Interactive Human Centered Artificial Intelligence: A Definition and Research Challenges

Abstract

Artificial Intelligence (AI) has become the buzzword of the last decade. Advances so far have been largely technical with a focus on machine learning (ML). Only recently have we begun seeing a shift towards focusing on the human aspects of artificial intelligence, centered on the narrow view of making AI interactive and explainable. In this paper I suggest a definition for "Interactive Human Centered Artificial Intelligence and outline the required properties. Staying in control is essential for humans to feel safe and have self-determination. Hence, we need to find ways for humans to understand AI based systems and means to allow human control and oversight. In our work, we argue that levels of abstractions and granularity of control are a general solution to this. Furthermore, it is essential that we make explicit why we want AI and what are the goals of AI research and development. We need to state the properties that we expect of future intelligent systems and who will benefit from a system or service. For me, AI and ML are very much comparable to raw materials (like stone, iron, or bronze). Historical periods are named after these materials as they fundamentally changed what humans can build and what tools humans can engineer. Hence, I argue that in the AI age we need to shift the focus from the material (e.g. the AI algorithms, as there will be plenty of material) towards the tools and infrastructures that are enabled which are beneficial to humans. It is apparent that AI will allow the automation of mental routine tasks and that it will extend our ability to perceive the world and foresee events. For me, the central question is how to create these tools for amplifying the human mind without compromising human values.