The United States remains the world’s leading investor in research and development. The National Science Foundation’s Science and Engineering Indicators 2026 document persistent dominance in knowledge production, but show that the diffusion of this knowledge depends on mediating institutions whose condition may diverge from the scientific frontier itself. A society can dominate global discovery while weakening its capacity to draw informed public decisions from it.
The Essentials
- The United States maintains its position as a global leader in scientific production, but research performance and democratic capacity to use science are two distinct goods.
- The NSF’s Science and Engineering Indicators 2026 show that knowledge diffusion depends on mediating institutions—scientific media, expert agencies, school curricula—whose condition may diverge from laboratory performance.
- Financing discoveries and financing the conditions for their reception are two different commitments, often conflated in public policy.
- Philosopher Mazarine Pingeot diagnoses this same divide for AI: producing tools without building the institutions that articulate what they change amounts to displacing our relationship to truth without offering antidotes.
- If the gap between production and mediation widens, high-science-intensity societies could make collective decisions less well-informed than their research budgets would suggest.
The Scientific Frontier Remains Held, But Mastery Is Concentrating
The United States spends more on research and development than any other country in the world. According to the NSF’s Science and Engineering Indicators 2026, the American share in global production of high-quality scientific articles remains considerable, and the country continues to form and attract a disproportionate share of active researchers at the frontier of their discipline. Patents, citations, university spinoffs: on each of these indicators, the United States maintains its top rank.
But this accounting measures production, not reception. It says how much new knowledge enters the world. It says nothing about how that knowledge passes through institutions to reach decision-makers, teachers, journalists, citizens. Yet it is in this trajectory that science becomes—or does not become—usable in public decision-making.
The history of American public health policy over the past twenty years illustrates this gap with uncomfortable clarity. Laboratories produced vaccines in record time, developed antivirals, mapped genomes at minimal costs. Meanwhile, growing portions of the population expressed distrust in the expert institutions supposed to relay these results. The scientific frontier had not budged. The mediation infrastructure, meanwhile, had been weakened by two decades of media disaffection, the rise of disinformation platforms, and underinvestment in science education.
Two Budgets for Two Functions That Don’t Add Up
There is a persistent confusion in the debate over science policy: a country’s effort is measured by its R&D spending, and from this one infers its capacity to decide in an informed way. This shortcut ignores a fundamental distinction.
Financing a molecular biology laboratory means financing the production of a result. Financing the National Agency for Health Safety so it can evaluate that result, compare it to other data, translate it into recommendations accessible to elected officials and physicians, means financing mediation. Financing scientific journalism capable of distinguishing a preliminary study from a consensus means financing reception. These are three distinct goods, produced by three different types of institutions, whose funding trajectories have no reason to be synchronized.
The NSF’s Science and Engineering Indicators, in their 2026 edition, document precisely this possible dissociation. They show that while indicators of research performance remain solid, knowledge diffusion depends on an ecosystem of intermediary institutions whose condition is treated as a secondary variable in public policy. Mediation budgets, science education, public communication, independent evaluation agencies, are structurally less visible than research budgets, harder to defend politically, and more vulnerable to austerity cycles.
The point is not that the United States has collectively abandoned these institutions. Many are holding up and working well. But their relative robustness, compared to the productive power of the laboratories they are supposed to extend, constitutes an imbalance worth naming.
Mazarine Pingeot’s Angle
Philosopher Mazarine Pingeot, in her work on artificial intelligence and democracy, diagnoses an analogous displacement for language technologies. Her central argument: generative AI tools are not regulated by technical safeguards; they call for political institutions capable of articulating what they change in our collective relationship to truth and decision-making. Producing models without building the conditions for their democratic reception amounts to modifying the ground on which the collective pact rests, without warning its inhabitants.
Transposed to the scientific question, the argument takes a precise form. Financing laboratories without financing the institutions that render their results credible and intelligible in public space amounts to producing knowledge without the conditions for its reception. Science arrives in a public space it does not control, carried by mediating institutions on which it depends but which it does not directly finance.
This reading is not universally shared. Part of economic liberalism, notably embodied in Tyler Cowen’s work on innovation and secular stagnation, defends the idea that information markets, if sufficiently open and competitive, naturally produce the intermediaries society needs. On this terrain, the market for scientific podcasts, specialized newsletters, and independent communicators testifies to genuine vitality. Demand exists, suppliers rise to meet it, and the state is not necessarily the best architect of this mediation.
