Scholarlink

<p>Algorithmic bias in AI systems is generally considered a technical problem. This paper argues that cognitive theories of human cognitive bias are a useful tool for understanding and regulating AI system fairness, without positing that AI systems are literally cognitive systems. This paper draws on cognitive science and economics to outline how human cognitive heuristics and biases under bounded rationality can inform understanding of: functionally similar error patterns in AI systems; biased ways that humans understand or act on AI system output; and human economic or institutional interests that may reinforce particular biased socio-technical systems. This paper uses cognitive bias theory, including theories of heuristics and biases developed by Kahneman &amp; Tversky [<CitationRef CitationID="CR22">18</CitationRef>], to outline a system for understanding AI system “bias-like” failures. It argues that this system should be based on well-studied human cognitive decision phenomena, but that it should be distinguished by a discussion of how human cognitive systems are different from AI systems. This paper argues that AI systems should be designed to be “bias aware” but that this awareness should be used to strategically target particular patterns of unfairness. This paper also discusses some of the implications for AI system design that this perspective may suggest. It further discusses some of the limitations of this cognitive perspective.</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

Human bias, machine bias: a cognitive lens on fair AI

  • 期刊论文

  • 发表日期：

    2026年5月6日

查看全文
  • Nyida Gyal

摘要

Algorithmic bias in AI systems is generally considered a technical problem. This paper argues that cognitive theories of human cognitive bias are a useful tool for understanding and regulating AI system fairness, without positing that AI systems are literally cognitive systems. This paper draws on cognitive science and economics to outline how human cognitive heuristics and biases under bounded rationality can inform understanding of: functionally similar error patterns in AI systems; biased ways that humans understand or act on AI system output; and human economic or institutional interests that may reinforce particular biased socio-technical systems. This paper uses cognitive bias theory, including theories of heuristics and biases developed by Kahneman & Tversky [18], to outline a system for understanding AI system “bias-like” failures. It argues that this system should be based on well-studied human cognitive decision phenomena, but that it should be distinguished by a discussion of how human cognitive systems are different from AI systems. This paper argues that AI systems should be designed to be “bias aware” but that this awareness should be used to strategically target particular patterns of unfairness. This paper also discusses some of the implications for AI system design that this perspective may suggest. It further discusses some of the limitations of this cognitive perspective.