Scholarlink

<p>Influential models conceive human thinking as an interplay between intuition and deliberation. Yet, it’s unclear how people actually perceive these types of reasoning. Across 13 studies (<i>n</i> = 239, 241, 240, 240, 241, 240, 184, 482, 479, 240 and 240 for Studies 1, 2, 3, 4, 5, 6, 7, 10, 11, 12 and 13, respectively), we examined whether humans favor intuition or deliberation and if this replicates in LLMs. Participants rated individuals’ reasoning quality in short vignettes that varied by reasoning type (fast-intuitive vs. slow-deliberative) and past accuracy (high, low, unspecified). Consistently, participants rated deliberative reasoning as superior to intuition, even when accounting for accuracy. Deliberative thinkers were seen as smarter and more trustworthy—a preference that held under time pressure and cognitive load, suggesting it arises intuitively. Studies with LLMs (ChatGPT 3.5 and 4) replicated the human preference pattern, indicating that AI language models capture human folk beliefs about reasoning. These findings suggest humans intuitively link deliberation with reliability and have implications for public trust in human and AI recommendations.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Humans and LLMs rate deliberation as superior to intuition on complex reasoning tasks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月29日

查看全文
  • Wim De Neys,
  • Matthieu Raoelison

摘要

Influential models conceive human thinking as an interplay between intuition and deliberation. Yet, it’s unclear how people actually perceive these types of reasoning. Across 13 studies (n = 239, 241, 240, 240, 241, 240, 184, 482, 479, 240 and 240 for Studies 1, 2, 3, 4, 5, 6, 7, 10, 11, 12 and 13, respectively), we examined whether humans favor intuition or deliberation and if this replicates in LLMs. Participants rated individuals’ reasoning quality in short vignettes that varied by reasoning type (fast-intuitive vs. slow-deliberative) and past accuracy (high, low, unspecified). Consistently, participants rated deliberative reasoning as superior to intuition, even when accounting for accuracy. Deliberative thinkers were seen as smarter and more trustworthy—a preference that held under time pressure and cognitive load, suggesting it arises intuitively. Studies with LLMs (ChatGPT 3.5 and 4) replicated the human preference pattern, indicating that AI language models capture human folk beliefs about reasoning. These findings suggest humans intuitively link deliberation with reliability and have implications for public trust in human and AI recommendations.