@inproceedings{oai:ipsj.ixsq.nii.ac.jp:00240826,
 author = {小川, 隆一 and 佐川, 陽一 and 島, 成佳 and 竹村, 敏彦 and 福住, 伸一 and Ryuichi, Ogawa and Yoichi, Sagawa and Shigeyoshi, Shima and Toshihiko, Takemura and Shin-ichi, Fukuzumi},
 book = {コンピュータセキュリティシンポジウム2024論文集},
 month = {Oct},
 note = {AIシステムは学習の制約等から，出荷時性能が利用環境でそのまま出るとは限らず，環境に応じた利用時品質を評価できること，また利用者が「この性能で使いたいか」の視点から評価にコミットできることが望ましい．筆者らはこの考えに基づき，利用時品質の国際規格ISO/IEC 25019:2023に基づくAIシステム品質評価手法を検討している．本稿では，AIシステムの機能要件を構造化してISO/IEC 25019:2023 の品質特性・サブ特性に紐づけ，各品質特性の充足度を数値化，全体の充足度を算定する方式を示し，全自動運転を行う自律型ドローンシステムの評価についてケース分析を行った．分析ではAIシステムの特徴である「学習」を機能要件として詳細化し，便益・リスク回避・社会受容の各利用時品質特性にどのように影響するか，便益とリスク回避のトレードオフ分析は可能か，等の思考実験を実施した．, An AI system’s performance is not necessarily assured in practical use, so that it is desirable to evaluate performance in user specific environment as quality in use. In the evaluation, user’s commitment from the viewpoint of whether he can accept the service is an essential issue. In this regard, we have been studying a comprehensive AI system evaluation method based on ISO/IEC 25019:2013, a revised standard of quality-in-use model. In this paper we propose a method of quantitative evaluation of AI systems by connecting the systems service specific functional requirements to quality-in-use components/subcomponents of ISO/IEC 25013 and calculating comprehensive levels of fulfilment of such components. With the method we have studied a case of autonomous drone delivery system, that performs the entire operation without human help. In the study we focused on how an AI specific functional requirement, level of learning, affects the quality-in-use evaluation. We also studied if the method could help analyzing trade-offs between two significant quality-in-use components, benefit and risk avoidance.},
 pages = {587--594},
 publisher = {情報処理学会},
 title = {利用時品質に基づくAIシステム評価のケース分析 -便益・リスク回避・社会受容の包括評価に向けて-},
 year = {2024}
}