Facilitating Responsible Use of Artificial Intelligence in the Public Sector Through Guidelines for Human- AI Interaction

Abstract

When developing Artificial Intelligence (AI) systems and implementing them in public services, guidelines for human-AI interaction can be instrumental in ensuring responsible use of these systems and AI intelligibility and responsibility, protecting the citizens’ rights and interests. Although there are as of yet no such guidelines established that are directed at the public sector, tech companies like Microsoft have established guidelines for the commercial sector. To investigate how commercial guidelines may help guide the development of public services that utilize AI, we have explored how Microsoft Human-AI Interaction Guidelines can be amended and used to guide the development of a sick leave duration prediction service. By developing a web application prototype for the specific case and conducting usability testing, we found that the guidelines needed amendments to better ensure responsible use of AI in a public service. Three considerations were discovered to be important when amending them: (i) the way users interact with the AI system, (ii) the diversity of citizens making use of the service, and (iii) the public service context.