Towards Continuous Explainability of Complex AI Systems: Requirements and Challenges

Preprints.org

Abstract

Increasing regulatory influences on artificial intelligence (AI) impose new requirements on the trustworthiness of complex AI systems, i.e., compound software systems using at least one AI method. When embedded in value chains and other regulatory scopes, it must be ensured that AI activities and AI-generated results remain explainable throughout the entire process. This work is motivated by the domain of AI-based requirements engineering (RE) in the automotive industry, where the overall objective is to meet homologation requirements with AI-generated contents influencing system development. In support thereof, this paper presents two contributions: an orientation framework for the alignment of individual use cases in this domain and a set of requirements for AI explanation design following the framework.

Authors 5

1. Andreas W. Mueller corresponding

   Schaeffler (Germany)

   Affiliation as printed

   Schaeffler Technologies AG & Co. KG , Herzogenaurach , Germany

2. Jonas Recker Aachen Laboratory for Machine Tools and Production Engineering (WZL)

   RWTH Aachen University

   Affiliation as printed

   Laboratory for Machine Tools and Production Engineering (WZL) , RWTH Aachen , Aachen , Germany

3. Alexander Keuper Aachen Laboratory for Machine Tools and Production Engineering (WZL)

   RWTH Aachen University

   Affiliation as printed

   Laboratory for Machine Tools and Production Engineering (WZL) , RWTH Aachen , Aachen , Germany

4. Günther Schuh Aachen Laboratory for Machine Tools and Production Engineering (WZL)

   RWTH Aachen University

   Affiliation as printed

   Laboratory for Machine Tools and Production Engineering (WZL) , RWTH Aachen , Aachen , Germany

5. Christoph Legat

   Technische Hochschule Augsburg

   Affiliation as printed

   Technical University of Applied Sciences Augsburg , Augsburg , Germany