[Paper Review] Principles to Practices for Responsible AI: Closing the Gap

Daniel Schiff, Bogdana Rakova+3 more|arXiv (Cornell University)|Jun 8, 2020

Ethics and Social Impacts of AI28 references62 citations

TL;DR

The paper analyzes why high-level responsible AI principles fail to translate into practice and proposes impact assessments, notably IEEE 7010, as a framework to close the gap, illustrated with a forest restoration case study.

ABSTRACT

Companies have considered adoption of various high-level artificial intelligence (AI) principles for responsible AI, but there is less clarity on how to implement these principles as organizational practices. This paper reviews the principles-to-practices gap. We outline five explanations for this gap ranging from a disciplinary divide to an overabundance of tools. In turn, we argue that an impact assessment framework which is broad, operationalizable, flexible, iterative, guided, and participatory is a promising approach to close the principles-to-practices gap. Finally, to help practitioners with applying these recommendations, we review a case study of AI's use in forest ecosystem restoration, demonstrating how an impact assessment framework can translate into effective and responsible AI practices.

Motivation & Objective

• Explain why responsible AI principles often fail to translate into organizational practices.
• Identify key explanations for the principles-to-practices gap including complexity, accountability, and disciplinary divides.
• Propose an effective framework for responsible AI centered on impact assessment.
• Argue that IEEE 7010's well-being impact assessment is a practical exemplar for closing the gap.
• Demonstrate applicability through a case study on AI for forest ecosystem restoration to illustrate actionable practices.

Proposed method

• Review existing responsible AI principles and the gap to practical adoption.
• Outline five explanations for the gap: complexity, accountability distribution, disciplinary divide, tool abundance, and labor division.
• Propose criteria for an effective framework (broad, operationalizable, flexible, iterative, guided, participatory).
• Advocate impact assessment, especially IEEE 7010, as the exemplar framework.
• Apply the framework in a case study on AI in forest ecosystem restoration to show practical translation.

Experimental results

Research questions

• RQ1 What factors explain why responsible AI principles are not translated into concrete practices within organizations?
• RQ2 How can an overarching framework address the principles-to-practices gap in AI development and deployment?
• RQ3 Can impact assessments, particularly IEEE 7010, operationalize responsible AI principles across lifecycle stages?
• RQ4 How does a case study in ecosystem restoration illustrate the use of a well-being impact assessment to guide responsible AI?
• RQ5 What guidance is needed to adapt impact assessments across sectors and use cases?

Key findings

• There are five main explanations for the gap: AI’s broad and complex impacts, muddled accountability, disciplinary divides, an abundance of tools, and organizational labor division.
• A broad, operationalizable, flexible, iterative, guided, and participatory impact-assessment framework can help translate principles into practice.
• IEEE 7010’s well-being impact assessment provides a practical exemplar by mapping domains to measurable indicators and stakeholder engagement.
• Impact assessments are adaptable to diverse use cases and can be iteratively updated as systems and contexts change.
• Guidance beyond a single tool is needed, favoring sector- and context-specific adaptations within a participatory framework.