Twenty million scientific articles. Four million patents. Fifty years of intellectual production spanning dozens of disciplines. The study published in 2023 in Nature by Yiling Lin, Carl Benedikt Frey and Lingfei Wu is one of the largest ever conducted on the geography of innovation. Its verdict is clear: geographically dispersed teams systematically produce fewer breakthrough discoveries than colocated teams, regardless of field or period.

This result comes at the worst possible time for those advocating the generalization of remote work. But it deserves better than being weaponized as a management argument against remote work, or as a blanket petition for returning to the office five days a week. What it illuminates is something more precise and more useful: the conditions under which truly new ideas emerge, and what scientific and organizational policies risk losing if they ignore this mechanism.

The Essentials

  • An analysis of 20 million articles and 4 million patents over 50 years shows that distant teams are systematically less likely to produce disruptive innovations (Lin, Frey, Wu, Nature, 2023).
  • The penalty does not bear on technical execution: distant teams publish as much, file as many patents, and their work is cited as much for incremental contributions.
  • The deficit concentrates on so-called “disruptive” innovations, measured by a bibliometric indicator that detects work that breaks an established trajectory rather than extending it.
  • The penalty tends to attenuate for teams whose members share a common history of in-person work before collaborating remotely.
  • The issue directly affects research center recruitment policies, the geography of European innovation, and team organization in the age of AI.

What Fifty Years of Dispersed Science Reveals

The Lin, Frey and Wu study rests on a measurement method developed in innovation economics over the past fifteen years: the CD index, or “disruption index.” The idea is simple. An article or patent is disruptive if it cuts other researchers off from prior literature: after it, people cite less of what existed before. It is consolidating if it builds on and strengthens the existing base. This is not a value judgment—incremental science is necessary—but a measure of the function a work plays in a field’s evolution.

On this indicator, the result is robust: the further geographically separated team members are at the time of collaboration, the less their production tends to be disruptive. The effect holds in sciences, engineering, and social sciences. It holds in the 1970s as in the 2010s. It is not attributable to author quality, team size, or disciplinary differences between collaborative and non-collaborative domains.

What the researchers then document is even more precise: distant teams find, codify, validate. They produce results. Their publication rate is not lower. Their articles are cited. But the citations they receive bear more on confirming and extending what exists than on opening new directions. This is the operational explanation that Frey and coauthors advance: geographic proximity fosters informal, unscheduled, often unconscious exchanges that bring disparate information into contact and generate novel combinations. The hallway, the cafeteria, the conversation that starts on one thing and shifts direction: this is not sentiment, it is a documented mechanism.

Informal Information Does Not Synchronize

This conclusion aligns with an older body of research on what innovation economists call “localized knowledge spillovers.” Geographic proximity transmits information that is not codified, not conscious, often impossible to put in an email or shared document. Alfred Marshall observed this phenomenon in nineteenth-century English industrial districts without formalizing it thus: “the secrets of the industry are in the air.” The Nature study provides the contemporary, quantified version.

The distinction between codifiable knowledge and tacit knowledge is at the heart of Carl Benedikt Frey’s thinking on innovation. In his earlier work, notably The Technology Trap (2019), Frey showed how major technological transitions encounter resistance precisely because they displace tacit skills that workers cannot easily transfer to new jobs. The same logic applies here in mirror: the tacit knowledge that feeds breakthroughs does not transfer easily through digital tools either. Zoom transmits words. It does not transmit peripheral context, weak signals, the reformulation that occurs when two people look at the same whiteboard.

The problem is not technical—it is structural. Remote collaboration tools have become highly efficient for defined tasks. They allow working together on code, text, a presentation. They allow coordinating, validating, delivering. Where they get stuck is on what no one has yet defined: the unplanned conversation that leads to a question that did not exist before being asked. It is precisely in this space, according to the study, that the colocation premium lodges.

The Nuance That Changes Everything for HR Policy

The study provides a precision that considerably modifies practical implications. The distance penalty attenuates, to a notable degree, for teams whose members share a common history of in-person work. Researchers who have worked together in person, who share implicit language, established trust, retain part of their capacity to produce breakthroughs even at a distance.

This is not a small nuance. This is a different mechanism. The tacit knowledge that circulates in physical proximity can, to some extent, be accumulated and stored in the relationship. It then becomes partially mobilizable at a distance. What colocation generates continuously, a preexisting relationship can partially compensate for, at least for a time.

For organizations, this opens a path. Not a return to the office as dogma, nor generalized remote work as comfort, but differentiated management according to work phase and the maturity of team relationships. Exploration and design phases—where breakthroughs are born—benefit from proximity. Execution, verification, documentation phases—where technical work happens—tolerate distance much better. And established teams can maintain part of their disruptive potential at a distance, provided the relationship was first built in person.

This is an organizational architecture that few enterprises or research institutions have formalized. Most have adopted uniform policies, either by managerial ideology (the office as proof of commitment) or by union or social pressure (remote work as an acquired right). The data suggest that neither of these logics corresponds to the problem’s actual structure.

What This Result Says About European Innovation

The issue extends far beyond the debate on work modes. It directly concerns the geography of innovation and scientific policies, particularly in Europe.

The European Union has massively invested in transnational scientific collaboration over the past thirty years. Horizon Europe, the research framework program, distributes billions of euros each year to consortiums bringing together teams in Berlin, Lisbon, Warsaw and Amsterdam on the same project. The logic is clear: break national silos, create critical mass, avoid duplication of effort. This is good logic for incremental science. It may be less suited to producing breakthroughs.

