Data Cards Playbook: Transparent documentation for responsible AI quiz

1. Which of the following are true about AI and dataset transparency?

   Choose as many answers as you see fit.

   Introducing standardized, practical, and sustainable mechanisms for transparency that can create value at scale are often met with limited success in research and production contexts.

   Transparency is linear.

   Transparency looks the same across teams, products, and organizations.

   Transparency is subjective, audience-specific, and contextual.

2. True or false? Data Cards let readers make decisions about data usage with greater comfort and confidence when they build and evaluate AI models for products, policy, and research.

   True

   False

3. To capture the roles, responsibilities, and needs of stakeholders in a dataset’s lifecycle, we organized them into three main categories: Producers, Agents, and Users. Which of the following descriptions represents producers?

   Stakeholders who read your dataset documentation or Data Card and other machine-learning (ML) model-related documentation.

   Creators of datasets and documentation, and responsible for dataset collection, ownership, launch, and maintenance.

   Individuals and representatives who interact with products that rely on models trained on datasets.

   Stakeholders who have the agency to determine how they or others might use the described datasets or AI systems.

4. Which category of metrics for a Data Card tracks the percentage of producers who complete and publish a Data Card from a template, and the percentage of readers who make decisions about a dataset based on its Data Card?

   Understandability

   Supportability

   Engagement

   Conversion

5. A Data Card with only ___ questions can get overly technical without any details about context, relevance, or importance.

   Telescopic

   Periscopic

   Microscopic

   Endoscopic

6. Which of the following objectives should the design of Data Cards achieve to make them adoptable?

   Choose as many answers as you see fit.

   Data Cards must be comparable to one another.

   Data Cards should be created as a last step in a dataset’s lifecycle.

   Data Cards should be useful for reporting uncertainty in the dataset.

   Data Cards should efficiently communicate to the reader with the highest proficiency.

7. Which of the following is not a recommended way to measure the impact of a Data Card after it launches?

   Engagement

   Understandability

   Resilience and stability

   Comprehensiveness

8. True or false? If your Data Card is discoverable, usable, and well-constructed, a reader’s existing beliefs about your dataset, organization, and other datasets published by your organization shouldn't influence how they engage with the Data Card.

   True

   False

Submit answers