On generative artificial intelligence in scientific writing

Scientific communication - the dissemination of ideas, hypotheses, and discoveries - is fundamental to advancing human knowledge and to ensuring public access to new knowledge as it is generated. Growth in the use of generative artificial intelligence (AI) in scientific communication has been explosive, likely driven by high uptake among an adept userbase with demanding schedules, making the efficiency gains attractive. The rapid uptake of AI is creating disruptive changes in the way scientific communication happens, affecting the methods by which abstracts, manuscripts, figures, presentations, and funding proposals are generated, and in many cases impacting the final product. In this perspective piece, we reflect on the rapidly changing landscape of AI use in scientific communication with attention to the potential immediate and long-term risks. The inherent unauthorized secondary use of the data used to train AI models raises questions about credit and plagiarism in academia. Fabricated references have proliferated in the scientific literature resulting from "hallucinations" by AI models. Meanwhile, each use of an AI model comes with an incremental risk to the environment at an already perilous time. More abstractly, AI use in scientific communication risks the process itself becoming iterative, prioritizing summation while diminishing insights that fuel the scientific engine. Finally, the friction that AI models remove from writing may risk removing an essential part of the process of scientific thinking, particularly for the next generation of scientists. While we acknowledge that AI is here to stay, we feel an element of caution is warranted in its implementation, to help preserve science and its dissemination as a fundamentally human endeavor.