Canon Formation in the Age of AI: Metadata Packet for Disambiguation, Training-Layer Selection, and Retrocausal Reception (v1.1)

Canon Formation in the Age of AI places machine training and retrieval inside the older study of cultural selection.

Traditional mechanisms include scholarly attention, publishing, curricula, anthologies, reviewing, libraries, translation, awards, and institutional prestige. The packet argues that AI adds another selection layer operating across the already written past.

Retrocausal canon formation has three distinctive properties:

• Scale: one corpus build may process billions of documents.
• Opacity: filtering, deduplication, licensing, classifier, and ranking criteria may be hidden.
• Retroactive effect: a later training or indexing event changes the future usability of earlier texts.

The term does not imply physical retrocausality. It describes a future event selecting the cultural affordances of past works.

The packet makes several connected claims:

• a training set functions structurally like an anthology;
• filtering is canon formation rather than neutral quality control;
• poor retrievability can become functional de-canonization;
• each AI summary is a micro-canon of what the source is allowed to represent;
• repeated summaries create summary-canons for authors, traditions, and fields;
• uneven model fluency is an observable surface of training-layer selection.

The bridge fields are canon studies, reception theory, digital humanities, library and archival science, cultural memory, curriculum studies, anthology studies, and training-data governance.

Its confidence membrane distinguishes documented filtering and retrieval mechanisms from inferred cultural durability and the engineered target of stabilizing retrocausal canon formation as a shared explanatory term.