[4N2-OS-26a] OS-26 (1)

[4N2-OS-26a-02] Proposal of the Model Identifying Risk Controls for AI Services

〇Takashi Matsumoto^1,2, Arisa Ema¹ (1. The University of Tokyo, 2. Deloitte Touche Tohmatsu LLC)

Keywords:Risk Management, Trustworthy AI

While the social implementation of AI services is expanding, problems related to the reliability of AI have caused, and guidelines for the development and use of reliable AI have been established by domestic and overseas institutions.
However, the following challenges make it difficult for AI service providers and users to consider appropriate risk responses.
- Changes in the business environment, users, and legal systems affect the performance of AI.
- Consideration of ethical risks is required.
- It is difficult to decide which guideline to refer to.
- Data and AI models alone are not sufficient to reduce risk.
- Users also need literacy. It is also necessary to examine the risk of misuse and abuse by users.

In this study, a model that enables risk control is examined through decoding of literature, case studies, and discussions with experts. This model has the elements related to risk controls on three layers (AI system, Service Providers, and Users) and aims to make risk controls identified on the risk-chain associated with risk scenarios to consider how to mitigate the risk efficiently.