From Origins to Frontier: A History of the Evolution of Scientific Institutions

How a 1660 London coffeehouse society, a German chemistry classroom, a 1945 wartime report, and a 1991 physics mailing list built the machinery that now decides what counts as known.

Before there was a journal, there was a table: the Royal Society's council reviewing a fellow's account of an experiment, 1660s. — Image prompt and art direction by Brecht Corbeel; generation pending.

Abstract

Scientific knowledge is not only produced by ideas; it is produced by institutions that decide what gets measured, funded, published, checked and specialized. This article traces four hinge points in that institutional history — the Royal Society's founding in 1660 and its Philosophical Transactions, the German university laboratory model of the nineteenth century, the post-1945 creation of large-scale government science funding culminating in the U.S. National Science Foundation, and the emergence of open-access and preprint publishing from the 1990s onward — and asks what each transition changed about how claims become accepted knowledge. It separates documented fact from vendor and advocacy claims, and closes with conditional, falsifiable expectations about where institutional science is headed next: open peer review, registered reports, and replication as a funded activity rather than an afterthought.

Science is often narrated as a history of ideas: a heliocentric model displaces a geocentric one, a germ theory displaces a miasma theory, a plate-tectonic model displaces a static-crust one. That narration is not wrong, but it leaves out the machinery that makes an idea checkable in the first place — the institutions that decide who gets to observe, who gets to challenge an observation, who pays for the equipment, and who gets to call a claim settled enough to build on. This is a history of that machinery: four hinge points, roughly a century apart, each of which changed what “doing science” institutionally required.

The four hinges are the founding of the Royal Society of London in 1660 and its 1665 journal Philosophical Transactions [1] [2]; the rise of the German university research laboratory in the nineteenth century, exemplified by Justus von Liebig’s teaching laboratory at Giessen [5]; the postwar creation of large-scale government science funding, crystallized in Vannevar Bush’s 1945 report Science: The Endless Frontier and the 1950 founding of the U.S. National Science Foundation [4] [3]; and the rise of preprint servers and open-access journals from 1991 onward, epitomized by arXiv and PLOS ONE [7] [8]. Threaded through all four is a fifth, quieter institutional story: the slow, incomplete, and still-contested build-out of replication as something anyone actually does, culminating in the 2015 finding that only 36 of 97 psychology results replicated at statistical significance [6] [9].

None of these four moments “caused” modern science on its own. Each solved a specific institutional bottleneck of its era, created a durable structure around that solution, and left a residue of problems that the next era’s institutions had to address. That pattern — bottleneck, structure, residue — is the throughline of this piece.

1660: a society before there was a journal

The Royal Society formed from informal gatherings of London- and Oxford-based natural philosophers who had been meeting since the 1640s; the first recorded meeting of what became the Society took place on 28 November 1660 after a lecture at Gresham College, and the group received its royal charter in 1662 [1]. Its motto, Nullius in verba — take nobody’s word for it — states the institutional problem it was built to solve: in the mid-seventeenth century there was no reliable mechanism for one natural philosopher to check another’s reported observation. A claim that a vacuum pump produced a given effect, or that a comet followed a given path, lived or died on private correspondence, personal reputation, and whoever happened to be in the room.

The Society’s answer was to make natural knowledge a collective, institutional activity rather than a private one: fellows brought instruments and specimens to meetings, performed experiments before witnesses, and recorded the proceedings. The corresponding secretary, Henry Oldenburg, began compiling members’ letters and experimental reports into a periodical, Philosophical Transactions, starting in 1665 — usually credited as the first (and still longest continuously running) scientific journal [2] [1]. This is a fact, not an inference from later prestige: Oldenburg’s stated purpose was to circulate accounts of experiments and observations to a wider audience than personal correspondence could reach, converting private letters into a dated, citable, and comparable public record.

What Philosophical Transactions did not yet have, in 1665, was peer review in any form a modern reader would recognize. According to the Society’s own historical account, formal external refereeing of submitted papers only began in 1832, when the Committee of Papers started sending manuscripts to referees rather than relying on the editor’s or council’s own judgment [1]. For roughly a century and a half, then, the journal functioned on editorial judgment and reputation, not systematic external review — an important corrective to any story that treats peer review as a single ancient invention rather than a slow accretion of practice across two hundred years. The word “peer review” itself is a twentieth-century coinage applied retroactively to a set of practices that had been evolving since Oldenburg’s time.

The residue this era left for later institutions to solve: publication answered the problem of circulating a claim, but it did not by itself answer who should be trusted to fund the work behind the claim, how to train the next generation of observers at scale, or how a claim gets checked rather than merely printed. Those became the next three centuries’ problems.

