Grounding Trustworthy AI: from high-level principles to technical requirements

Lawfulness, ethics and robustness stand as the overarching principles of Trustworthy AI. The Requirements for Trustworthy AI (TAI Requirements), that are also partially included in the EU proposal for a Regulation on Artificial Intelligence (AI Act), further specify these principles.

This proposal focuses on the requirements that an AI system should be compliant with in order to be considered trustworthy. As neither the scientific nor the industrial communities have yet developed robust guidelines, theories or techniques on how to measure whether an AI system complies with these requirements, or on how to create trustworthy-by-design AI systems – the main objective of the project is to translate these requirements into measurable KPI’s and bring high-level principles into practicable technical and non-technical methodologies.
To this end the project will 1) provide metrics and tools to measure and possibly repair existing AI systems w.r.t. trustworthiness 2) provide a controlled experimentation environment to concretely test trustworthiness and 3) define well-founded methodologies on how to develop trustworthy-by-design AI systems.