Scholarlink

<p>The open literature continues to voice concerns with regard to the fundamental challenge of opaqueness in machine learning (ML). This dilemma emerges from the tension between harnessing the predictivity of algorithms and maintaining algorithmic oversight. From this lens, this paper sheds light on key philosophical aspects of the problem of limited interpretability, highlights the difficulties in ensuring reliable deployment, and presents a framework to overcome the aforementioned challenge. The proposed framework integrates three elemental standpoints: <i>Dataism</i>, reflecting the unwavering reliance on data in ML for decision-making; <i>Skepticism</i>, ensuring vigilant scrutiny of model outcomes and bias; and <i>Intuition</i>, underlining the experiential wisdom embedded in domain expertise. Through mapping these standpoints onto the proposed DSI framework, we show how each standpoint offers distinct and converging benefits. This paper showcases the proposed framework through a theoretical analysis that focuses on ML deployments to demonstrate how a balanced consideration of the three standpoints can help alleviate concerns surrounding interpretability and contextual understanding. Finally, this study also provides a philosophical and technical critique of the proposed framework and shares strategies for melding data-driven decision making with human oversight to serve as a blueprint for transparent ML practices – especially in engineering domains.</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

Dataism, skepticism, and intuition for interpretable machine learning

  • 期刊论文

  • 发表日期：

    2025年8月28日

查看全文
  • M. Z. Naser

摘要

The open literature continues to voice concerns with regard to the fundamental challenge of opaqueness in machine learning (ML). This dilemma emerges from the tension between harnessing the predictivity of algorithms and maintaining algorithmic oversight. From this lens, this paper sheds light on key philosophical aspects of the problem of limited interpretability, highlights the difficulties in ensuring reliable deployment, and presents a framework to overcome the aforementioned challenge. The proposed framework integrates three elemental standpoints: Dataism, reflecting the unwavering reliance on data in ML for decision-making; Skepticism, ensuring vigilant scrutiny of model outcomes and bias; and Intuition, underlining the experiential wisdom embedded in domain expertise. Through mapping these standpoints onto the proposed DSI framework, we show how each standpoint offers distinct and converging benefits. This paper showcases the proposed framework through a theoretical analysis that focuses on ML deployments to demonstrate how a balanced consideration of the three standpoints can help alleviate concerns surrounding interpretability and contextual understanding. Finally, this study also provides a philosophical and technical critique of the proposed framework and shares strategies for melding data-driven decision making with human oversight to serve as a blueprint for transparent ML practices – especially in engineering domains.