Until recently, research rhythm followed a broadly linear path: read the literature, form a hypothesis, design the experiment, publish the results. Creativity often lived at the margins, in half-written notes or sporadic conversations, while most thinking happened in isolation. Science was not always organized this way. The first academies of science, such as the Royal Society, thrived on correspondence and disputation, and knowledge emerged through dialog and collective reasoning. Over time, that conversational culture gave way to specialization, formal reporting, and the economy of publication. Large language models (LLMs) now may be reopening that space for dialog, although the interlocutor is nonhuman.
Recently, Anna Viktorovna Gavrilova from the Department of Biosciences at the University of Milan and Carlo Galli from the Department of Medicine and Surgery, Histology and Embryology Lab at the University of Parma published a commentary article “Conversing with machines: How AI is changing the way scientists think” in Quantitative Biology. The authors argue that large language models (LLMs) have turned computation into dialog, an epistemic reconfiguration rather than a gain in convenience. Reasoning becomes dialogical, evolving through feedback between representation and interpretation, and LLMs work as mediating artifacts that shape what can be inferred and how hypotheses are articulated.
Within quantitative work, each exchange builds on the previous one, so hypothesis refinement and interpretation can happen inside a single dialog. Modeling has always been cyclical, but most of its reasoning stayed implicit. The conversational layer makes part of it visible and linguistically documented: responses can be interrogated, revised, or traced, and logs stored alongside the data and code they helped produce, an “epistemic provenance” analogous to version control.
The authors also ask who takes part. Tasks once requiring scripting or bioinformatics support can now start through dialog, a step toward epistemic equity for underfunded institutions, though linguistic precision becomes the new epistemic threshold. Because models reproduce biases in their training data, “inclusive AI” risks becoming a new mechanism of exclusion. Creativity is the second front: trained across fields and genres, LLMs act as agents of analogy, retrieving forgotten conceptual bridges and reintroducing serendipity into a risk-averse research culture.
The final section is cautionary. Fluency can deceive: models generate plausible text rather than verified truth, so fabricated citations or stylistic precision can create an illusion of rigor. The deeper threat is habituation, where reflection itself is outsourced, a cognitive offloading that concerns reasoning rather than memory. A practical response is to treat dialog as part of the scientific record. The commentary closes on transparency, critical literacy, and humility: “To converse with AI is to participate in an experiment in co-thinking,” the authors write, and the human remains the ethical and epistemic anchor.
DOI
10.1002/qub2.70032