stemmed from rapidly deploying larger quantities of information This jibes with previous results in LLMs effectiveness at combatting disinformation: https://mitsloan.mit.edu/ideas-made-to-matter/mit-study-ai-chatbot-can-reduce-belief-conspiracy-theories >in a series of four preregistered experiments Lo…”

AI systems out-persuade expert humans原评论 ↗
讨论串评论kbt22 · 2026-06

“I disagree with the conclusion. AI systems out-persuaded expert humans in 2026. Should AI systems be prevalent in political debates, each of us will become experts at dealing with this, and it's not clear AI systems will out-persuade humans in the long run.”

AI systems out-persuade expert humans原评论 ↗
讨论串评论Gaishan · 2026-07

“"We therefore hypothesized that AI’s throughput advantage allows it to pack conversations with more fact-checkable claims than human persuaders can, and that this is what drives its persuasive edge. This “fact-density” account predicts that (i) constraining AI should selectively reduce persuadees’ perception of AI’s informational contribution to the conv…”

AI systems out-persuade expert humans原评论 ↗
被提交为 story · 5 次