Scholarlink

<p>Over 148 responsible AI (RAI) frameworks have been catalogued, yet AI-driven discrimination continues to compound social harm, concentrate power, and undermine equity for marginalized communities. Existing frameworks are designed for academic, technical and legal specialists, treating discrimination as a static property to be assessed at a single point, with no shared language for academics, executives, developers, auditors, and policymakers who collectively determine AI outcomes across AI lifecycles. This paper introduces the <i>Discrimination DRIFT</i> framework, derived from a structured narrative review of empirical literature across five sectors. DRIFT models discriminatory outcomes as directional: positive DRIFT reduces bias and advances equity through deliberate stakeholder action; negative DRIFT compounds social harm and entrenches existing power imbalances when those actions are absent. The framework provides accessible, jargon-free language for cross-disciplinary communication across job roles and organisational levels, supported by an aide-mémoire for retention and deployment. DRIFT is an acronym for Diverse Teams, Representative Data, Independent Audit, Freedom to Challenge, and Transparency. Each component maps directly to documented mechanisms of discriminatory harm and to legislative requirements including the EU AI Act and GDPR. The framework brings together siloed RAI governance practices into a single practical resource, supporting actionable organizational change grounded in responsible AI principles. This applied lens offers relevant guidance for policymakers, practitioners, and researchers alike.</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

Discrimination DRIFT: A Practical Framework for Responsible AI

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月26日

查看全文
  • Sarah Wyer

摘要

Over 148 responsible AI (RAI) frameworks have been catalogued, yet AI-driven discrimination continues to compound social harm, concentrate power, and undermine equity for marginalized communities. Existing frameworks are designed for academic, technical and legal specialists, treating discrimination as a static property to be assessed at a single point, with no shared language for academics, executives, developers, auditors, and policymakers who collectively determine AI outcomes across AI lifecycles. This paper introduces the Discrimination DRIFT framework, derived from a structured narrative review of empirical literature across five sectors. DRIFT models discriminatory outcomes as directional: positive DRIFT reduces bias and advances equity through deliberate stakeholder action; negative DRIFT compounds social harm and entrenches existing power imbalances when those actions are absent. The framework provides accessible, jargon-free language for cross-disciplinary communication across job roles and organisational levels, supported by an aide-mémoire for retention and deployment. DRIFT is an acronym for Diverse Teams, Representative Data, Independent Audit, Freedom to Challenge, and Transparency. Each component maps directly to documented mechanisms of discriminatory harm and to legislative requirements including the EU AI Act and GDPR. The framework brings together siloed RAI governance practices into a single practical resource, supporting actionable organizational change grounded in responsible AI principles. This applied lens offers relevant guidance for policymakers, practitioners, and researchers alike.