Figure 1. Henry Oldenburg's Philosophical Transactions, begun in 1665, turned private letters into a dated, citable public record [@philtrans-standrews-history]. — Image prompt and art direction by Brecht Corbeel; generation pending.

The nineteenth century: the laboratory becomes a classroom

For roughly a century and a half after the Royal Society’s founding, the production of new natural knowledge in Europe remained largely the work of individual gentleman-scholars, wealthy amateurs, or small private circles — a research and training bottleneck that limited how many people could be taught to do original experimental work, and how fast. The bottleneck broke in Germany, and the clearest documented case is Justus von Liebig’s teaching laboratory at the University of Giessen, which he built starting in the 1820s after studying with Joseph Louis Gay-Lussac in Paris [5].

Liebig’s specific institutional innovation, as described by the Science History Institute, was to reproduce at scale the kind of hands-on laboratory apprenticeship he had experienced under Gay-Lussac, but “for many more students at a time” — building what amounted to a model laboratory for training graduate students in organic and analytical chemistry within the walls of a teaching university, rather than in a private cabinet or an isolated workshop [5]. Research and instruction, previously separate activities performed by separate kinds of people, were fused into a single room with ranked benches, shared apparatus, and a common set of problems students worked through together under supervision. Liebig also took over and edited a chemistry journal, Annalen der Chemie und Pharmacie, which became the field’s leading publication within a few years, coupling his teaching laboratory to a dedicated outlet for the results it produced [5].

The Science History Institute’s account is explicit that this model “was widely imitated in Europe and later on in the United States” [5] — a fact about diffusion, not a claim about Liebig personally inventing the university laboratory from nothing. The broader “German model” of combining research and graduate training inside the university, of which Giessen is one of the best-documented instances, became the template that American research universities explicitly copied when they reorganized around graduate schools and the PhD in the second half of the nineteenth century.

What this era changed institutionally: it converted the researcher’s laboratory from a private workshop into a reproducible pedagogical unit — a place that could train a chemist and simultaneously produce chemistry, at a throughput no single master-apprentice relationship could match. The residue it left: it scaled training and output, but it did almost nothing to change how such laboratories were paid for. Nineteenth-century laboratory science, in Germany as elsewhere, still depended overwhelmingly on university appointments, private patronage, and industrial sponsorship rather than any systematic public funding mechanism aimed at basic research for its own sake.

Figure 2. Justus von Liebig's teaching laboratory at Giessen trained many students at once at the bench, fusing research and instruction into one room [@sciencehistory-liebig-wohler]. — Image prompt and art direction by Brecht Corbeel; generation pending.

1945–1950: the state becomes the funder of basic research

Before the Second World War, American scientific research institutions depended primarily on philanthropic endowments and private industrial funding, an arrangement that tended to steer inquiry toward questions aligned with a funder’s existing interests rather than toward curiosity-driven basic research [3]. The wartime mobilization of scientists into projects like radar and the Manhattan Project demonstrated, to the U.S. federal government’s own satisfaction, that large coordinated science produced results no individual patron could have bought.

In November 1944, President Franklin D. Roosevelt asked Vannevar Bush, director of the wartime Office of Scientific Research and Development, to propose how the government could carry that wartime capacity for organized research into peacetime. Bush’s answer, delivered in July 1945 as the report Science: The Endless Frontier, recommended establishing a federal foundation to fund basic scientific research and the training of scientific personnel [4] [3]. This is a primary-source-verifiable fact: the report itself is archived by NSF and is explicit in its recommendation.

It took five more years of legislative negotiation — chiefly over how much control scientists themselves versus political appointees would have over grant decisions — before Bush’s proposal became law. President Harry S. Truman signed the National Science Foundation Act of 1950 (Public Law 81-507) on May 10, 1950, creating the NSF as an independent federal agency to fund basic research and science education across disciplines [3]. The institutional innovation here is specific and separable from Bush’s report: peer-reviewed, competitive, discipline-spanning federal grant-making became a standing government function rather than a wartime emergency measure, and it created a funding architecture — proposal, external review panel, program officer, award — that other governments (and, eventually, other U.S. agencies such as NIH) would adapt.

It is worth being precise about what is fact here and what is a later interpretive framing. That Bush wrote the report, that it recommended a funding foundation, and that NSF was created in 1950 are documented facts drawn from NSF’s own institutional history and the primary report text [4] [3]. The claim sometimes made — that Bush’s report represents a single clean philosophical vision of “basic research as the wellspring of all applied progress” that has guided science policy unchanged for eighty years — is a much more contested historical and policy analysis, and this article does not adopt it; the report’s own recommendations were substantially modified by Congress before enactment, and science funding since 1950 has never been governed by Bush’s report alone.

