Understanding our Model Cards and Their Role in Responsible AI

[Ashley Snyder]

At ServiceNow, we are committed to creating AI technology that is human-centered, inclusive, transparent, and accountable. One way we achieve this is through the use of model cards. But what exactly are model cards, and why are they crucial for Responsible AI?

 

What is a Model Card?

A model card is like a nutritional label for an AI model. Just as a nutritional label provides consumers with important details about the contents and nutritional value of food products, a model card offers comprehensive information about an AI model. This transparency is vital for understanding the capabilities and limitations of AI models. A model card includes several key components:

• Model Purpose: This describes what the model is designed to do, its intended use cases, and the problems it aims to solve. It helps users understand the context in which the model can be applied effectively.
• Training Data: Provides information about the data used to train the model, including its sources, any preprocessing steps, and its diversity. This ensures users are aware of the dataset's scope and potential biases.
• Performance Metrics: Details how well the model performs, often broken down by different user groups or scenarios. Metrics might include accuracy, precision, recall, and other relevant performance indicators, helping users evaluate the model's reliability.
• Limitations: Highlights any known weaknesses or situations where the model might not perform well. This section is crucial for setting realistic expectations and preventing misuse.
• Ethical Considerations: Discusses potential ethical impacts of the model’s use, including any biases identified during development and steps taken to mitigate them. This promotes responsible and fair use of AI.

By providing this information, model cards help stakeholders understand an AI model's strengths and limitations, fostering trust and enabling more informed decisions.

 

 

Model Cards and Responsible AI

Responsible AI is about developing and deploying AI systems in ways that are ethical, transparent, and accountable. Model cards play a significant role in this effort by:

1. Enhancing Transparency: They provide vital details about AI models, making it clear how they work and under what conditions they perform best.
2. Ensuring Accountability: By documenting the model’s development and performance, model cards hold creators accountable, allowing for traceability in case of unintended harm or bias.
3. Promoting Fairness: Performance metrics often include evaluations across different demographic groups, highlighting any biases and encouraging the development of fairer AI systems.
4. Guiding Ethical Use: By outlining potential ethical considerations, model cards provide guidance on the responsible and appropriate use of AI models, preventing misuse and harm.

 

At ServiceNow, we help provide information to empower AI choices by providing product documentation and model cards that explain our AI systems in clear, understandable terms. To learn more about ServiceNow's dedication to Responsible AI, visit our Responsible AI Guidelines. 

 

How model cards align with the NIST AI Risk Management Framework

The National Institute of Standards and Technology (NIST) has developed the AI Risk Management Framework (AI RMF) to provide guidance on managing risks associated with AI systems. This framework emphasizes four key functions: Govern, Map, Measure, and Manage. Model cards align with these functions as follows:

• Govern: Model cards support governance by ensuring that AI models are developed and used transparently and accountable.
• Map: They help map out the purpose, scope, and context of AI models, making it easier to understand and manage their impact.
• Measure: By providing detailed performance metrics, model cards help measure the effectiveness and fairness of AI models.
• Manage: They assist in managing risks by highlighting limitations and ethical considerations, enabling proactive risk mitigation.

For more information on the NIST AI Risk Management Framework, you can visit their official documentation.

 

Resources

• ServiceNow Model Cards
• NIST AI Risk Management Framework
• ServiceNow Responsible AI Guidelines