Provenance Question-based AI Transparency and Accountable AI Governance

Laura Waltersdorfer, Dominique Hausler, Tanja Auge

Keywords: AI transparency, Provenance, Provenance question, AI governance

TL;DR: A framework supporting deriving provenance requirements from provenance questions for improved AI system transparency and governance.

Abstract: Ensuring transparency in Artificial Intelligence (AI) systems is critical for building trust and accountability. However, implementing technical governance and transparency in complex AI systems remains a challenge due to vague requirements, missing know-how and time resources. Provenance questions (PQs), outlining transparency requirements of a system, can play a key role in counteracting this. Nevertheless, the implementation of technical transparency and suitable PQs in complex AI systems pose significant challenges. This paper presents an approach for the formalisation and transformation of PQs, aimed at improving AI system transparency. This involves a question analysis on a linguistic and provenance level, based on the W7 model. To this end, we propose two definitions for simple and complex PQs to map them to PROV-O concepts, followed by a discussion of a reference architecture.

Submission Number: 17