[论文解读] Responsible AI Governance: A Systematic Literature Review

Amna Batool, Didar Zowghi+1 more|arXiv (Cornell University)|Dec 18, 2023

Ethics and Social Impacts of AI被引用 9

一句话总结

tldr: 对61项研究的系统性文献综述，涵盖框架、模型、工具与政策，通过3W1H（Who-What-When-How）视角分析治理覆盖范围并识别差距。

ABSTRACT

As artificial intelligence transforms a wide range of sectors and drives innovation, it also introduces complex challenges concerning ethics, transparency, bias, and fairness. The imperative for integrating Responsible AI (RAI) principles within governance frameworks is paramount to mitigate these emerging risks. While there are many solutions for AI governance, significant questions remain about their effectiveness in practice. Addressing this knowledge gap, this paper aims to examine the existing literature on AI Governance. The focus of this study is to analyse the literature to answer key questions: WHO is accountable for AI systems' governance, WHAT elements are being governed, WHEN governance occurs within the AI development life cycle, and HOW it is executed through various mechanisms like frameworks, tools, standards, policies, or models. Employing a systematic literature review methodology, a rigorous search and selection process has been employed. This effort resulted in the identification of 61 relevant articles on the subject of AI Governance. Out of the 61 studies analysed, only 5 provided complete responses to all questions. The findings from this review aid research in formulating more holistic and comprehensive Responsible AI (RAI) governance frameworks. This study highlights important role of AI governance on various levels specially organisational in establishing effective and responsible AI practices. The findings of this study provides a foundational basis for future research and development of comprehensive governance models that align with RAI principles.

研究动机与目标

• Synthesize existing literature on AI governance to understand who controls AI governance, what is governed, when governance occurs in the AI lifecycle, and how governance is enacted.
• Categorize governance solutions by frameworks, models, tools, standards, policies, and guidelines across multiple levels (team, organization, industry, national, international).
• Assess limitations and challenges in current AI governance solutions and identify gaps for comprehensive Responsible AI governance (RAI).
• Provide a structured basis for future RAI governance frameworks aligned with ethical principles and diverse stakeholder involvement.

提出的方法

• Conduct a systematic literature review following Kitchenham guidelines.
• Use 3W1H (Who, What, When, How) to analyze 61 selected papers (empirical and non-empirical).
• Classify governance solutions into five levels of governance and group stakeholders by governance level.
• Perform data extraction on publication year, study type, and 3W1H coverage to identify completeness of governance solutions.

实验结果

研究问题

• RQ1 RQ1: What governance frameworks, models, tools, and policies for AI are offered in the literature?
• RQ2 RQ2: Which target AI application domains are considered in the existing AI governance approaches/solutions?
• RQ3 RQ3: What are the limitations and challenges of AI governance discussed in the literature?

主要发现

• 61 studies were analyzed, with 27 published in 2023 and a minority providing complete 3W1H answers (only 5 studies).
• Most governance solutions (19) address organization-level governance, while national and international levels have fewer, and none provide complete 3W1H coverage.
• Data and systems are the most frequently governed pillars; humans and processes receive comparatively less emphasis in the literature.
• There is a strong focus on how to govern (methods/tools) relative to who/when/what, indicating gaps in stakeholder roles and lifecycle-stage coverage.
• A variety of AI governance frameworks, models, tools, and policy guidelines exist, but ethical principles and stakeholder involvement at different lifecycle stages remain inconsistently addressed.
• Common application domains include healthcare and robotics, with many studies not specifying domains.