KEY LEARNINGS
- Transparency is not just about revealing code; it is about providing meaningful information so stakeholders can understand how AI affects them.
- A 'layered' approach to transparency is essential, providing simple summaries for end users and detailed technical documentation for regulators.
- Standardized documentation tools like Model Cards and Datasheets for Datasets are becoming the industry norm for disclosing system capabilities and data sources.
- Regulations like the EU AI Act and GDPR increasingly mandate disclosure, particularly regarding the existence of automated decision-making.
- Organizations can balance transparency with intellectual property protection by explaining the logic of decisions without revealing proprietary weights or algorithms.
- 📄Google Research: Model Cards for Model ReportingOriginal Google research paper on Model Cards.
- 📄Microsoft: Datasheets for DatasetsResearch paper on standardized dataset documentation.
- 🌐NIST AI Risk Management Framework: Transparency CharacteristicsNIST guidance on transparency in AI systems.
- Mitchell, M., et al. (2019). Model Cards for Model Reporting. Proceedings of the Conference on Fairness, Accountability, and Transparency.
- Gebru, T., et al. (2021). Datasheets for Datasets. Communications of the ACM.
- European Parliament and Council. (2024). Regulation (EU) 2024/1689 (EU AI Act).
