May 6, 2021

“Towards Auditable AI Systems“: Fraunhofer HHI presents whitepaper

The Fraunhofer Heinrich Hertz Institute (HHI), together with TÜV Association and the Federal Office for Information Security (BSI) have published the jointly developed whitepaper entitled "Towards Auditable AI Systems". The whitepaper outlines a roadmap to examine artificial intelligence (AI) models throughout their entire lifecycle. By identifying the possibilities and limitations of a reliable safety assessment for AI systems, the researchers contribute to the EU-wide discussion  on a general regulation for AI technologies.

AI plays an increasingly important role as a vital part of various decision-making and control systems. However, highly efficient AI technologies such as deep learning pose the so-called black-box problem. Their decisions cannot be sufficiently traced and verified. Yet the security, robustness, interpretability and reliability of AI systems is crucial in safety-critical application areas such as mobility, biometrics and medicine.

To ensure that AI-based systems meet these requirements, an AI auditing process as well as specific AI certification procedures need to be developed. To this end, Fraunhofer HHI, TÜV Verband and BSI organized the workshop "Auditing AI Systems: From Basics to Applications" that took place in October 2020. Here, international AI experts met at the Berlin CINIQ Center/Forum Digital Technologies to lay the foundation for auditable AI. The participants have now published the outcome of their discussions in the latest white paper.

"As one of the leading research institutes in the field of explainable AI (XAI), we were able to create the scientific foundation for AI auditing and AI certification," says Dr. Wojciech Samek, head of the "Artificial Intelligence" department at Fraunhofer HHI and initiator of the AI for Good "Trustworthy AI" workshop series. In addition, Fraunhofer HHI is a leading member of the ITU/WHO focus group "AI for Health". Here, the institute contributes to the development of an audit procedure for AI applications in medicine. Within this framework, an innovative AI testing platform was recently completed.