Accountability for AI-controlled decision making in public administration

The researchers are examining the issue of accountability from a philosophical and legal perspective in relation to automated or AI-controlled decision making in public administration.

Even now many decision-making processes in public administration are based on simple algorithms. Tax bills, entitlement to social welfare payments and qualification for citizenship are already among the decisions taken by algorithms. 

Strides in AI technology have led to the development of systems based on increasingly advanced algorithms. These systems have limited transparency, can process large quantities of data, are flexible, and can to some extent design themselves.

If automated or AI-controlled decision-making systems become more common, they might conceivably replace a large proportion of human decision making in public administration. This prospect raises fundamental philosophical issues about accountability: who or what should be held accountable for wrong decisions and why?

The researchers in this interdisciplinary project distinguish three scenarios: 

  1. Artificial systems based on simple algorithms. Although humans are generally held accountable in these situations, the legal position in Sweden remains unclear. 
  2. Artificial systems as self-learning and self-developing algorithms. In this case entirely novel legal and philosophical questions must be addressed. For instance, decisions made by AI systems at this level cannot necessarily be understood later on by humans. Yet someone or something must be accountable.
  3. A hypothetical scenario in which AI systems exhibit behavior that cannot be distinguished from human behavior, or even possess something that cannot be equated with human consciousness. Would it be possible to hold a hypothetical entity of this kind accountable for its decisions from a legal and philosophical perspective?

The researchers aim to examine accountability from a philosophical and legal viewpoint in all three cases. They will ascertain whether there is a need to find new rules governing accountability for autonomous systems, or modify existing ones. To do so, they will analyze both how accountability works under Swedish laws, and how moral accountability and moral agency can be understood by means of philosophical reasoning. Their analysis will be based on accountability under Swedish law, and a philosophical analysis of moral responsibility and moral agency. 

The results of the analysis will result in specific proposals for modified categories of accountability and new forms of accountability that can be applied both generally and more specifically in a Swedish context under the law. The philosophical and legal analysis is also expected to yield a deeper understanding of the ethical, legal and societal implications of introducing AI systems in public administration.

Project:
The Artificial Public Servant

Principal investigator:
Sandra Friberg

Co-investigators:
Johan Eddebo
Lars Karlander
Oliver Li

Institution:
Uppsala University

Grant:
SEK 6 million