What is an AI Model Card? Model Cards vs Authority Artifacts
A model card is a short document describing a machine learning model: what it is for, how it was evaluated, and its known limitations. Model cards were proposed in 2019 ("Model Cards for Model Reporting"). An authority artifact is different: it states the rules a system must act within, and systems check it before acting.
Describing versus constraining
A model card tells a reader what a model is like. It does not stop the model doing anything. An authority artifact is evaluated: an action outside its rules is denied or sent to a person.
Using both
A model card is useful evidence about a model whose output feeds a decision. The authority artifact holds the rule that decides what to do with that output, and records which rule applied.
Example
A vendor supplies a credit model with a model card reporting 87% accuracy.
The bank files the model card and uses the model's output directly.
The bank keeps the model card, and seals the rules that act on the model's score: thresholds, escalations and exceptions, each citing its policy clause.
Frequently asked questions
Related concepts
See it in practice
Upload a policy, confirm its rules, seal them: any system can then check an action against them.