[4P1-OS-17b] OS-17

[4P1-OS-17b-04] AI Algorithm Transparency Toolkit for AI Governance

〇Masayuki Okamoto¹, Tonfi Yuki^2,3, Hiraku Murayama, Kirihito Yajima⁴, Yuta Nishigaki⁵, Hiroki Habuka^6,7, Naoko Munakata⁸ (1. Toyota Motor Corporation, 2. SmartNews, Inc., 3. ZeLo, a Foreign Law Joint Enterprise, 4. Aflac Life Insurance Japan Ltd., 5. The Legal Training and Research Institute of Japan, Supreme Court of Japan, 6. Kyoto University, 7. Smart Governance Inc., 8. Graduate School of Public Policy, The University of Tokyo)

Keywords:Trasparency, AI governance, AI regulation, AI guideline

We introduce a toolkit that presents a catalog of disclosure items regarding AI algorithm transparency. This toolkit is designed based on international discussions on the development of AI regulations and AI guidelines, various risk events generated by AI algorithms, and previous studies on AI algorithms. The toolkit is structured to serve as a systematic disclosure guide that allows discretionary selection of disclosures in terms of AI algorithm providers, users, and risks. We intend that this toolkit will be helpful for providers responding to self- and co-regulation and practicing agile governance in AI governance. This paper presents an overview of this toolkit, including disclosure items along the AI development life cycle, perspectives on utilization, granularity of transparency, and use cases. Although this toolkit mainly focuses on AI algorithms using machine learning or deep learning models, the discussion points are generally the same for other AI techniques, e.g., rule-based and knowledge-based models.