AI data transparency: an exploration through the lens of AI incidents

Data transparency significantly lags behind other forms of AI transparency: The paper cites a 2023 report showing that data transparency is notably behind other transparency areas in popular foundation models, indicating this is a widespread and persistent problem across the AI industry.

Companies provide minimal public documentation about their data practices: The research found that "low data transparency persists across a wide range of systems," meaning organizations building AI systems are not adequately disclosing information about what data they used, where it came from, or how it was processed.

Lack of transparency makes it harder to investigate AI problems: The paper identifies a critical barrier: when companies don't explain their models and systems clearly, it becomes extremely difficult for researchers, regulators, and the public to investigate data-related concerns when AI systems cause harm or raise red flags.

There's no standardized way to track data transparency across different AI systems: The researchers emphasize "a need to develop systematic ways of monitoring AI data transparency that account for the diversity of AI system types," suggesting the field currently lacks consistent standards or frameworks for measuring and comparing data transparency.

Understanding stakeholder needs is essential for improvement: The paper calls for better understanding of what information both data suppliers (those providing data) and data users (those using AI systems) actually need, suggesting that effective transparency solutions must be tailored to real-world requirements rather than imposed top-down.