AI under great uncertainty: implications and decision strategies for public policy

Decisions where there is not enough information for a well-informed decision due to unidentified consequences, options, or undetermined demarcation of the decision problem are called decisions under great uncertainty. This paper argues that public policy decisions on how and if to implement decision-making processes based on machine learning and AI for public use are such decisions. Decisions on public policy on AI are uncertain due to three features specific to the current landscape of AI, namely (i) the vagueness of the definition of AI, (ii) uncertain outcomes of AI implementations and (iii) pacing problems. Given that many potential applications of AI in the public sector concern functions central to the public sphere, decisions on the implementation of such applications are particularly sensitive. Therefore, it is suggested that public policy-makers and decision-makers in the public sector can adopt strategies from the argumentative approach in decision theory to mitigate the established great uncertainty. In particular, the notions of framing and temporal strategies are considered.