AI-generated "two-headed shrimp" appears in a popular science book; People's Daily: the root cause is not AI but publishing gatekeeping

“Well, now look at this—two-headed shrimp have appeared. Hurry up and apply to register them as a new farmed species.” Recently, a reader complained on social media that the “South China Sea Marine Life Popular Science Series,” which they had purchased, contained an erroneous image of a “two-headed shrimp” in its introduction to the black tiger shrimp. Staff at the publisher involved responded that they are notifying the entire internet to take the book off shelves and are recalling the books involved.

The book’s preface states that the vast majority of the images in the book were generated by AI (artificial intelligence). Problems caused by AI are not isolated cases; they have occurred many times in recent years. This exposes the current problems of unclear and incomplete boundaries and norms for AI use. At present, China has launched the formulation of several standards, such as the “Academic Publishing Specification: Labeling and Statements for the Use of Generative Artificial Intelligence,” the “Security Requirements for Artificial Intelligence Applications in the Publishing Industry,” and the “Requirements for Artificial Intelligence Corpus Processing in the Publishing Industry.” It is believed that as institutions become increasingly sound, the use of AI will become more standardized and rigorous.

Technology is only a tool, not the subject of responsibility; AI has no original sin—the problem lies with users and gatekeepers. To a certain extent, AI is merely a magnifying glass; its intervention magnifies safety hazards that already existed. Even without AI, many problems in publications are often exposed online. For example, a “Romance of the Western Chamber” published by a certain publisher was pointed out to contain more than 300 problems, and a “Three Hundred Tang Poems” launched by a certain publisher was found to contain multiple errors in the correspondence between simplified and traditional characters.

At present, some publishing institutions excessively pursue popularity and traffic, fearing “missing a good opportunity,” which has led to the publication cycles of many books being compressed and the “three reviews and three proofreadings” mechanism being suspended. Ultimately, “publishing speed” and “economic benefits” override everything, and quality becomes the victim. This distorted profit-seeking logic and the lack of requirements for quality are the fundamental causes of the errors.

Safeguarding publishing safety and quality is a consensus and even more a bottom line. Publishing institutions must strictly implement their principal responsibilities, firmly tighten and solidify the gatekeeping responsibilities in processes such as editing, review, and proofreading, and build layer upon layer of defense from manuscript to finished book; for AI applications, they must carefully study and assess, add verification and proofreading steps, and provide labeling and notification so that consumers know whether a book is an individual creation or the product of human-machine collaboration. Regulatory authorities should take the initiative and act proactively, improve the normalized whole-chain regulatory mechanism, and use strong supervision and severe penalties to force the industry to achieve self-purification.

Whether the quality gate of publishing and distribution can be guarded does not depend on whether some new technology is used, but on whether workers in the industry still respect this gate and whether they have left “cracks” in it. Only by making reverence a habit, giving institutions teeth, and making technology serve people can we live up to the fragrance of books and to readers.

(Excerpted and edited from Yunnan Daily; original title: “A Popular Science Book Grows ‘Two-Headed Shrimp’: The ‘Cracks’ in the Publishing Industry Should Be Patched”)