AI has stopped being a lab tool and become the engine of discovery, a map of 128 research fronts across 11 fields finds

Pathologists now let software flag suspicious slides; farmers let an app name a diseased leaf. Both belong to the same shift, according to a global map of research published in The Innovation: artificial intelligence has stopped being a specialist's add-on and become the engine of scientific work.

The map is the 2025 edition of Research Fronts, an annual series the Chinese Academy of Sciences has produced with Clarivate since 2014. Its unit is the "research front": a cluster of heavily co-cited papers, defined here as work in the top 1% of citations for its field and year, plus the papers that later cite them. Because the clusters are drawn from citation behaviour rather than editorial labels, they track what working scientists actually build on.

Starting from 13,830 fronts in Clarivate's Essential Science Indicators for 2019–2024, the team kept 110 hot fronts, ten for each of 11 broad fields, and added 18 emerging ones whose core papers average 2023 or later. Thirty-one were then read in depth by subject, policy and strategy experts.

The headline finding is about AI. It appears in all 11 fields, and not as a single front among many. In mathematics the hot list is dominated by physics-informed neural networks and neural operators — machine learning welded to differential equations. In biology, AI-driven prediction and design of biomolecular complexes draws on AlphaFold and successors. In clinical medicine, five of the seven emerging fronts are AI applications, four of them systematic appraisals of ChatGPT in diagnosis, radiology, ophthalmology and nursing education. In agriculture, deep-learning detection of plant pests was an emerging front in 2024 and has now graduated to the hot list — a technology reaching mainstream practice.

Other fronts mark different kinds of momentum: the planetary boundaries framework in ecology, high-throughput single-cell methods in biology, high-temperature polymer dielectrics and halide solid electrolytes for all-solid-state batteries in materials, petawatt-class lasers and high-temperature superconductivity in physics, dark matter detection and gravitational theory in astronomy, 6G semantic communication in information science, and generative AI risk management in the social sciences.

The authors are careful about what the list can bear: a snapshot of "only a fraction" of a landscape containing tens of thousands of fronts, chosen by citation counts that lag discovery and by expert judgement at the margins. What they claim is narrower and defensible: that organized basic research — sky surveys, national "AI for science" programmes, large instruments — is now a major source of breakthroughs, and that AI's spillover between fields deserves study in its own right rather than as a per-discipline footnote.

The full reports were released in Chinese and English in December 2025, and the paper is aimed at the people who allocate money and instruments. Its value is less a prediction than a mirror: it shows where much of the world's most-cited science is now accumulating.