Paper short abstract:

The paper presents reflections and interim results of ongoing interdisciplinary research on presumptions and foundations of human control of AI. The “AI-Cockpit” is a research project aimed at building an interface that grants humans oversight and control over systems of automated decision making.

Paper long abstract:

Within the framework of the project “AI-Cockpit”, a technology for an AI-system-interface will be developed that enables both human oversight and control over the automated decisions of various AI-systems. Our interdisciplinary research team is thus faced with the challenge to further conceptionalize the idea of “Human in Command of AI” and at the same time generate practical implementations of our premise. We follow a human-centered research approach using participatory methods. In addition, we will develop a generic “cockpit model” that can be used across individual cases and in the design of new AI-systems. Our approach is strictly aligned with the requirements of the EU-AI Act.

In three different application areas (Human Resources, Smart City, Care), first prototypes will be developed which provide insights across the individual areas and enable the development of free/libre open-source software. Our research network is composed of computer scientists, Industrial and organizational psychologists, STS scholars, cultural anthropologists and neuroscientists who in cooperation with our practice partners contribute to the design of a cockpit model.

For our project we have chosen a special variant of interdisciplinary work: Our synthesis-based-research project is accompanied by practitioners who make their own significant contribution to the research. In all fields of application, we go beyond general interdisciplinary cooperation and ask people in the respective fields about their needs, their experiences and assessments in terms of sustainable cooperation (vgl. Klausner/Niewöhner 2020, 163). This creates "epistemic partnerships" (Bieler et al. 2021, 525) which promise added value in the development of human-centered AI systems.