Semantic AI Disclosure (SAID)

Semantic AI Disclosure (SAID) is a lightweight, standardized shorthand used to quickly communicate the degree of AI involvement in the creation of content.

Much like Semantic Versioning tracks software changes, SAID provides a clear, unobtrusive way to establish transparency and provenance. The annotations are format-agnostic and can be applied broadly to an entire project, or granularly to specific sections of text, code, or media.

The SAID Scale

Use the following marking system to indicate the degree of AI usage:

(0) - No AI Use: AI tools were not used.
(I) - AI Consultation: AI used for consultation.
(II) - AI Proofreading: AI was used for proofreading / spell checking.
(III) - Interactive AI Generation: Some content was generated by AI through interactive prompts, with significant human guidance and revision.
(IV) - Extensive AI Generation: Most or all content was generated by AI interactively but without significant revision.
(V) - Autonomous AI Generation: Content was automatically generated by AI without human interaction.

Combining Tags

These annotations can be used in a broad or granular way, and users may combine multiple tags. For instance, if someone uses AI for both consultation (I) and proofreading (II), they could write (I, II) or (I+II). Alternatively, users may wish to default to the highest level of intervention.

Summary

Applicability: SAID annotations are format-agnostic. They can be applied to text, images, video, code, or any other form of communication.
Adoption Model: The system is advisory in nature. It does not dictate policy, but rather provides a universal framework for transparent disclosure.
Granularity: Annotations are flexible. They can be applied broadly to an entire project or granularly (parsimoniously) to specific sections, paragraphs, or assets.

Background

In 2023, while working on edits for the AI Roadmap, Generative AI was still in its infancy but already proving to be a powerful tool for drafting professional documents. I found myself needing a way to quickly signal "I wrote this" without having to write out a lengthy disclaimer every time. I realized that as these tools became ubiquitous, professionals would need a standardized way to communicate their use of AI across varying degrees of involvement.

I envisioned a set of short codes that would immediately convey the level of AI integration to the reader. The goal was to create a qualification that was highly informative yet unobtrusive—a shorthand that wouldn't distract from the content itself. In 2024, I formally documented an early version of this scale in subsequent work on AI Guidelines within a shorthand scale for the City of Chicago's AI Policy Framework.