Core Concepts

AI Model Cards 101: An Introduction to the Key Concepts and Terminology

Increase transparency by communicating key information about machine learning models using AI Model Cards.
In: Core Concepts

Model cards are a documentation format used in AI to provide standardized information about a machine learning model. They are similar to data sheets in other domains, such as product data sheets or material safety data sheets.

Model cards were introduced by researchers at Google in 2018 as a way to address the need for greater transparency and accountability in AI. Since then, they have gained popularity in the AI community as a way to promote responsible AI development and to help users make informed decisions about the use of machine learning models.

Model Cards Promote Transparency and Accountability

When creating a model card, it is crucial to strike a balance between simplicity and technical detail. It is essential to consider the target audience, which will vary depending on the AI system's intended purpose.

For instance, a model card for an AI system designed to assist medical professionals in interpreting x-rays to improve the diagnosis of musculoskeletal injuries is likely to be reviewed by various groups, including medical professionals, scientists, patients, researchers, policymakers, and developers of comparable AI systems. As a result, the model card may assume some familiarity with healthcare and AI systems.

What is covered in a Model Card?

Model cards typically cover several key items related to the machine learning model, including:

  1. Model details: This includes basic information about the model, such as its name, version, and type (e.g. neural network, decision tree, etc.), as well as the intended use case.
  2. Model architecture: This describes the structure of the model, including the number and type of layers, activation functions, and other key architectural choices.
  3. Training data and methodology: This covers information about the data used to train the model, such as the size of the dataset, the data sources, and any preprocessing or data augmentation techniques used. It also includes details about the training methodology, such as the optimizer used, the loss function, and any hyperparameters that were tuned.
  4. Performance metrics: This includes information about the model's performance on various metrics, such as accuracy, precision, recall, and F1 score. It may also include information about how the model performs on different subsets of the data (e.g. certain classes or samples).
  5. Potential biases and limitations: This covers any potential biases or limitations of the model, such as imbalanced training data, overfitting, or biases in the model's predictions. It may also include information about the model's limitations, such as its ability to generalize to new data or its suitability for certain use cases.
  6. Responsible AI considerations: This covers any ethical or responsible AI considerations related to the model, such as privacy concerns, fairness and transparency, or potential societal impacts of the model's use. It may also include recommendations for further testing, validation, or monitoring of the model.

The specific items covered in a model card may vary depending on the context and intended use of the model, but the goal is to provide transparency and accountability in the development and use of machine learning models.

Examples of Model Cards

In recent times, an increasing number of companies are choosing to share their models through open-source channels. One of the latest popular examples is Meta's LLama model. It has become common practice for models to be accompanied by a Model Card and a Research Paper upon release. HuggingFace has emerged as the largest public repository of open-source models, with the majority including Model Cards. This is particularly relevant for enterprises seeking to adopt and tailor these pre-trained models, as it is essential to comprehend the model's capabilities and any potential risks involved.

Model Cards are the Secret to Responsible and Effective AI Development

Some examples of Model Cards:

  • Meta LLama: A foundational, 65-billion-parameter large language model
  • OpenAI GPT-3: Generative Pretrained Transformer or “GPT”-style autoregressive language model with 175 billion parameters.
  • Google Face Detection: Model part of the Google Cloud Vision API

IBM's approach to AI Model documentation

IBM announced AI FactSheets, a document that contains factual information about the creation and deployment of an AI model or service. It can include details about the model's purpose and criticality, characteristics of the dataset, and actions taken during its development and deployment. Different roles in the machine learning lifecycle, such as the business owner, data scientist, model validator, and model operator, contribute information to the FactSheet. For instance, the business owner can specify the model's intended use, the data scientist can explain data manipulation activities, the model tester can describe testing measurements, and the model operator can provide performance metrics.

Collectively, these facts provide a comprehensive account of the model's construction, similar to how a school transcript or resume offers a more complete understanding of a student or job applicant. Check an example of a Mortgage Evaluator Model or Weather Forecaster Model. The AI Factsheets are created and updated automatically and can be exported in multiple formats such as Tabular Data, Markdown, Slides, or PDF. There are also templates to accommodate requests from the existing regulator in certain geographies and industries. This is really taking the concept of Model Cards to the Enterprise scale.

Persona view of IBM AI FactSheets
Persona view of IBM AI FactSheets

Why Model Cards are a must-have for AI Development

Overall, model cards are a useful tool for promoting responsible and transparent AI development and use, and can help ensure that machine learning models are used in a fair and equitable manner.

  1. Transparency and accountability: Model cards help promote transparency and accountability in AI development by providing standardized documentation of the machine learning model. This helps users better understand the model's strengths and limitations, as well as any potential biases or ethical considerations.
  2. Better decision-making: Model cards can help users make informed decisions about the use of machine learning models, such as whether a particular model is suitable for their use case, or whether further testing or validation is necessary before deploying the model in a production environment.
  3. Reduced bias and errors: By providing information about potential biases or limitations of the model, model cards can help developers and users identify and mitigate these issues, reducing the risk of biased or erroneous results.
  4. Increased collaboration and knowledge sharing: Model cards can facilitate collaboration and knowledge sharing among AI developers and users, by providing a standardized format for documenting and sharing information about machine learning models.
  5. Promotion of responsible AI: Model cards promote responsible AI by encouraging developers and users to consider the ethical and societal implications of the model's use, and to take steps to address any potential issues or concerns.

Get started today with Model Cards

If you're interested in using model cards in your organization, there are several steps you can take to get started. First, identify the machine learning models currently being used in your organization and determine whether model cards would be useful for documenting and sharing information about these models.

Next, consider developing a template or standard format for creating model cards, to ensure consistency and completeness of the documentation. You may also want to consider using tools or platforms that support the creation and sharing of model cards, such as the Model Card Toolkit developed by Google or the Model Card documentation on HuggingFace.

Finally, be sure to communicate the importance of model cards to stakeholders in your organization, such as developers, data scientists, and decision-makers. By promoting the use of model cards, you can help ensure that your organization is following best practices for responsible and transparent AI development.

In conclusion, model cards are a powerful tool for promoting responsible and transparent AI development, and can help organizations ensure that machine learning models are used in a fair and equitable manner. By adopting model cards in your organization, you can help build trust in AI and drive better decision-making for your business.

Unlocking the Full Potential of AI: A Deep Dive into Model Cards
Unlocking the Full Potential of AI: A Deep Dive into Model Cards
Written by
Armand Ruiz
I'm a Director of Data Science at IBM and the founder of I love to play tennis, cook, and hike!
More from

Introducing Large Vision Models - LVMs

LLMs have transformed text processing in AI and machine learning. Now, Large Vision Models (LVMs) are emerging, set to similarly revolutionize image processing and interpretation.

The History of AI

Foundation models are pivotal in AI evolution and essential in today's tech. In this post, we will understand AI's history is key to its future direction.

Accelerate your journey to becoming an AI Expert

Great! You’ve successfully signed up.
Welcome back! You've successfully signed in.
You've successfully subscribed to
Your link has expired.
Success! Check your email for magic link to sign-in.
Success! Your billing info has been updated.
Your billing was not updated.