Users are the North Star for AI Transparency

"Transparency" means different things to different people, creating confusion between policymakers and AI developers. The paper identifies multiple competing definitions of AI transparency across the field, which leads stakeholders to "talk past each other" rather than solving actual problems. This fragmentation means policies may demand transparency without specifying what that actually means in practice.

The paper proposes a "north star" for AI transparency: systems should be user-centered, user-appropriate, and honest. Rather than pursuing abstract transparency goals, the authors argue the field should prioritize what users actually need to understand and trust AI systems, tailored to different users' knowledge levels and needs, while ensuring information is truthful.

Current transparency solutions often don't address underlying user concerns. Practitioners frequently develop technical transparency measures (like explainability tools) that may not match what policymakers or affected users actually want to know about how AI systems work and make decisions about them.

Clearer common language is essential for responsible AI governance. By mapping existing transparency concepts back to user-centered principles, the paper provides a framework for policymakers, technologists, and other stakeholders to communicate concrete demands and expectations, rather than continuing to work from incompatible definitions.

The field needs to move from vague transparency demands to specific, user-beneficial goals. The paper emphasizes that effective AI regulation and responsible development require clarity about why transparency matters (protecting users) and what transparency should accomplish, not just that it should exist.