Scholarlink

<p>The opacity of AI systems poses a fundamental epistemic challenge: how can we justifiably trust systems whose decision processes are inscrutable? Rather than relying on inherently limited explainability, I propose an institutional epistemology of warrant. Drawing on speech act theory and social epistemology, I argue that institutional sources that constitute (rather than merely describe) reality can ground epistemic warrant without algorithmic transparency. The framework addresses both retrieval scenarios (existing constitutive sources are cited and verified) and generative scenarios (AI outputs acquire force only after institutional validation). It shifts attention from internal mechanisms to verifiable linkages with authoritative institutional frameworks. Despite important limitations (risks of fabricated constitutive sources, vulnerability to institutional decline, and ethical concerns about algorithmically originated normative content) the account offers a philosophically grounded alternative where comprehensibility is unattainable. The framework is intended for domains where constitutive authority enjoys recognized legitimacy, presupposing rather than creating institutional trust. The framework presupposes existing institutional trust; yet by making institutional warrant more visible and auditable, it also has the potential to reinforce that trust <i>ex post</i>.</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

Constitutive knowledge sources: an institutional approach to epistemic trust in opaque AI systems

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月11日

查看全文
  • Lior Gazit

摘要

The opacity of AI systems poses a fundamental epistemic challenge: how can we justifiably trust systems whose decision processes are inscrutable? Rather than relying on inherently limited explainability, I propose an institutional epistemology of warrant. Drawing on speech act theory and social epistemology, I argue that institutional sources that constitute (rather than merely describe) reality can ground epistemic warrant without algorithmic transparency. The framework addresses both retrieval scenarios (existing constitutive sources are cited and verified) and generative scenarios (AI outputs acquire force only after institutional validation). It shifts attention from internal mechanisms to verifiable linkages with authoritative institutional frameworks. Despite important limitations (risks of fabricated constitutive sources, vulnerability to institutional decline, and ethical concerns about algorithmically originated normative content) the account offers a philosophically grounded alternative where comprehensibility is unattainable. The framework is intended for domains where constitutive authority enjoys recognized legitimacy, presupposing rather than creating institutional trust. The framework presupposes existing institutional trust; yet by making institutional warrant more visible and auditable, it also has the potential to reinforce that trust ex post.