A short label that says what an AI model is and is not good for

orig. “Model Cards for Model Reporting” · Margaret Mitchell, Simone Wu, Andrew Zaldivar, Timnit Gebru

Fairness & Bias Beginner 3 min read Written, reviewed by Marginalia Editorial

In the margin

Heads up

Food comes with a nutrition label. This paper proposes the same idea for AI models, so people know where they work and where they fail.

What's going on

A model card is a short document that ships with a model. It says what the model was built for, how it was tested, and where it performs worse, for example across different groups of people. The goal is to surface gaps in fairness and reliability before the model is used in the real world, rather than after something goes wrong.

Why it matters

Model cards are now common practice at major AI labs and on model-sharing sites. They are a simple, low-cost step toward using AI responsibly, and they make it harder to quietly ship a model that fails for some people. This is a practical piece of the fairness and accountability conversation.

Source

Margaret Mitchell, Simone Wu, Andrew Zaldivar, Timnit Gebru, Google

View on arXiv ↗ Read the PDF ↗

Still fuzzy on a sentence?

Paste it and we'll explain it even more simply.

Test your understanding

Member notes Sign in ↗

ME Marginalia Editorial TEAM

The editorial team picked this paper because it is worth your time, and an AI drafted this note. We check each one and rewrite it when the draft misses the mark.