91Â鶹¹ú²ú¾«Æ·×ÔÅÄ

This session aims to identify the key elements of meaningful transparency while exploring it in different contexts of uses of AI systems, looking to all stages of the AI system life cycle.  

Stakeholders typically agree on the need to ensure transparency in order to increase accountability and trust in AI systems and platforms. However, divergence emerges when it comes to the practical implementation of transparency in AI systems. 

For instance, regarding the algorithmic moderation tools, transparency is considered an essential element for improving platform governance, understanding and addressing digital harms and for ensuring users' rights in the context of AI systems used in content moderation.  

As noted by the UNESCO Draft Guidance for regulating digital platforms, the Windhoek+30 Declaration on Information as a Public Good calls for the promotion of increased transparency of relevant technology companies and media viability, principles unanimously endorsed by UNESCO’s Members States. Yet, transparency is a remarkably poorly defined concept, and there is a risk that poorly formulated regulations, even when centered on transparency, may contribute to further internet fragmentation or fail to address harms effectively. As noted by the IGF Coalition on Platform Responsibility 2022 Outcome Document, to make transparency meaningful, any regulatory policy should clarify the object of transparency, its audience, why disclosing the information is essential, and its goals. In addition, more information is not always better: transparency should not just be measured in terms of quantity, but in terms of quality. 

defines procedural and transparency obligation as a multilevel system involving the entire AI life cycle, from decision making processes to transparent ethical impact assessment; from explainability to the eventual need of sharing of code or datasets in cases of serious threats of adverse human rights impacts.