Accountability of Generative AI: Exploring a Precautionary Approach for "Artificially Created Nature"

Transparency alone cannot guarantee accountability: The paper finds that while transparency in AI systems is valuable, it is "not a sufficient condition for accountability" — meaning that simply being able to see how an AI works doesn't automatically ensure the system can be held responsible for its actions or failures.

Current AI systems are too complex to fully understand: Generative AI relies on "complex mechanisms that make it difficult for even experts to fully trace the reasons behind the outputs," creating a fundamental challenge where even specialists cannot explain why the system produced a particular result.

Unexplainable AI should be treated like natural phenomena: The paper uses the metaphor of "artificially created nature" to suggest that when AI systems become too opaque to understand, they should be regulated similarly to how we manage natural risks — with caution and precaution rather than assuming we can control them.

The precautionary principle should guide AI policy: Rather than waiting for problems to occur, the authors advocate for a precautionary approach that assumes potential risks from generative AI and acts to prevent harm before it happens.

Public participation is essential for managing AI risks: The paper proposes that "a platform for citizen participation is needed" to address generative AI risks, suggesting that accountability requires input from the broader public, not just technical experts and companies.