Scholarlink

<p>Machine learning (ML) models are commonly portrayed as capturing stable patterns in data, yet they operate in inherently dynamic and evolving contexts. Drawing upon Jacques Derrida’s philosophical concepts of <i>différance</i>, <i>trace</i>, iterability and the archive, I critically examine the phenomenon of concept drift (or data drift) in machine learning, illuminating how models perpetually defer meaning and embody context-dependent interpretations. Derridean philosophy reveals fundamental limitations in prevalent artificial intelligence (AI) narratives– such as the presumed universality of scalable data, convergence toward fixed solutions and context-free, data-centric epistemologies– highlighting instead the inherent contingency and provisionality of computational knowledge. By analysing concrete examples from spam detection, facial recognition bias and predictive policing, I demonstrate explicitly how ML systems, as Derrida would have predicted, are interpretive processes, always relational, provisional and contextually dependent, carrying traces of absent contexts. Practically, embracing Derridean insights explicitly reshapes ethical AI practices, demanding continuous interpretive vigilance, transparent uncertainty communication, iterative and adaptive retraining infrastructures, preservation of contextual diversity and context-sensitive evaluation metrics. This philosophical reframing explicitly transforms AI from a static technological product into an ongoing interpretive conversation: ethically accountable, contextually responsive and perpetually open to reinterpretation.</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

Différance and Data Drift: The Trace of Change in Machine Learning

  • 期刊论文

  • 发表日期：

    2025年7月9日

查看全文
  • Nofie Iman

摘要

Machine learning (ML) models are commonly portrayed as capturing stable patterns in data, yet they operate in inherently dynamic and evolving contexts. Drawing upon Jacques Derrida’s philosophical concepts of différance, trace, iterability and the archive, I critically examine the phenomenon of concept drift (or data drift) in machine learning, illuminating how models perpetually defer meaning and embody context-dependent interpretations. Derridean philosophy reveals fundamental limitations in prevalent artificial intelligence (AI) narratives– such as the presumed universality of scalable data, convergence toward fixed solutions and context-free, data-centric epistemologies– highlighting instead the inherent contingency and provisionality of computational knowledge. By analysing concrete examples from spam detection, facial recognition bias and predictive policing, I demonstrate explicitly how ML systems, as Derrida would have predicted, are interpretive processes, always relational, provisional and contextually dependent, carrying traces of absent contexts. Practically, embracing Derridean insights explicitly reshapes ethical AI practices, demanding continuous interpretive vigilance, transparent uncertainty communication, iterative and adaptive retraining infrastructures, preservation of contextual diversity and context-sensitive evaluation metrics. This philosophical reframing explicitly transforms AI from a static technological product into an ongoing interpretive conversation: ethically accountable, contextually responsive and perpetually open to reinterpretation.