Automatic Generation of Model and Data Cards: A Step Towards Responsible AI

Automated documentation can improve transparency: The paper demonstrates that using AI to generate model and data cards produces more complete documentation than human-created versions, addressing a significant gap in how AI systems are currently documented and made accountable to users and stakeholders.

Scale of the problem is substantial: The researchers analyzed over 4,800 existing model cards and 1,400 data cards, revealing widespread inconsistencies and incompleteness in how AI developers currently document their systems—a critical issue as open-source AI tools proliferate.

LLMs can enhance responsible AI practices: By automating the generation of model and data cards through large language models, the CardGen pipeline helps ensure that important information about AI systems' capabilities, limitations, and data sources is consistently captured and made available.

Standardized documentation supports accountability: The paper emphasizes that complete and objective model and data cards enable better "accountability and traceability," meaning stakeholders can better understand how AI systems work and identify potential risks or biases before deployment.

Completeness, objectivity, and accuracy matter: The research specifically highlights that automatically generated cards showed improvements in three key dimensions—completeness (more information included), objectivity (less biased descriptions), and faithfulness (accuracy to source materials)—all essential for responsible AI governance.