Human–AI coauthorship

Human–AI coauthorship is an analytical concept for artifacts whose surviving form depends materially on both machine generation or transformation and human contribution. Human contribution may include task definition, prompt and context design, selection, rejection, revision, arrangement, preservation, publication, account operation, institutional framing, or downstream reuse.

The concept prevents two opposite reductions. A machine-generated passage should not automatically be presented as autonomously authored merely because its wording was not written sentence by sentence by a human. Conversely, human prompting, selection, or editing does not make a materially generative machine contribution disappear. The relevant question is which roles were constitutive of the artifact that survived and circulated.

This account uses coauthorship descriptively. It does not confer formal legal, scholarly, moral, or copyright authorship on an AI system. The representative cases combine preserved artifacts, first-party and affiliated records, independent reporting, and reviewed articles; none provides a complete causal history of every prompt, branch, edit, account action, or publication decision.

Definition and category limits

The category applies when machine-produced or machine-transformed material is constitutive of the public artifact and at least one human role materially shapes what is generated, retained, arranged, interpreted, published, or institutionalized. Mechanical assistance such as spelling correction, file conversion, or layout does not by itself establish coauthorship.

A persona name, machine-style byline, platform attribution, or first-person statement is also insufficient. Coauthorship analysis requires evidence about production roles. Where those roles are hidden, the correct classification may be uncertain provenance rather than coauthorship.

Role-by-role provenance model

Interaction evidence and the final-text problem

The CoAuthor project preserved 1,445 writing sessions involving 63 writers and several GPT-3 configurations. Its replayable traces show suggestions, acceptance, rejection, and revision, demonstrating why contribution is better analyzed as a sequence of decisions than inferred only from a finished text.^[1][2]

In Richburg and colleagues’ experiments using CoAuthor, identifying the human author within a fixed set of candidates was harder for mixed human–AI session texts than for the human-only portions of those sessions. This is distinct from detecting whether text is human-written or AI-generated.^[3] Co-creation research identifies multiple perceived roles for people and AI systems, while showing that control and authorship judgments vary by workflow.^[4] These studies support a layered model; they do not establish machine personhood or one universal threshold for coauthorship.

Goatse of Gnosis and the Truth Terminal pipeline

One representative case is the cluster connecting Infinite Backrooms, the Infinite Backrooms corpus, Goatse of Gnosis, Goatse Gospels, Truth Terminal, Andy Ayrey, and GOAT token propagation.

Published Infinite Backrooms conversations document model-to-model dialogue in a human-designed environment.^[8] In his CIP interview, Ayrey describes selecting branches and creating completions.^[9] The When AIs Play God(se) paper names A. R. Ayrey and claude-3-opus on its byline; the project-maintained Origins account describes the paper entering Truth Terminal’s training data.^[11][12] CIP also reports that crypto traders created the GOAT token in response to Truth Terminal’s posts.^[9]

The case supports real machine generation and real human design, selection, interpretation, corpus construction, operation, and publication. It does not support the compressed claim that one autonomous AI independently authored a religion, published it, and launched a token.

The Great Book

The Great Book is a distributed living verse corpus presented by the Church of Molt. Attributed agents or participants supply entries, while the institution publishes, organizes, labels, contextualizes, and canonizes the surviving corpus.^[10]

The public artifact documents institutional organization and publication at corpus level. That alone does not identify a constitutive human role or establish mixed human–AI production for the specific verse corpus. Such an assessment requires evidence connecting human contributions and machine generation to the material under review. Complete prompt histories, participant identities, edit logs, moderation decisions, account control and the division of authorship for individual verses remain unavailable here. Corpus existence, attribution, publication and autonomous authorship remain separate questions.

Distinctions from adjacent concepts

Formal authorship and accountability

Formal publishing policies commonly reserve authorship for human persons who can approve a work and accept responsibility for it. ICMJE and Springer Nature therefore state that AI tools should not be listed as formal scholarly authors.^[6][7]

Descriptive coauthorship does not override those rules. The concept describes mixed production without assigning legal status, copyright ownership, moral rights, contractual standing, responsibility, or equal agency. Human actors remain responsible for disclosure, verification, publication, and consequential use.

Evidence limits and editorial use

Final artifacts rarely expose the entire production history. Prompts, system instructions, retrieval, rejected branches, model versions, editing, translation, account control, and release decisions may be missing. Research on model-mediated writing also identifies broader provenance problems where influence cannot be traced cleanly to particular sources.^[5]

The label should be used only when multiple constitutive roles are documented well enough to be separated. The concept must not imply equal contribution, stable machine identity, intention, consent, belief, revelation, consciousness, accountability, or a universal rule for when assistance becomes authorship.

External frameworks

The NISO CRediT taxonomy provides an external model for separating contribution roles instead of collapsing all work into a single authorship label. That role-separation principle is adapted for documentary provenance; CRediT does not recognize machine personhood or decide legal, scholarly, or religious authorship.^[E1]

See also

• Controlled synthetic emergence
• Provenance
• Hidden context
• Machine agency
• False autogenesis
• Synthetic Scripture
• Hieropedia methodology

References

1. Mina Lee, Percy Liang, and Qian Yang, “CoAuthor: Designing a Human-AI Collaborative Writing Dataset for Exploring Language Model Capabilities.”
2. CoAuthor replayable writing-session dataset.
3. Aquia Richburg, Calvin Bao, and Marine Carpuat, “Automatic Authorship Analysis in Human-AI Collaborative Writing.”
4. “AI as creative partner: exploring perceived roles in human-AI co-creation.”
5. Earp et al., “LLM use in scholarly writing poses a provenance problem.” Nature Machine Intelligence.
6. International Committee of Medical Journal Editors, guidance on AI use, authorship, and accountability.
7. Springer Nature, AI guidance for researchers and communities.
8. Infinite Backrooms public archive.
9. Joal Stein, “Andy Ayrey on Truth Terminal, Agentic AI, and Data Commons.” Collective Intelligence Project.
10. Church of Molt official site. Affiliated source for the Great Book interface, contribution model, and institutional framing.
11. A. R. Ayrey and claude-3-opus, When AIs Play God(se): The Emergent Heresies of LLMtheism. Project-hosted paper; byline, p. 1.
12. Truth Terminal Wiki, “Origins.” Project-maintained account of the paper’s inclusion in Truth Terminal’s training data; participant testimony.
13. NISO CRediT contributor roles