This is not an argument against international collaboration—the study does not say that homogeneous or national teams perform better. It says that physical proximity, regardless of team composition, is a factor in breakthrough. The answer is not withdrawal; it is the design of meeting spaces. Joint laboratories, researcher residencies, multi-pole campuses where international teams meet physically for defined phases: these devices exist, but remain marginal in European scientific policies dominated by the virtual consortium logic.

Europe is already struggling to produce technological breakthroughs at the scale of its investments. As an analysis of European innovation dynamics showed, the continent has not produced a technology giant in fifty years—not for lack of spending, but for lack of creative destruction, risk-taking, breakthrough. If the geographic organization of scientific work contributes to this deficit, it is a lever that policies can activate, provided they name it clearly.

AI Changes the Parameters, Not the Fundamental Constraint

A counter-argument regularly surfaces in this debate: artificial intelligence will change the game. If language models can synthesize corpora, detect connections between fields, suggest combinations no one had envisioned, then perhaps the informal whiteboard conversation becomes less determinative.

The argument deserves serious consideration, but it conflates two things. AI excels at traversing a known space of possibilities and finding optima in that space. It is less comfortable delimiting a new space of possibilities, posing a question no one has yet formulated. It is precisely this last capacity that Frey and coauthors associate with physical proximity: not execution in a defined space, but redefinition of the space itself.

Whether AI can reproduce the serendipity of laboratory hallways remains open, and it would be an error to settle the question too quickly in either direction. What is documented today is that existing digital tools—far more mature than generative AI—have not eliminated the colocation premium on breakthroughs. Nothing in current literature suggests generative AI achieves this in the near term. It probably amplifies incremental productivity, enriches the execution and documentation phase. It does not replace, for now, the informal contact that generates truly new questions.

The deeper question is this: if AI takes over an increasing share of technical execution work, as several recent studies on work organization in the age of AI suggest, then the relative value of design work increases. And if colocation is a determining factor in the quality of this design work, then the premium of physical proximity does not decrease with automation—it increases.

A Political Economy of Return to the Office That Must Be Built

Daron Acemoglu, whose work on technology and power usefully complements Frey on this terrain, has insisted on a question often sidestepped: who captures the gains of progress, and who decides how it is organized? The same question applies here. The remote work debate is today dominated by two logics: the employee’s right to choose their work location, and the employer’s short-term interest in measurable individual productivity. Neither addresses the question of breakthrough collective innovation, which is a common good—not only for the company concerned, but for the economy as a whole.

The results of the Nature study are produced by economists from Oxford and Pittsburgh, not by CEOs seeking to justify return to the office. They describe a real mechanism, with implications that organizations and public policies can integrate without adopting uniform or ideological positions.

The concrete question is not “how many days in the office,” but: which work phases require proximity, which teams need to build a relationship in person first before working remotely, and how do research institutions and enterprises design their meeting spaces to maximize the chances of breakthrough? These are questions of organization, not work philosophy. And the data now exist to treat them as such.

What the study does not say, however, is how to reconcile this constraint with the reality of labor markets where talent is globally dispersed. Forcing a team to colocate may mean not recruiting some of the best available researchers. This is a real tension, without universal answer. It suggests at minimum that colocation should be conceived as a targeted investment—residencies, in-person sprints, co-located design phases—rather than as a permanent rule or absence of rule.

Research centers that begin piloting these hybrid architectures—intensive colocation in exploratory phase, distributed work in execution phase—will offer in a few years the natural data to test whether this intuition holds at scale. Until then, the Lin, Frey and Wu study provides a solid empirical foundation that scientific policies and human resources directors would be wrong to ignore in the name of comfort or vested rights.

Sources

  1. Lin Y., Frey C.B., Wu L. — “Remote collaboration fuses fewer breakthrough ideas”, Nature, 2023: https://www.nature.com/articles/s41586-023-06767-1
  2. Frey C.B. — The Technology Trap: Capital, Labor, and Power in the Age of Automation, Princeton University Press, 2019
  3. Funk R.J., Owen-Smith J. — “A dynamic network measure of technological change”, Management Science, 2017 (CD index, method for measuring disruption)
  4. Acemoglu D., Johnson S. — Power and Progress: Our Thousand-Year Struggle over Technology and Prosperity, PublicAffairs, 2023
  5. Marshall A. — Principles of Economics, 1890 (localized spillovers, industrial districts)
  6. Horizon Europe Programme, European Commission — financing data for transnational collaborative research: https://research-and-innovation.ec.europa.eu/funding/funding-opportunities/funding-programmes-and-open-calls/horizon-europe_en
  7. Lin Y., Frey C.B., Wu L. — Full arXiv preprint: https://arxiv.org/pdf/2206.01878
  8. Official University of Oxford statement, November 30, 2023: https://www.ox.ac.uk/news/2023-11-30-remote-collaborations-deliver-fewer-scientific-breakthroughs-oxford-co-led-research
  9. NSF Public Access Repository: https://par.nsf.gov/biblio/10477205-remote-collaboration-fuses-fewer-breakthrough-ideas
  10. Oxford Research Archive (ORA): https://ora.ox.ac.uk/objects/uuid:e7bbc792-25da-4caf-9c7d-fc0684f7dcf5