Artificial Intelligence in Scientific Publishing: A Tool to Be Governed, Not a Substitute for Scientific Judgment

Technological innovation has rarely entered human society without resistance.Some of the concerns that accompanied technologies now regarded as indispensable may appear almost incomprehensible today, although they were taken very seriously at the time.The Industrial Revolution provides perhaps the best-known example.In early nineteenth-century England, the Luddites destroyed textile machinery that they perceived as threatening their jobs, skills, and livelihoods.Their reaction was not simply an irrational fear of machines: it also reflected legitimate concerns about profound changes in working conditions and the distribution of the benefits of technological progress.Nevertheless, mechanization continued and eventually transformed manufacturing and society.Other innovations followed similar paths.Railways were initially regarded by some as potentially dangerous to human health.The automobile encountered resistance and restrictive legislation: the British Locomotive Act of 1865 famously required, under certain circumstances, a person carrying a red flag to precede a self-propelled vehicle.Electricity, elevators, vaccination and, much more recently, the Internet all encountered scepticism, resistance or fear before becoming part of everyday life.History does not teach us that every technological innovation is necessarily beneficial.There have been technologies whose harmful consequences became apparent only after widespread adoption.Nor should legitimate concerns about technological change be dismissed as another form of Luddism.History does, however, suggest that genuinely useful technologies tend eventually to become incorporated into ordinary life and professional practice.Artificial intelligence (AI), and particularly generative AI based on large language models (LLMs), appears likely to follow this pattern.The scientific community is now debating whether and how AI should be used in research, scientific writing, peer review and editorial decision-making.Positions range from enthusiasm to considerable concern.Both attitudes are understandable.AI systems can perform tasks that only a few years ago would have seemed extraordinary.At the same time, their limitations-including inaccurate information, fabricated references, bias, lack of transparency and the generation of extremely convincing but incorrect statements-are equally real [1][2][3][4].The question, therefore, should probably no longer be whether scientists will use AI.They already do, and its use will almost certainly increase.A more useful question is: what should we allow AI to do for us, and what must we never delegate to it? Assistance and Delegation Are Not the Same ThingThis distinction, in our view, is fundamental.AI can be an extraordinarily useful scientific assistant.Consider the preparation of a manuscript.AI can identify grammatical and spelling errors, improve the readability of English, suggest more concise formulations, identify repetitions, detect inconsistencies between different sections of a manuscript and help organize complex material.It can assist in searching the literature and in identifying potentially relevant references, provided that these references are subsequently checked against the original sources.