AccountabilityEthicsTransparencycs.CY
Principles to Practices for Responsible AI: Closing the Gap
Daniel Schiff, Bogdana Rakova, Aladdin Ayesh, Anat Fanti, Michael Lennon
June 8, 2020
arXiv.org
130 citations
11 influential
Key Takeaways
Key Takeaways: Principles to Practices for Responsible AI
-
The "principles-to-practices gap" is a real problem: While many companies have adopted high-level AI ethics principles, the paper identifies a significant disconnect between these stated principles and their actual implementation in day-to-day organizational practices.
-
Multiple barriers prevent implementation: The gap stems from five distinct challenges, including disciplinary divides (e.g., between ethicists and engineers) and an overwhelming number of available tools and frameworks, making it unclear which approaches to prioritize.
-
Impact assessments offer a practical solution: The authors recommend using impact assessment frameworks as a concrete way to translate principles into action. These frameworks should be broad (covering multiple stakeholder concerns), operationalizable (actually doable), flexible, iterative, guided by clear processes, and participatory (involving diverse voices).
-
Real-world application is possible: A case study on AI use in forest ecosystem restoration demonstrates how these impact assessment principles can work in practice, showing that responsible AI implementation is achievable beyond theoretical discussions.
-
Closing the gap requires organizational commitment: Moving from principles to practices isn't just a technical problem—it requires companies to fundamentally change how they develop and deploy AI systems, with structured processes and genuine stakeholder engagement.