News

Back
12/08/2021 | Gender/Diversity in Informatiksystemen

Full Grant Re­ceived from Volk­swa­gen­Stif­tung for Col­lab­or­at­ive Pro­ject "AI Forensics: Ac­count­ab­il­ity through In­ter­pretab­il­ity in Visual AI Sys­tems"

The project is part of the initiative "Artificial Intelligence and the Society of the Future".

We are delighted to announce that the research group Gender/Diversity in Informatics Systems (GeDIS) led by Prof. Dr. Claude Draude has received the full grant in the funding line “Artificial Intelligence and the Society of the Future” of the VolkswagenStiftung for the project "AI Forensics: Accountability through Interpretability in Visual AI Systems.” The project is coordinated by Prof. Dr. Matteo Pasquinelli (HfG Karlsruhe) and the consortium further involves  Prof. Dr. Leo Impett (Cambridge University), Prof. Dr. Fabian Offert (University of California Santa Barbara) and Prof. Dr. Noura Al Moubayed (Durham University). The joint grant is endowed with a budget of ca.1.400.000 EUR for a period of three years starting in 2022.

The project proposes AI forensics as a new sociotechnical framework for the analysis and critique of AI systems. This implies the design and development of new AI forensics tools for the investigation of large AI datasets (that are impossible to analyze manually), models (reverse-engineering the black box effect), and applications (exposing their sociohistorical context), making state-of-the-art results from explainable and interactive machine learning accessible to researchers in the humanities and beyond.

The resulting open source AI forensics toolkit is designed to be used by, and shared with civil organisations, academia and industry to study the social impact and implications of real-world AI systems. The AI forensics toolkit will also support four sociotechnical investigations that will examine the production pipeline of face recognition, interpretability and accountability of AI in the humanitarian sector, bias-aware design methods for AI systems, and the role of visual AI models in the sciences. Through a close integration of machine learning research and sociotechnical analysis, the project establishes a new and comprehensive methodological framework to equip the humanities and social sciences with the means to investigate the implications of 21st century artificial intelligence.

Further information can be found in the press release of the VolkswagenStiftung (in german).