Signs for Ethical AI: A Route Towards Transparency

AI systems need standardized transparency labels similar to traffic signs — The paper proposes a three-level transparency scheme with visual signs at the first level designed for "quick and universal interpretability," making it easier for ordinary people to understand what AI systems can and cannot do without requiring technical expertise.

Current AI regulation focuses only on data privacy, leaving other critical aspects unaddressed — While GDPR covers data privacy, the paper identifies significant gaps in regulation around "AI capabilities, purpose and source," suggesting that transparency requirements need to expand beyond existing privacy frameworks to cover how AI actually works.

Most people lack basic understanding of AI despite its widespread use — The authors note that "society remains mostly ignorant of the capabilities and standard practices of AI today," despite AI directly impacting billions of people in sectors like finance, health, security, and advertising, highlighting a critical knowledge gap between AI deployment and public awareness.

A tiered information approach can serve different user needs — The proposed scheme uses three levels (visual signs, factsheets with limited details, and full information access), allowing people to quickly understand AI systems at a glance while enabling those who want deeper information to access it.

Transparency should be built into AI systems from the start, not added later — The paper calls for "transparent by design" principles to be integrated during software development, suggesting that transparency is most effective when considered during initial design rather than as an afterthought.