BSI
Methodological Guide to Data Quality in AI Systems
With QUAIDAL, the BSI has published a catalog for the quality assurance of training data in AI applications, which supports compliance with regulatory requirements such as the European AI Regulation.
The European AI Regulation defines quality requirements for AI training data that cover aspects such as relevance, accuracy and completeness. The document series "QUAIDAL - Qualitycriteria for AI Trainingdata in AI Lifecycle" from the German Federal Office for Information Security (BSI) aims to translate these abstract quality requirements into concrete building blocks, measures and metrics & methods. It is intended to support compliance with regulatory requirements and increase technical traceability in the development process of AI systems.
BSI President Claudia Plattner: "We must ensure that applications with artificial intelligence meet high quality requirements. This is the only way we can produce and use trustworthy AI. With this catalog, the BSI offers very specific assistance that starts at the grassroots level."
QUAIDAL is aimed in particular at providers of high-risk AI systems, for which the AI Regulation defines detailed requirements in terms of documentation, data management and continuous quality assurance. In addition, this modular concept can be flexibly expanded in the future to take account of new technological developments.
The draft of the guidelines is now available on the BSI website and in two machine-readable GitHub repositories.










