Four researchers propose a common language, aligned with ISO/IEC standards, to catalogue AI uses and their risks, and to facilitate compliance with several regulations at once.
The European AI Act requires providers of "high-risk" systems to manage and document the risks they pose. However, it is necessary to be able to describe them in a coherent, interoperable and machine-searchable way, while new risks are constantly emerging. Delaram Golpayegani, Harshvardhan J. Pandit, Declan O'Sullivan and Dave Lewis present two tools designed for this purpose.
An ontology and a vocabulary
AIRO (AI Risk Ontology) is used to describe an AI use case and the risks associated with it; VAIR (Vocabulary of AI Risks) complements it with ready-to-use concept taxonomies. Their design is based on two international standards: ISO/IEC 23894, dedicated to AI risk management, and ISO/IEC 22989, which sets out the terminology.
To promote their reuse, the two resources have been integrated, following a community consensus process, into the Data Privacy Vocabulary (DPV), already used to document compliance with the GDPR. DPV thus extends its scope from personal data to AI, with the ambition of allowing the same documentation to meet several regulatory regimes.
Read the full article on the original source
Go to the source →