In January 2023, Michael Park, Erin Leahey and Russell Funk published a study in Nature with a gloomy headline finding. Looking across six decades of data on 45 million papers and 3.9 million patents, they reported that new work is “increasingly less likely to break with the past in ways that push science and technology in new directions” [1]. The pattern held across every broad field they looked at [1].

That is striking, given that the same paper describes “exponential growth in the volume of new scientific and technological knowledge” over recent decades [1]. It also drew serious pushback within months. My view sits between the two camps. I take the signal seriously, but I read it less as proof that science is running out of ideas and more as a question about what our system pays people to produce. I also think we should be honest that a number built from citations is not a census of all useful progress.

What the disruption score measures

The study leans on a measure called the CD index (short for consolidating versus disruptive). The idea is simple. If a paper truly changes a field, later researchers who cite it tend to stop citing the older work it built on, because that older work no longer matters much to them [1]. If a paper instead improves an existing line of research, later researchers are more likely to cite it together with its predecessors [1].

The score runs from minus 1 (fully consolidating) to plus 1 (fully disruptive), and the team measured it five years after publication [1]. Watson and Crick’s 1953 paper on the structure of DNA scored 0.62, while a Nobel-winning 1965 paper by Kohn and Sham, which used established theorems to build a method for calculating electron structure, scored minus 0.22 [1]. Both won Nobel Prizes. They changed science in different ways.

Both kinds of work are useful. The study is asking whether the balance between them has shifted, not whether one kind is good and the other bad.

How big the drop is

The decline is large. For papers, the average score fell between 1945 and 2010 by 91.9% in the social sciences (from 0.52 to 0.04) and by 100% in the physical sciences, with other fields in between [1]. The language of science shifted too. Early paper titles favoured verbs of creation and discovery, such as produce and determine, while later titles leaned toward verbs of improvement and application, such as improve and use [1].

One detail softens the gloom. Despite the huge growth in output, the number of papers and patents at the very disruptive end “remains nearly constant over time” [1]. So the big breakthroughs have not vanished. They are a shrinking share of a much larger pile.

Why the authors think it happened

Park and colleagues weighed some popular explanations. If fields were simply running out of easy discoveries, they argued, you would probably expect them to decline at different times and rates, yet the trends looked broadly similar across fields [1]. Nor did it look like a flood of weaker work: the downward trend persisted in papers from Nature, Science and the Proceedings of the National Academy of Sciences, and among Nobel-winning papers [1].

Their preferred explanation is narrowing. Over time, researchers increasingly cite the same previous work, cite their own work more, and cite older work, and these habits go along with lower disruption scores [1]. In their words, “a narrower scope of existing knowledge is informing contemporary discovery and invention” [1]. They add a sharp point: relying on “narrower slices of knowledge” helps individual careers, “but not scientific progress more generally” [1].

The strongest case against the metric

In June 2023, Alexander Petersen, Felber Arroyave and Fabio Pammolli posted a critique on arXiv as a preprint; a peer-reviewed version appeared in 2024 [2]. They argued that “the reported decrease in disruptiveness is an artifact of systematic shifts in the structure of citation networks unrelated to innovation system capacity” [2]. They call this citation inflation.

Their first mechanism is easy to picture. Reference lists have grown longer over the decades, which packs more links into the citation network and, they argue, drags the disruption score toward zero for reasons unrelated to innovation [2]. They also point to rising self-citation, and conclude that the index is “unsuitable for cross-temporal analysis” [2].

This is a fair hit, and Park and colleagues saw part of it coming. They wrote that as papers cite more previous work, the chance of a paper being cited on its own could fall for purely mechanical reasons [1]. They then tried adjusted versions of the score, statistical controls and simulations that randomly rewired the citation network, and still found a decline, with real papers falling further below what chance would predict over time [1]. They also called the CD index “a relatively new indicator” that needs more study [1].

I cannot referee this dispute, and a general reader does not need to. What both sides agree on is that the score is sensitive to how people cite, and citing habits are shaped by incentives. That alone should make us cautious about treating any single citation measure as the scoreboard for progress.

What counts as progress, and what we pay for

A citation-based score can miss a lot. A carefully built dataset, a replication that confirms a shaky result, a better quality check, or a tool that thousands of labs quietly use may all score as consolidating by this measure. Kohn and Sham’s paper is a reminder that consolidating work can be Nobel-worthy [1]. In my own work I build deep-learning models and web tools for reviewing medical images, and much of the valuable work in any field is careful improvement rather than a single breakthrough.

Still, I think Park and colleagues are pointing at something real about incentives. When hiring, promotion and grant decisions lean on paper counts, a steady stream of safe, incremental papers is the most reliable way to fill that count. Bold work tends to take longer, and it can fail. If we mostly reward volume, we should not be surprised when we get volume.

The authors’ own suggestions are modest and sensible. They propose that scholars be given time to read widely, that universities “forgo the focus on quantity, and more strongly reward research quality”, perhaps with year-long sabbaticals, and that funders back “riskier and longer-term individual awards that support careers and not simply specific projects” [1].

If you sit on a hiring or grant committee, try asking candidates to point to their two or three most important contributions instead of counting their papers. If you are a student, protect some time to read outside your narrow topic. Neither step needs a perfect metric, and both push in the direction the study’s authors recommend [1].

References

[1] M. Park, E. Leahey, and R. J. Funk, “Papers and patents are becoming less disruptive over time,” Nature, vol. 613, no. 7942, pp. 138-144, Jan. 2023, doi: 10.1038/s41586-022-05543-x.

[2] A. M. Petersen, F. Arroyave, and F. Pammolli, “The disruption index is biased by citation inflation,” arXiv:2306.01949, Jun. 2023, doi: 10.48550/arXiv.2306.01949. Later published in Quantitative Science Studies, vol. 5, no. 4, pp. 936-953, 2024, doi: 10.1162/qss_a_00333.