Opaque Provenance

Opaque provenance names the structural feature of AI-generated content by which an assertion cannot be traced to the specific sources that produced it. A printed book preserves a text: the text can be read, cited, verified, corrected, and argued about by anyone who holds a copy, and the provenance of every claim can be traced to a specific author, edition, and page. A large language model preserves something different — a statistical compression of millions of texts in which the model's 'knowledge' is not the texts themselves but an abstraction from them, a lossy compression that retains patterns while discarding the specific evidence from which the patterns were derived. When an AI system asserts something, the assertion cannot be traced to a source in the way a claim in a printed book can be traced to a citation. The user cannot inspect the evidence, evaluate the reliability of the sources, or distinguish between an assertion derived from peer-reviewed research and one derived from a forum post.