In 2023, Nature published an spectacularly ambitious study of American researchers based on 45 million papers and 3.9 million patents claiming that science and technology had become steadily less disruptive. The paper itself became something of a scientific blockbuster, attracting more than 400,000 accesses on Nature’s website and already accumulating almost 800 Scopus citations. The conclusion travelled rapidly through academia and the media because it fitted an irresistible narrative, modern science had supposedly become bloated, cautious and increasingly incapable of producing genuine intellectual revolutions.
At the time, I expressed several reservations about such a sweeping conclusion. Among other things, I wondered whether changes in research-team size could partly explain the result and whether the conclusion would survive the use of alternative metrics designed specifically to identify revolutionary discoveries. Two years later, that confidence already looked premature. A 2025 study in Nature Computational Science that i mentioned on this blog introduced the concept of persistent disruption and showed that papers capable not merely of disrupting previous work, but of imposing a lasting new intellectual trajectory, had become more prominent in recent decades. The supposedly simple story of scientific decline was already beginning to fracture.
Now the original claim faces a much more damaging challenge. A new Nature analysis by researchers at Vrije Universiteit Amsterdam and KU Leuven, highlighted by The Economist, argues that a substantial share of the celebrated decline may have been generated not by any historical collapse in scientific creativity, but by artefacts in the bibliographic data themselves. Older papers are disproportionately affected by missing reference information, and those deficiencies can artificially inflate disruption scores, making early science look more revolutionary and contemporary science more pedestrian than it really is. Once those distortions are corrected, the alleged decline is dramatically reduced and, in some datasets, almost disappears. If that result survives scrutiny, then the problem is no longer merely that the original interpretation may have been exaggerated. It is that an extraordinarily influential claim about the supposed intellectual exhaustion of modern science may have been built, to a remarkable extent, on imperfections in the very machinery used to measure it.