What this era changed: research funding became a routine government responsibility, decoupled (in principle) from a specific patron’s commercial or political interest, judged instead by a rotating body of expert peer reviewers. The residue: as government funding scaled up the volume of research being produced, it also scaled up the pressure to publish that research quickly and often, which — as the next section documents — created new incentives that peer review at the publication stage was not originally built to resist.

Figure 3. Vannevar Bush's 1945 report proposed the federal funding architecture that became the National Science Foundation in 1950 [@bush-endless-frontier] [@nsf-history]. — Image prompt and art direction by Brecht Corbeel; generation pending.

What scaling exposed: the replication problem

None of the three institutional innovations above — a learned society’s journal, a teaching laboratory, or a competitive grant system — was designed to systematically check, after publication, whether a claimed result actually holds up when someone else tries to reproduce it. For most of the twentieth century, replication was assumed to happen informally, as a side effect of other scientists building on a result, rather than funded or organized as a distinct institutional activity.

The clearest documented test of that assumption is the Open Science Collaboration’s 2015 Reproducibility Project in psychology, published in Science. A large, coordinated group of researchers attempted to replicate 100 studies sampled from three major psychology journals. Of the 97 original studies that had reported a statistically significant effect, only 36 replications reproduced a statistically significant effect in the same direction — a raw success rate of roughly 37 percent — and even among the studies rated as successfully replicated, effect sizes in the replication were on average about half the magnitude reported in the original study [6] [9]. The Center for Open Science, which coordinated the project, attributed the gap substantially to structural incentives: publication and career advancement reward novel, statistically significant findings far more than they reward confirmatory replication work, so few researchers had strong professional reasons to attempt replications before this coordinated project made it a funded, credited activity in its own right [9].

It is important to separate what this result actually shows from stronger claims sometimes layered onto it. The 2015 finding is a documented empirical result about one field (psychology), using one method (direct replication attempts on a specific journal sample), reported in a peer-reviewed venue. It is not, by itself, a general finding that “most of science is false,” a claim the study’s own authors did not make and that would require separate, field-by-field replication programs to establish. What the result does support, as an institutional diagnosis, is narrower and better evidenced: publication and funding institutions built primarily to reward novelty had not, by the early 2010s, built comparably strong institutional mechanisms to reward and fund confirmation. That is a gap in institutional design, not a verdict on any individual finding.

Figure 4. The 2015 Reproducibility Project found that only 36 of 97 original psychology results replicated at statistical significance [@osc-2015-reproducibility] [@cos-osc-blog]. — Image prompt and art direction by Brecht Corbeel; generation pending.

1991–2006: circulation and review move outside the subscription journal

The fourth hinge addressed a different bottleneck than replication: the speed and cost of circulating results at all, and who gets to gatekeep that circulation. In 1991, physicist Paul Ginsparg created what became arXiv, an electronic repository allowing researchers to post manuscripts — initially in high-energy physics, later across physics, mathematics, computer science and other fields — for free public access before or independent of formal peer-reviewed publication [7]. arXiv’s own description is explicit that submissions are moderated for topical relevance and scholarly form but are “not peer-reviewed by arXiv” itself; responsibility for the content rests with the authors [7]. By the mid-2020s the archive hosted more than three million scholarly articles across eight subject areas [7]. The institutional shift here is that a claim could reach the entire research community, dated and citable, before any journal’s editorial process had rendered a verdict on it — inverting the sequence that had held since Oldenburg’s 1665 model of “review, then circulate.”

A parallel but distinct shift happened on the peer-reviewed side of publishing. The Public Library of Science, founded in 2000 by Nobel laureate Harold Varmus, biochemist Patrick Brown, and computational biologist Michael Eisen, launched PLOS ONE in December 2006 as an open-access, author-funded journal with a deliberately unusual editorial criterion: reviewers were instructed to judge methodological soundness, not to predict a paper’s ultimate importance, leaving that judgment to the community after publication rather than to editors and referees before it [8]. This “mega-journal” model — broad-scope, rigor-focused rather than novelty-focused review, funded by author publication fees rather than reader subscriptions — was, by the encyclopedia’s account, imitated by numerous subsequent open-access journals and became, by 2010, the world’s largest single journal by volume of articles published [8].

Both arXiv and PLOS ONE are best understood as institutional responses to the same underlying cost structure that had built up around subscription journals over the twentieth century: circulation was slow, access was gated behind subscription prices that many institutions and most individual readers could not afford, and the editorial criterion of “predicted importance” gave reviewers and editors a subjective lever that could reject methodologically sound but unfashionable work. Neither innovation eliminated peer review — arXiv sits alongside it rather than replacing it, and PLOS ONE still requires it — but each reduced a specific institutional friction: arXiv reduced the friction of waiting for review before any reader could see a result; PLOS ONE reduced the friction of an editor’s subjective importance judgment standing between a methodologically sound paper and print.

