National Institute of Standards and Technology (July 2024). Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile. NIST AI 600-1.

Source: https://doi.org/10.6028/NIST.AI.600-1

Relevant passage: section 2.7, “Human-AI Configuration,” p. 9. The profile describes both over-reliance on and unnecessary aversion to generative AI as risks. This is guidance, not evidence about how often either occurs.

Human-AI configuration concerns how people and AI systems work together. NIST describes two possible risks involving generative AI, which produces text or other content. Relying on an output means using it to help decide what to believe or do. Over-reliance means relying on an output beyond what its support justifies. Suppose generative AI produces the outputs in the earlier fictional visit-record task, and its past count is checked against the records. Relying on its estimate for next month solely because that count checks out would illustrate over-reliance: checking one claim does not by itself confirm another. NIST also warns that experts may unnecessarily avoid generative AI and miss useful assistance. These are risks to examine in a particular task, not measurements of how often they occur.

Learn Before