Responsible AI

Abstract

The principles-first approach adopted by most regulators in the field of AI has created an opportunity for collaboration between regulators and industry - with that dialogue allowing regulators to refine and converge on basic principles of responsible AI, as well as gain insight on whether new and specific regulation is needed to govern the use of AI. The benefit of such collaborations is the emergence of learnings and best practices, which when shared can strengthen regulators and industry’s understanding of AI technology and promote the responsible use of AI. From Microsoft’s own responsible AI journey, and through participation in some collaborative initiatives, an overarching learning was the importance of context in evaluating the regulatory implications of an AI application. Not all responsible AI principles will be equally relevant in each context, and regulatory issues are best assessed based on the specific use of AI being deployed, and the jurisdiction and industry in which the AI application is being used. Another key learning is that a materiality-based approach can help focus efforts on more sensitive AI use cases, compared with approaches that treat all uses of AI the same and put them through the same internal process. It was also observed that, in the financial services industry in particular, existing internal risk management and governance frameworks can provide a useful starting point when assessing the implications of using AI. Another learning is the complementary roles of financial institutions and technology partners in promoting transparency and accountability. Additionally, there is great value in bringing a diverse set of perspectives into discussions on responsible AI regulatory principles. A final learning is that it is helpful to ground the dialogue on responsible AI by focusing on actual use cases and examples.