AI Risk Management Framework (AI RMF 1.0)
A voluntary framework from the U.S. National Institute of Standards and Technology for organizations managing risk across an AI system’s lifecycle, built around four functions (Govern, Map, Measure, Manage) and seven characteristics of trustworthy AI. AI Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1) is NIST’s own companion profile mapping this framework onto risks specific to generative AI.
License: Public domain — U.S. federal government work (17 U.S.C. § 105).
Key points
- Human-AI configurations span a spectrum from fully autonomous AI decision-making to a human expert making the final call with AI as one input — roles need to be defined explicitly per system rather than left implicit. See Human-AI Interaction.
- Transparency, explainability, and interpretability are distinct questions: transparency answers “what happened,” explainability answers “how” a system reached its output, interpretability answers “why” that output and what it means to the user. See Explainable AI.
- Human-AI pairing can amplify bias or improve on it depending on how it’s organized — in some perceptual-judgment tasks the AI component amplifies a human’s existing bias, while deliberately structured human-AI teaming can instead produce complementary, better-than-either-alone results. See Human-AI Interaction.
- Appendix A names “human factors experts (e.g., UX/UI design)” as a distinct actor category across the AI lifecycle, responsible for human-centered design, active end-user involvement, and evaluating end-user experience.
- Mitigating harmful bias doesn’t by itself make a system fair or accessible — a demographically balanced system can still be inaccessible to people with disabilities or shut out by the digital divide. See Accessibility.