Simon Kiorgaard

The use of principles for ethical analysis and evaluation in AI ethics is widespread. Although common, this tendency is not without its critics. The most frequent criticisms point to alleged impracticalities stemming from the principles’ abstractness, and generality in scope. The principal aim of this paper is to respond to these criticisms, with a view to then setting the current principle-based approach upon firmer footing. To do this, I examine the method of principlism in bioethics to which the principle-based approach in AI ethics has often been compared. As the bioethics literature is comparatively older than the AI ethics literature, my expectation here is that the former body of work will have insights that I can draw on to address the aforementioned criticisms. Based on this investigation, I recommend that the current ad hoc principle-based approach in AI ethics be supported by and worked into a more rigorous theoretical framework capable of systematically closing the gap between abstract ethical reasoning and practical AI governance. Key to this framework, and inspired by the literature of bioethics, is the distinction between general principles and particular rules, and the related methods of rule specification and reflective equilibrium which mediate between them.