Scholarlink

<p>This study systematically examines the inherent trade-offs between explainability and performance in high- stakes autonomous AI systems operating within critical domains such as healthcare, criminal justice, finance, and aviation. Employing a PRISMA protocol-led systematic literature review, 45 peer-reviewed articles were selected through rigorous inclusion and exclusion criteria. The findings reveal a persistent tension: while complex models, especially deep learning architectures, achieve superior predictive accuracy, they often lack interpretability, undermining trust, accountability, and regulatory compliance in sensitive decision-making environments. The study emphasizes that explainability is not a peripheral concern but a foundational requirement in applications with profound ethical, legal, and societal implications. It advocates for the development and adoption of hybrid and context-sensitive AI frameworks that balance predictive performance with interpretability. By addressing the governance, user trust, and technical dimensions of this trade-off, the research offers critical insights for both AI developers and policymakers working toward responsible, transparent, and human-aligned AI deployment.</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

The paradox of explainability vs. performance in high-stakes autonomous AI systems: a systematic review of trade-offs, regulatory gaps, and emerging solutions

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月13日

查看全文
  • Uday Nedunuri,
  • Abhijit Das Gupta,
  • Debashis Guha

摘要

This study systematically examines the inherent trade-offs between explainability and performance in high- stakes autonomous AI systems operating within critical domains such as healthcare, criminal justice, finance, and aviation. Employing a PRISMA protocol-led systematic literature review, 45 peer-reviewed articles were selected through rigorous inclusion and exclusion criteria. The findings reveal a persistent tension: while complex models, especially deep learning architectures, achieve superior predictive accuracy, they often lack interpretability, undermining trust, accountability, and regulatory compliance in sensitive decision-making environments. The study emphasizes that explainability is not a peripheral concern but a foundational requirement in applications with profound ethical, legal, and societal implications. It advocates for the development and adoption of hybrid and context-sensitive AI frameworks that balance predictive performance with interpretability. By addressing the governance, user trust, and technical dimensions of this trade-off, the research offers critical insights for both AI developers and policymakers working toward responsible, transparent, and human-aligned AI deployment.