Model Card
Documentation describing an AI model capabilities, limitations, and appropriate uses.
Definition
A model card is standardized documentation that describes an AI model's capabilities, limitations, intended uses, and potential risks. Model cards help users understand what a model can and cannot do, enabling appropriate use and setting realistic expectations.
Good model cards include information about training data, performance across different demographics, known failure modes, and ethical considerations. They're becoming a standard practice for responsible AI development and deployment.
Why It Matters
AI models have non-obvious limitations and failure modes. Model cards help users avoid misuse by clearly communicating what to expect and what to avoid.
For organizations, model cards demonstrate responsible AI practices and help downstream users make informed decisions.
Examples in Practice
A model card warns that an image classifier performs poorly on certain skin tones, helping users avoid discriminatory applications.
Documentation reveals a language model's training data cutoff, preventing users from expecting knowledge of recent events.
A model card's intended use section clarifies that a summarization model shouldn't be used for legal document analysis.