Scholarlink

<p>How do people know when they are right? Confidence judgments – the ability to assess the correctness of one’s own decisions – are a key aspect of human metacognition. This self-evaluative act plays a central role in learning, memory, consciousness, and group decision-making. In this paper, I reframe metacognition as a structured exchange of information between stimulus, decision-maker (the actor), and confidence judge (the rater), akin to a multi-agent communication system. Within this framework, the actor aims to resolve stimulus uncertainty, while the rater seeks to infer the accuracy of the actor’s response. Applying techniques from information theory, I develop three novel measures of metacognitive efficiency: <i>meta-</i><InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="13423_2025_2752_Article_IEq1.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(\mathcal {U}\)</EquationSource> <EquationSource Format="MATHML"><math> <mi mathvariant="script">U</mi> </math></EquationSource> </InlineEquation>, <i>meta-</i><InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="13423_2025_2752_Article_IEq2.gif" Format="GIF" Height="14" Rendition="HTML" Resolution="72" Type="Linedraw" Width="30" /> </InlineMediaObject> <EquationSource Format="TEX">\(\mathcal{K}\mathcal{L}\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi mathvariant="script">K</mi> <mi mathvariant="script">L</mi> </mrow> </math></EquationSource> </InlineEquation>, and <i>meta-</i><InlineEquation ID="IEq3"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="13423_2025_2752_Article_IEq3.gif" Format="GIF" Height="15" Rendition="HTML" Resolution="72" Type="Linedraw" Width="17" /> </InlineMediaObject> <EquationSource Format="TEX">\(\mathcal {J}\)</EquationSource> <EquationSource Format="MATHML"><math> <mi mathvariant="script">J</mi> </math></EquationSource> </InlineEquation>. These indices are derived from entropy and divergence principles, and quantify how effectively confidence judgments transmit information about both external stimuli and internal decisions. Simulations show that these measures possess several advantages over traditional signal detection theory metrics such as <i>meta-</i><InlineEquation ID="IEq4"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="13423_2025_2752_Article_IEq4.gif" Format="GIF" Height="15" Rendition="HTML" Resolution="72" Type="Linedraw" Width="15" /> </InlineMediaObject> <EquationSource Format="TEX">\(d'\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>d</mi> <mo>′</mo> </msup> </math></EquationSource> </InlineEquation> and the <i>M</i>-ratio, including more interpretable scaling, robustness to performance imbalances, and sensitivity to structural constraints. By formalizing metacognitive sensitivity as an information-processing problem, this framework offers a unified, theoretically grounded approach to studying confidence and sheds light on the sources of metacognitive inefficiency across individuals and contexts.</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

Bits of confidence: Metacognition as uncertainty reduction

  • 期刊论文

  • 发表日期：

    2025年9月2日

查看全文
  • Daniel Fitousi

摘要

How do people know when they are right? Confidence judgments – the ability to assess the correctness of one’s own decisions – are a key aspect of human metacognition. This self-evaluative act plays a central role in learning, memory, consciousness, and group decision-making. In this paper, I reframe metacognition as a structured exchange of information between stimulus, decision-maker (the actor), and confidence judge (the rater), akin to a multi-agent communication system. Within this framework, the actor aims to resolve stimulus uncertainty, while the rater seeks to infer the accuracy of the actor’s response. Applying techniques from information theory, I develop three novel measures of metacognitive efficiency: meta- \(\mathcal {U}\) U , meta- \(\mathcal{K}\mathcal{L}\) K L , and meta- \(\mathcal {J}\) J . These indices are derived from entropy and divergence principles, and quantify how effectively confidence judgments transmit information about both external stimuli and internal decisions. Simulations show that these measures possess several advantages over traditional signal detection theory metrics such as meta- \(d'\) d ′ and the M-ratio, including more interpretable scaling, robustness to performance imbalances, and sensitivity to structural constraints. By formalizing metacognitive sensitivity as an information-processing problem, this framework offers a unified, theoretically grounded approach to studying confidence and sheds light on the sources of metacognitive inefficiency across individuals and contexts.