Authors: Priyabanta Sandulu, Andrea Šipka, Sergey Redyuk, Sebastian J. Vollmer
Published in: 28th European Conference on Artificial Intelligence (ECAI) TRUST-AI: The European Workshop on Trustworthy AI (2025)
The rapidly evolving regulatory landscape in AI presents significant challenges to establishing and maintaining trust. AI practitioners face a substantial burden in understanding and operationalizing abstract requirements. Existing solutions often lack concrete strategies for effective risk mitigation. We address these gaps by proposing an AI debugger, powered by an expandable knowledge base, that identifies risks and suggests actionable mitigation with little overhead to the end-user. A Human-in-the-Loop component supports adaptive decision-making, and the unique Requirement & Knowledge Engineering pipeline suggests the mapping between abstract guidelines and actionable specifications, pending validation by the end-user. Our framework aims to reduce the compliance overhead and streamline the development of trustworthy AI systems.
@inproceedings{sandulu2025,
title = {Actionable Trustworthy AI with a Knowledge-based Debugger},
author = {Priyabanta Sandulu and Andrea Šipka and Sergey Redyuk and Sebastian J. Vollmer},
year = {2025},
abstract = {The rapidly evolving regulatory landscape in AI presents significant challenges to establishing and maintaining trust. AI practitioners face a substantial burden in understanding and operationalizing abstract requirements. Existing solutions often lack concrete strategies for effective risk mitigation. We address these gaps by proposing an AI debugger, powered by an expandable knowledge base, that identifies risks and suggests actionable mitigation with little overhead to the end-user. A Human-in-the-Loop component supports adaptive decision-making, and the unique Requirement & Knowledge Engineering pipeline suggests the mapping between abstract guidelines and actionable specifications, pending validation by the end-user. Our framework aims to reduce the compliance overhead and streamline the development of trustworthy AI systems.},
}