Putting a Face on Algorithms: Personas for Modeling Artificial Intelligence

Abstract

We propose a new type of personas, artificial intelligence (AI) personas, as a tool for designing systems consisting of both human and AI agents. Personas are commonly used in design practices for modelling users. We argue that the personification of AI agents can help multidisciplinary teams in understanding and designing systems that include AI agents. We propose a process for creating AI personas and the properties they should include, and report on our first experience using them. The case we selected for our exploration of AI personas was the design of a highly automated decision support tool for air traffic control. Our first results indicate that AI personas helped designers to empathise with algorithms and enabled better communication within a team of designers and AI and domain experts. We call for a research agenda on AI personas and discussions on potential benefits and pitfalls of this approach.