Human Control and Discretion in AI-driven Decision-making in Government

Abstract

Traditionally public decision-makers have been given discretion in many of the decisions they have to make in how to comply with legislation and policies. In this way, the context and specific circumstances can be taken into account when making decisions. This enables more acceptable solutions, but at the same time, discretion might result in treating individuals differently. With the advance of AI-based decisions, the role of the decision-makers is changing. The automation might result in fully automated decisions, humans-in-the-loop or AI might only be used as recommender systems in which humans have the discretion to deviate from the suggested decision. The predictability of and the accountability of the decisions might vary in these circumstances, although humans always remain accountable. Hence, there is a need for human-control and the decision-makers should be given sufficient authority to control the system and deal with undesired outcomes. In this direction this paper analyzes the degree of discretion and human control needed in AI-driven decision-making in government. Our analysis is based on the legal requirements set/posed to the administration, by the extensive legal frameworks that have been created for its operation, concerning the rule of law, the fairness-non-discrimination, the justifiability and accountability, and the certainty/predictability.