Figure 5. arXiv, launched in 1991, and open-access journals such as PLOS ONE, launched in 2006, moved review and circulation outside the older subscription model [@arxiv-about] [@plos-one-wikipedia]. — Image prompt and art direction by Brecht Corbeel; generation pending.

Reading the four hinges together

Laid end to end, the pattern is legible. 1660–1665 solved the problem of turning private observation into a shared, dated, citable record, but left peer review as an informal, editor-driven practice for a century and a half and did nothing about training capacity or funding. The nineteenth-century German laboratory model solved the training-and-throughput problem by fusing research and instruction in one room, but left funding dependent on patronage and university appointment rather than any systematic public mechanism. The 1945–1950 creation of NSF solved the funding problem by making competitive, peer-reviewed federal grants a standing government function, but the resulting increase in publication volume and career pressure exposed a replication gap that funding and publication institutions had not been built to address. And the 1991–2006 preprint and open-access wave addressed circulation speed and access cost, without on its own resolving either the replication gap or the deeper question of what “peer reviewed” should mean when review can now happen in public, in real time, on a posted preprint, rather than privately before publication.

Two present-day institutional experiments follow directly from that unresolved thread and deserve to be named as experiments, not settled solutions. Registered reports — a review format in which a study’s methodology is peer-reviewed and provisionally accepted before data collection, precisely to remove the incentive to publish only statistically significant results — are a direct, traceable institutional response to the replication gap the 2015 psychology study documented [6]. Open peer review, in which referee reports and reviewer identities are published alongside a paper, is a direct response to the same transparency pressure that produced preprint servers and open-access mega-journals. Both are still minority practices relative to conventional closed peer review at subscription and hybrid journals, and their long-run effects on the same replication metrics the 2015 study used remain to be measured by future, similarly coordinated replication projects — a test this article does not have the data to prejudge.

What is fact, what is claim, and what is still open

To be explicit about the register of each claim above: that the Royal Society was founded in 1660, chartered in 1662, and began Philosophical Transactions in 1665 under Henry Oldenburg is documented institutional fact [1] [2]. That formal external refereeing at that journal began only in 1832 is a specific, separately documented fact that corrects any assumption of continuous modern peer review since the seventeenth century [1]. That Liebig’s Giessen laboratory fused research and training and was widely imitated is documented by historical-of-science institutions, though the exact causal weight of Giessen specifically versus other German laboratories of the same period is a matter of ongoing historical scholarship, not a single settled ranking. That Bush’s 1945 report recommended a federal research foundation and that NSF was signed into law in 1950 are primary-source and institutional facts [4] [3]; that Bush’s report constitutes the single governing philosophy of all subsequent science policy is a stronger interpretive claim this article does not endorse. That 36 of 97 psychology studies replicated at significance in the 2015 project is a specific, peer-reviewed empirical result about one field and one method [6]; generalizing it to “most published research is false” is a claim beyond what that study alone supports. And that arXiv and PLOS ONE changed circulation speed and access cost is documented [7] [8]; whether open science practices broadly have yet closed the replication gap the 2015 study identified is, honestly, an open empirical question that would require a comparable coordinated replication effort across fields and years to answer, and this article makes no prediction about its outcome beyond noting that registered reports and open peer review are the two institutional mechanisms currently designed to test it.

The institutional history of science, read this way, is not a straight line of progress toward an already-perfected method. It is a sequence of specific fixes to specific bottlenecks — circulation, training, funding, verification — each of which left the next bottleneck exposed for a later generation of institution-builders to find.

Sources

1. The Royal Society. History of the Royal Society. Royal Society (2024).
2. A History of Scientific Journals project, University of St Andrews. History of Philosophical Transactions. University of St Andrews (2020).
3. National Science Foundation. History - About NSF. National Science Foundation (2024).
4. Vannevar Bush. Science: The Endless Frontier. Office of Scientific Research and Development / U.S. Government Printing Office (1945).
5. Science History Institute. Justus von Liebig and Friedrich Wöhler. Science History Institute (2023).
6. Open Science Collaboration. Estimating the reproducibility of psychological science. Science (2015). DOI: 10.1126/science.aac4716.
7. arXiv / Cornell University. About arXiv. arXiv (2024).
8. Wikipedia contributors. PLOS One. Wikipedia (2024).
9. Center for Open Science. Open Science Collaboration: Reproducibility Project: Psychology. Center for Open Science (2015).

Originally published at https://absolutedigitalpublishers.com/articles/from-origins-to-frontier-a-history-of-the-evolution-of-scientific-institutions.