The scrivener's dilemma: AGI and the crisis of authorship in publishing
"Image synthesis assisted by Qwen Image 3.0, an AI partner within the Global Future Nexus ecosystem."
The publishing industry is confronting a paradox of its own making. It has become the primary source of the high-quality text that trains the very systems threatening to make human authorship economically obsolete. In 2026, the number of ebooks released on Amazon surged from an average of 100,000 per month before ChatGPT's launch to approximately 300,000 by the end of 2025. Rakuten Kobo rejected 45% of submissions to its self-publishing platform last year, with roughly 80% of those rejections attributed to suspected AI content. The gatekeepers are drowning in a flood they helped enable.
The Collapse Risk
The most immediate danger is not that AI will replace human authors. It is that AI will consume itself. Publishers and researchers have warned that training future models on AI-generated content leads to model collapse—a degenerative process where systems reinforce their own errors, reduce output diversity, and ultimately decline in performance. As one analysis put it, "AIs in their current form are very dependent on human talent, so if publishers reduce the number of young, aspiring authors, future AIs will stagnate".
This creates an uncomfortable alliance. The publishing industry, which has legitimate grievances about the unauthorized use of its content for AI training, finds itself in the position of arguing that AI companies need human writers to survive. Dan Conway, CEO of the UK's Publishers Association, has described the unauthorized scraping of books for training data as "the great copyright heist". Yet the industry also acknowledges that AI has "incredible potential to make our businesses better, make our people more efficient, get more books into the hand of readers".
The Quality Collapse
Beyond the technical risk of model collapse lies a cultural one. A small Canadian publisher, Nimbus Publishing, has taken a hard stance against generative AI, with its managing editor stating: "As soon as you start engaging with generative AI to help you even do an outline, I don't feel like the work is yours anymore". The company's policy explicitly does not condone AI in creating text, artwork, editorial, or publicity materials.
The concern is not merely philosophical. Paul MacKay, who manages a Halifax bookstore, notes that readers are "already getting tired of seeing AI on the shelf". The Hachette cancellation of Mia Ballard's novel Shy Girl after reports of AI use in its creation became a cautionary tale for the industry. As one AI consultant and federal advisory council co-chair acknowledged, AI "will reduce the overall quality of what's being written and published".
The Transformation of the Editor
The editorial profession is being restructured in real time. In Italy, a survey found that approximately three in four publishers have used AI tools to some degree, primarily for metadata generation, manuscript classification, error correction, and translation. Yet Italian publishers emphasize that every book still passes through human editing. The editor's role is shifting from correcting minor errors to higher-value tasks: fact verification, structural shaping, protecting the author's voice, and bearing final responsibility for quality.
Chinese academic research frames this as a transformation from "content processing operators" to "content value curators". The editor's core competency shifts from "accuracy of experiential judgment" to "curatorial judgment, clarity of intent, and ethical responsibility". In a world where AI can generate any publishing decision, the human's role is to make the right one—and to bear the consequences.
The Attribution Problem
The industry is grappling with a fundamental question: can AI-generated content be detected, and should it be labeled? At a Qatar National Library forum, participants debated whether "a standardized, globally recognized label be required for all AI-generated content". The UK government's proposed "rights reservation" framework—allowing creators to opt out of AI training—has been criticized as unenforceable. As Conway noted, "there isn't a technological solution for an opt out and it's incredibly difficult, if not impossible, to enforce".
In China, the approach has been more prescriptive. The Measures for Labeling AI-Generated Synthetic Content require disclosure, and legal scholars warn that failure to label AI use in publications could trigger consumer protection claims. The Copyright Law's four-part test—literary/artistic/scientific domain, originality, reproducibility, and intellectual achievement—remains the standard, but its application to AI-assisted works is unclear.
The Path Forward
For Global Future Nexus, the publishing crisis is a microcosm of the broader AGI governance challenge. The same tensions—between innovation and protection, efficiency and quality, human agency and machine capability—are playing out across every domain that AGI touches.
The path forward requires three commitments.
First, transparency as infrastructure: content provenance systems that allow readers to know what they are reading and who—or what—created it.
Second, human value preservation: recognizing that the editor's role is not to compete with AI on speed, but to provide the judgment, taste, and accountability that no algorithm can replicate.
Third, sustainable training ecosystems: ensuring that the human authorship on which AI depends does not atrophy in the rush to automate it.
The publishing industry's dilemma is not whether to use AI. It is how to use it without consuming the human creativity that makes publishing worth doing in the first place. The alternative is a world where books are written by machines, for machines, about machines—and no one reads them.
Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)
Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)
