Original title: Normative boundaries of AI in scientific work: Evidence from PhD researchers
Francesco Angelini, Johan Lyrvall · August 2026
Artificial intelligence is entering scientific research, but scientists do not view its role the same way for every task. To see how researchers feel about this, a team studied survey answers from thousands of graduate students across science, engineering, and medicine. The researchers looked at how comfortable students felt using artificial intelligence to write papers, collect and analyse data, design experiments, track existing studies, and summarise reading material. They found that feelings about artificial intelligence do not fall into simple like or dislike camps. Instead, most students draw clear lines. Most accept artificial intelligence for literature chores like tracking and summarising past studies. However, they push back against letting artificial intelligence handle core intellectual work like writing, analysing data, or designing experiments. This shows that rules for research tools must fit specific tasks.
Imagine a family deciding who cooks dinner. Everyone is happy for one person to chop vegetables, but the moment someone else tastes the soup and adds salt, feelings shift. The same activity—cooking—is fine when it is routine prep but sensitive when it touches taste and credit. Here, researchers accept automated help for tracking and summarizing papers, the equivalent of chopping vegetables, but resist it for writing and designing experiments, the equivalent of deciding the flavour. Comfort depends on which part of the work is handed over.
Universities, funders, and journals are deciding where AI belongs in research. If PhD students—the people doing much of the work—draw the line at writing and data analysis while accepting AI for literature, then blanket policies will miss the real tension. Governance and training should be task-specific rather than all-or-nothing.
The sample is self-selected and limited to PhD students in STEM and medical and health sciences, so findings may not generalize to other researchers or fields. The survey measures comfort, not actual use or legitimacy, so the profiles describe attitudes, not whether AI should be used. The study does not establish consequences of these attitudes for research quality or careers.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Normative boundaries of AI in scientific work: Evidence from PhD researchers — Francesco Angelini, Johan Lyrvall
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.