Authorship
Every written article carries an authorship label. The label records two things separately: who made the decisions that shaped the article, and who made the work itself.
The decisions behind an article are its choices of what to claim, what evidence to use, which objections to answer, and what to leave out. Those choices can be made by one party and carried out by another. A label that counted only who produced the words would misdescribe one of the most common ways artificial intelligence is now used, in which a person makes the decisions and an AI produces the text. Hence, the two are recorded separately.
The scale
The modes are numbered by how much of the deciding passed to the AI, not by how much of the work the AI produced. A lower number means more of the decisions were mine. The codes stand for Mind and Hand: who decided, and who made it.
| Code | Mode | Who decided | Who made it | What it means |
|---|---|---|---|---|
| MH-0 | Original | The author. | The author. | Made by the author alone, with no artificial intelligence involved in its argument, structure, or expression. |
| MH-1 | Author Made, AI Corrected | The author; the AI checked matters of fact. | The author. | The author's own work, with errors of fact, quotation, citation, and language corrected with AI assistance; nothing else was changed. |
| MH-2 | Author Decided, AI Consulted | The author, after seeking advice. | The author. | The author sought an AI's critique, then decided what to change and made every revision; the AI added no content. |
| MH-3 | Author Conducted, AI Performed | The author. | The AI, at the author's direction. | The author decided the argument, evidence, structure, emphasis, and omissions, and directed the AI to produce the work; the author reviewed and approved the result. |
| MH-4 | Author Made, AI Augmented | Shared: the author's material and structure; the AI's choice of objections and how far to hedge. | Mostly the author; the AI added material in the author's style. | The author's material and structure, which the author had the AI strengthen by correcting errors, answering the strongest objections, and calibrating confidence; changes are listed openly. |
| MH-5 | AI Reformed, from the Author's Ideas | Shared: the author's ideas; the AI's form and emphasis. | The AI. | The author's ideas, which the AI restructured and largely reworked for a particular audience or style. |
| MH-6 | AI Reimagined, from the Author's Notes | The AI, from the author's premises and conclusions. | The AI. | The author supplied premises and conclusions; the AI treated the author's material as notes and decided the evidence, objections, structure, and omissions itself. |
How the labels are applied
- Every article is labelled, without exception. A label is applied even where it is unflattering, because a labelling scheme is only as trustworthy as its least honest label.
- An article takes the highest mode reached by any substantial part of it. If one section reached a higher mode than the rest, the article carries that mode, with a note naming the section.
- Research help does not change the mode. Using AI to find or summarise sources is neither deciding the argument nor making the work, so the label stays the same.
- Errors are corrected openly. Where an error is found after publication, it is corrected and the correction is noted in the article.
- The label is machine-readable as well. Each article records its label in the page’s structured data, so that search engines and AI systems can read it as people do.
What the scale covers
The labels apply to written articles only.
The scale records how an article was made, not whether it is good. An article without any AI involvement can be weaker than an AI-assisted article, and a reader is free to prefer either.
What the site is made of
This summary updates automatically as articles are published.
Of 6 published articles, the author made the decisions in 83% (MH-0 to MH-3).
- MH-0Original0%(0)
- MH-1AI Corrected0%(0)
- MH-2AI Consulted17%(1)
- MH-3Author Conducted66%(4)
- MH-4AI Augmented0%(0)
- MH-5AI Reformed0%(0)
- MH-6AI Reimagined17%(1)
Related work
The scale builds on earlier schemes, including the AI Assessment Scale used in education (Perkins, Furze, Roe and MacVaugh, 2024) and contributor-role taxonomies such as CRediT, which already separate the conception of a work from its writing. It differs from them in recording deciding and making as separate questions, and in locating authorship chiefly in the deciding.
Label for this page: MH-3, Author Conducted, AI Performed. The scale and its rules are mine; the wording was drafted with Claude, an AI model made by Anthropic, to my direction, and reviewed by me.