Future of Creativity

Originality and Provenance in the AGI Era

What 'original' means when models are trained on enormous corpora of prior work, and how provenance systems are emerging in response.

Risograph illustration of human and AI creative collaboration — Plate / Creative direction is becoming a higher-leverage skill.

Executive summary

Originality has always built on prior work. AGI-class models make the inheritance explicit and contestable. New provenance systems — content credentials, watermarks, training-data attribution — are emerging to support both creators and audiences.

Key concepts

Originality as combination
Training-data lineage
Content credentials
Watermarking
Attribution

Originality as combination

Most originality is novel combination, framing, and execution of inherited ideas. AI makes the inheritance visible at a new scale, which is uncomfortable but not novel.

Training-data lineage

Models inherit from training corpora. Active research is improving attribution and detection of memorised content.

Content credentials and watermarks

Standards like C2PA attach verifiable provenance to media. Watermarking is improving but remains imperfect.

What audiences want

Audiences mostly want to know what they are looking at. Good provenance helps; perfect verification is not yet possible.

Key takeaways

01Originality is novel combination, not first creation.
02Training-data inheritance is now visible and contestable.
03Provenance standards are emerging.
04Audiences value disclosure even where verification is imperfect.

Frequently asked questions

Is AI-generated work plagiarism?

Sometimes, when it memorises specific source work. Mostly it is novel combination of patterns, which sits in the same category as most creative work.

How can I verify what I see?

Look for content credentials (C2PA), source-verifiable distribution, and provenance metadata. Treat unverified content as such.