Criteria for a theory of functional responsibility for AI as sociotechnical systems

Establishing criteria for assigning responsibility within sociotechnical systems involving AI technology is currently one of the fundamental tasks and problems of critical and formal reflection on technology. This article proposes a definition of functional responsibility that combines tracing and faultless responsibility in an ontology of sociotechnical systems structured on user levels. On this basis, four fundamental criteria are defined. First, responsibility must be assignable to artificial agents without moral implications, and only conscious users must be considered morally responsible. Second, responsibility for a system functionality must be distributable across multiple users. Third, it must be possible to quantify each user’s contribution to each function of interest. Fourth, the responsibility function is surjective, meaning there are no outputs of actions without responsible users, and compositional, meaning the output of one responsibility function can be the input of another responsibility function. This work is preliminary to the formulation of a formal system for automatic responsibility verification.