On challenges and approaches to test AI systems

‍

This article featured in the renowned Datenschutz und Datensicherheit magazine (Issue 04/2025) and is written in English.

‍

The rapid evolution of AI, from rule-based systems to foundation models, has introduced significant complexities and challenges in testing AI systems. This article discusses inherent trade-offs such as standardization versus customization and measurement accuracy versus resource constraints.

‍

QuantPi’s Chief Scientist and Co-Founder, Dr. Antoine Gautier, Head of Policy and Grants, Lukas Bieringer and Senior Researcher, Dr. Stefan Klößner introduce an abstract approach to design test scenarios and tests for standardized yet customizable assessment.  

Read the full article to explore:

• The complexities and challenges in testing AI systems.  
• Two fundamental trade-offs in real-world testing of AI systems.
• An abstract approach to test biases and robustness aspects of AI models. 
• A standardized, yet customizable manner for consistent and comparable test results.  

Fill out the form to access the full article.

‍

‍

‍

‍

‍

About the authors

‍

Dr. Antoine Gautier

Antoine Gautier is Chief Scientist and Co-Founder at QuantPi, a leading European technology provider for AI testing. He holds an M.Sc. in applied mathematics from EPFL and did his PhD in the machine learning group at Saarland University. His research on nonlinear eigenvector problems and centrality measures in complex networks has been presented at top tier math and computer science conferences such as NeurIPS and SIAM Review. He serves as PI in multiple European and German research projects in the context of trustworthy AI and co-authors standardization documents on AI testing. 

‍

Lukas Bieringer

Lukas Bieringer works as Head of Policy & Grants at QuantPi. He holds an M.Sc. from Heidelberg University and previously served as Head of Technology Transfer at the CISPA Helmholtz Center for Information Security. As a member of multiple working groups at German national level, he co-authored standardization documents in the context of AI testing such as the 2nd version of the German standardization roadmap for AI and DIN SPEC 92006 on requirements for AI test tools. His research on the practical relevance of security threat models for AI systems has been published at leading computer science venues such as ICML, AAAI, and SOUPS.

Dr. Stefan Klößner

Stefan Klößner works as a Senior Researcher at QuantPi. After pursuing diplomas in Business Administration and Mathematics, he did his PhD in Economics at Saarland University. He also holds a habilitation in Econometrics and Statistics at Saarland University, where he is still lecturing regularly. His research results in various areas of economics, econometrics and statistics have been presented at numerous conferences and published in renowned journals such as the Journal of Business and Economics Statistics. 

‍