Comments on NIST AI RMF RFI
Abstract
In traditional systems engineering and risk, we have a model of the system to which we can apply statistical and decision theoretic approaches to risk management. With AI Systems, both the system structure and system state are evolving, and the time constants on the dynamics of systems state and systems structure are different. All of that contributes to the complexity of AI systems. One of the greatest challenges is getting actors to see the whole system and hold the inherent complexity. Many want to approach AI systems and their risks linearly, tracking cause and effect. With AI, a necessary shift is to consider emergent issues and risks as components of interconnected and interacting systems rather than as independent issues with unrelated consequences. Addressing a risk likely means creating new vulnerabilities and new systems tradeoffs. Improvements in management of AI-related risks requires new approaches that reflect a whole systems perspective. As part of that, organizations need new approaches that broaden the scope of risk-based decisions to include opportunistic risk as well as possible threats.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2021
- Accession Number
- AD1148048
Entities
People
- Brett Tucker
- Carol Smith
- Nathan Van Houdnos
- Rachel Dzombak
- Ramayya Krishnan
Organizations
- Carnegie Mellon University