DFG-Project Accountable AI
Accountable Artificial Intelligence-based Systems: A multi-perspective analysis
Aim of the research project:
Developments in the field of artificial intelligence (AI) offer new innovative opportunities to contribute to the well-being and progress of individuals and society. However, due to a large number of incidents involving AI (e.g. discrimination through AI predictions), the accountability of AI is becoming increasingly important. Accountability generally means that actions carried out can be clearly assigned to a person. Applied to AI, accountable AI-based information systems (AAIS) refer to a socio-technical relationship structure of people who interact with AI technologies in order to perform certain tasks, whereby the actions within the interaction can be clearly attributed to a person. This is intended to ensure accountability in order to hold someone legally responsible in the event of a failure of the AI-based IS.
While calls for the development and embedding of mechanisms to create accountability in AI-based IS are growing louder, research on AAIS is still in its infancy. Based on the current state of research, there is conceptual ambiguity about AAIS due to the multitude of definitions of terms and differing opinions on what goals should be pursued by creating accountability. On the other hand, there is a lack of validated findings and explanatory models on how the use of accountability mechanisms affects the development, operation and use of AI. This results in three research questions that are to be answered as part of this project: (1) “Which facets of accountability are relevant for AAIS?”, (2) “What influence does accountability have on the perception of the development, operation and use of AI?” and (3) “How does accountability affect the behavior of AI users and architects?”.
In order to develop a holistic conceptualization of AAIS, the applicants plan to identify facets of accountability that are decisive for the effectiveness of accountability mechanisms within an AAIS. This will provide a conceptual and methodological toolkit for further accountability research. The second research question is to be answered by examining both positive and negative perceptions of accountability mechanisms by the actors. The investigations are based on the accountability facets and are theoretically underpinned by accountability theory. The insights gained make a theoretical contribution in the form of multi-perspective explanatory models for the effect of accountability. To answer the third research question, real-world research will be conducted to investigate how the behavior of AI architects and users changes when they can be held accountable through accountability mechanisms.
Applicant:
Contact person:
Sponsor:
