Scholarlink

This chapter examines how artificial intelligence systems in neuroscience, despite their transformative potential, systematically perpetuate and amplify existing societal inequities rather than addressing them. This chapter documents a comprehensive “bias propagation pipeline” through which inequities enter, amplify, and perpetuate throughout AI development. This pipeline begins with historically biased research practices that favor Western, educated populations and culminates in deployed systems that demonstrate differential performance across demographic groups. Through systematic analysis of major neuroimaging datasets including the Human Connectome Project, ADNI, UK Biobank, and others, this chapter reveals profound demographic imbalances. Racial minorities, diverse geographic populations, and neurodivergent individuals remain severely underrepresented despite being the populations these technologies are designed to serve. These dataset biases manifest as real-world disparities in diagnostic accuracy, treatment recommendations, and accessibility. Documented performance gaps reach 40-fold differences in error rates across demographic groups in healthcare AI applications. This chapter outlines comprehensive technical approaches to bias mitigation, including improved data collection strategies, algorithmic fairness constraints, stratified evaluation frameworks, and participatory design methodologies. However, it argues that technical solutions alone are insufficient without a fundamental reconceptualization of neurological “normality” and a shift from fairness-oriented to justice-oriented approaches that embrace neurological diversity as natural human variation. This chapter concludes with a transformative vision for inclusive neurological AI that requires unprecedented collaboration across disciplines and communities, centering marginalized voices in development processes and ensuring global equity in technology access and benefits. The central argument echoes Elie Wiesel’s observation that neutrality helps oppressors rather than victims, demonstrating that passive approaches to AI development inevitably perpetuate discrimination while active commitment to equity and justice represents the only path toward neurological AI systems that truly serve all of humanity’s diverse neurological landscape.

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

Bias and Justice in Neurological AI

  • 章节

  • 发表日期：

    2025年

查看全文
  • Thorsten Rudroff

摘要

This chapter examines how artificial intelligence systems in neuroscience, despite their transformative potential, systematically perpetuate and amplify existing societal inequities rather than addressing them. This chapter documents a comprehensive “bias propagation pipeline” through which inequities enter, amplify, and perpetuate throughout AI development. This pipeline begins with historically biased research practices that favor Western, educated populations and culminates in deployed systems that demonstrate differential performance across demographic groups. Through systematic analysis of major neuroimaging datasets including the Human Connectome Project, ADNI, UK Biobank, and others, this chapter reveals profound demographic imbalances. Racial minorities, diverse geographic populations, and neurodivergent individuals remain severely underrepresented despite being the populations these technologies are designed to serve. These dataset biases manifest as real-world disparities in diagnostic accuracy, treatment recommendations, and accessibility. Documented performance gaps reach 40-fold differences in error rates across demographic groups in healthcare AI applications. This chapter outlines comprehensive technical approaches to bias mitigation, including improved data collection strategies, algorithmic fairness constraints, stratified evaluation frameworks, and participatory design methodologies. However, it argues that technical solutions alone are insufficient without a fundamental reconceptualization of neurological “normality” and a shift from fairness-oriented to justice-oriented approaches that embrace neurological diversity as natural human variation. This chapter concludes with a transformative vision for inclusive neurological AI that requires unprecedented collaboration across disciplines and communities, centering marginalized voices in development processes and ensuring global equity in technology access and benefits. The central argument echoes Elie Wiesel’s observation that neutrality helps oppressors rather than victims, demonstrating that passive approaches to AI development inevitably perpetuate discrimination while active commitment to equity and justice represents the only path toward neurological AI systems that truly serve all of humanity’s diverse neurological landscape.