The “creeping normality” of large language models is quietly reshaping the life sciences
Large language models (LLMs) are gradually transforming research in the life sciences in ways that extend far beyond improving productivity, and becoming a new normal before scientists have agreed where the limits of its use should be.
A new study (https://doi.org/10.1002/fee.70059), by an international team of scientists, describes this phenomenon as the "creeping normality" of generative artificial intelligence (AI), a process in which profound changes become accepted because they occur incrementally through small, subtle steps. “While generative AI has rapidly become part of the everyday workflows of many researchers, its long-term effects on almost all aspects of science have received comparatively little attention”, says Ivan Jarić, researcher from the Biology Centre of the Czech Academy of Sciences, and lead author of the study. “Routine reliance on these tools could fundamentally reshape the foundations of scientific practice and culture”, he adds.
The article identifies several domains where such changes are already emerging. One is scientific collaboration. Researchers increasingly use LLMs for support that was conventionally provided by colleagues and informal professional networks, like brainstorming ideas, troubleshooting analyses, and seeking expert feedback. An increasing reliance on LLMs over human collaboration is expected to lead to reduced motivation to seek external expertise from other research fields or regions, or on a particular taxonomic group or methodology, which may in turn weaken cross-disciplinary interactions, reduce exposure to diverse perspectives, and encourage more siloed and less innovative science.
AI use may also fundamentally change how scientists seek information and generate ideas. “As LLMs increasingly replace search engines, encyclopaedias, and even literature searches, researchers are going to become more dependent on AI-generated feedback, limited only to the questions one thinks of asking”, explains Susan Canavan from the University of Galway, another author of the study. “This could unintentionally reinforce confirmation bias, create intellectual echo chambers, and contribute to greater uniformity and lower originality in scientific language, approaches, and creative thinking.”
Image. The proposed appropriateness of AI controls at different stages of the scientific process.
Beyond cognition, the article explores the consequences for scientific training and career paths. While lowering barriers for researchers who are early in their careers or lack access to coding or statistical expertise, the use of LLMs can generate new types of inequalities and reinforce existing ones. It will also intensify deskilling in core competencies, with some of the fundamental skills in the life sciences being outsourced to LLMs, such as literature searching and synthesis, natural-history reading and knowledge acquisition, taxonomic judgement, coding and debugging, and statistical reasoning. The authors suggest that it could even influence hiring decisions and career opportunities by reducing the need and motivation to open new PhD and postdoctoral positions, as various types of work will be increasingly delegated to LLMs.
“Rather than advocating for or against AI, we call for the life sciences community to establish clear boundaries for its appropriate use”, suggests Michael Bertram from the Swedish University of Agricultural Sciences and Stockholm University, another author of the study. “LLMs should, for example, be widely embraced for routine tasks such as proofreading, language editing, and workflow annotation, and, if underpinned by proper human control, also for advanced tasks like literature synthesis and data extraction. However, AI use should be undesirable in activities requiring independent scientific judgement, including research prioritization, peer review, funding decisions, and ethical considerations.”
The authors conclude that the transformation of science by generative AI is already underway. The urgent question is not whether AI will become part of scientific research, but where the scientific community chooses to draw the line. Defining those boundaries now, they argue, will be essential for preserving creativity, diversity, accountability, and human judgement as core scientific values in the life sciences.
For more detailed information, see the article published in Frontiers in Ecology and the Environment:
Jarić, I., Pipek, P., Canavan, S., Firth, J.A. and Bertram, M.G. (2026). The creeping normality of AI in the life sciences. Frontiers in Ecology and the Environment https://doi.org/10.1002/fee.70059
Contact: Ivan Jarić, ivan.jaric@hbu.cas.cz
