Im Rahmen der „Foundational Research on AI Systems“ untersucht das us-amerikanische National Institute of Standards and Technology (NIST) bereits seit geraumer Zeit unterschiedliche Aspekte von Systemen mit künstlicher Intelligenz (KI-Systeme). Am 20. August 2020 veröffentlichte das NIST ein White Paper als „draft report“ mit vier Prinzipien zur Nachvollziehbarkeit von Entscheidungen durch KI-Systeme. Diese Four Principles of Explainable Artificial Intelligence (Draft NISTIR 8312) lauten wie folgt:
(i) Explanation: AI systems should deliver accompanying evidence or reasons for all outputs.
(ii) Meaningful: Systems should provide explanations that are understandable to individual users.
(iii) Explanation Accuracy: The explanation should correctly reflect the system’s process for generating the output.
(iv) Knowledge Limits: The system only operates under conditions for which it was designed or when the system reaches a sufficient level of confidence in its output.
(ii) Meaningful: Systems should provide explanations that are understandable to individual users.
(iii) Explanation Accuracy: The explanation should correctly reflect the system’s process for generating the output.
(iv) Knowledge Limits: The system only operates under conditions for which it was designed or when the system reaches a sufficient level of confidence in its output.
Zu diesen Prinzipien hält das NIST u.a. folgendes fest: „(…) These principles are heavily influenced by an AI system’s interaction with the human receiving the information. The requirements of the given application, the task, and the consumer of the explanation will influence the type of explanation deemed appropriate. Our four principles are intended to capture a broad set of motivations, applications, and perspectives".
Das NIST nimmt bis 15. Oktober 2020 Kommentare zum „draft report“ entgegen.
Michal Cichocki
Das NIST nimmt bis 15. Oktober 2020 Kommentare zum „draft report“ entgegen.
Michal Cichocki