The argument is sound in its register. But it stumbles on an asymmetry. Digital platforms that distribute scientific information optimize for engagement, not rigor. The attention economy favors content that polarizes rather than content that nuances. An independent scientific journalist, however competent, does not have the institutional infrastructure of a public evaluation agency to investigate a case in depth.
Market mediation and institutional mediation produce different goods, whose effects do not substitute for one another.
Institution as a Condition of Credibility
The credibility of a scientific result beyond the laboratory rests on the network of institutions that confer public authority upon it: the peer-reviewed journal, the independent evaluation agency, the university that certifies, the investigative journalism that questions, the school that trains minds to distinguish correlation from causation.
These institutions have a long history and their construction took decades. Their weakening produces effects that do not appear immediately in scientific performance indicators: laboratories continue to publish, while receiving institutions deteriorate.
American federal funding for basic research via the NSF or NIH has, for decades, been a model studied worldwide. Budgets are substantial, evaluation processes robust, agency independence defended. But these same agencies have seen, repeatedly, their recommendations circumvented or ignored in high-visibility public decisions, not because their work was incorrect, but because the political institutions charged with receiving them had lost, or deliberately set aside, the capacity to integrate them. Digital participation in public decisions documented by Beth Simone Noveck supposes precisely that institutions accept to share not only information, but the reasons that support it.
The salient point of the Science and Engineering Indicators 2026 lies here: they measure what the United States produces, not what American society is capable of absorbing in its decisions. These two measures are related, but the second does not automatically follow from the first.
The Weakening of Reception Capacity
The American case results from cumulative choices and remains an observable configuration. Several signals are documented, even if none alone constitutes proof of institutional collapse.
Federal funding for local journalism in the United States has declined substantially since 2005, with job losses in regional newsrooms. Yet local journalism is often the first relay through which scientific recommendations on public health, environment, or food safety reach citizens who do not read specialized journals. Its disappearance is not offset by newsletters from specialists addressing already-informed publics.
Science school curricula have experienced significant variations by state, sometimes introducing controversies on subjects—evolution, climate change—for which scientific consensus is robust. This variability does not affect laboratories. It affects the cohort of future citizens who will, in ten or twenty years, need to evaluate public policies with scientific content.
Federal expert agencies have suffered cycles of political disaffection, where their recommendations were either ignored or actively contested by officials who formally depended on them. The budget of the Environmental Protection Agency, FDA, or CDC has experienced repeated compression that did not prevent these institutions from functioning, but did affect their capacity to investigate complex cases within politically useful timeframes.
None of these elements taken in isolation is decisive. Together, they sketch a trend: investments in knowledge production and investments in institutions that render it usable are evolving at different speeds, sometimes in divergent directions.
The Coming Decades Will Test These Balances
The fundamental question posed by the Science and Engineering Indicators 2026, even if the report does not formulate it in these terms, is a question of institutional architecture. Over the 2030-2040 horizon, high-science-intensity societies will need to address collective decisions of growing complexity: energy transition, AI regulation, pandemic management, climate adaptation. Each of these decisions presupposes that political institutions are capable of investigating files with high scientific density, distinguishing what is established from what is uncertain, and honestly communicating that uncertainty to citizens who must ultimately validate the choices.
Two trajectories are plausible. In the first, democracies invest in both legs simultaneously: research and mediation. They maintain well-funded independent expert agencies, support quality scientific journalism, reform curricula to teach not only results but reasoning. The scientific frontier remains held, and reception capacity follows. In this trajectory, American scientific power genuinely translates into decision-making advantage.
In the second trajectory, the imbalance intensifies. Laboratories continue to produce at high intensity while mediating institutions crumble under the combined effect of budget cuts, media fragmentation, and political polarization. Society then has cutting-edge science that it uses unevenly: certain actors—companies, well-connected experts, well-funded administrations—access results and integrate them; a large portion of the population navigates an informational space that mixes verified results and unfounded claims without the tools to distinguish them. Scientific advantage becomes a club good rather than a public good.
Generative artificial intelligence complicates this picture by adding a variable that neither the NSF Indicators nor traditional science policies anticipate well. When language models produce scientifically plausible but factually inaccurate content at scale and at zero marginal cost, mediating institutions—fact-checking, investigative journalism, evaluation agencies—face a new burden for which their funding has not been sized. Pingeot is right to point out that this dynamic calls for a political and institutional response, not merely a technical one. But formulating that response requires precisely the collective capacities that the imbalance between production and mediation is eroding.
The signals to watch are concrete: evolution of federal expert agency budgets, trajectory of employment in science journalism, results of surveys on trust in scientific institutions, and especially governments’ and states’ ability to effectively integrate scientific recommendations within politically relevant timeframes. These signals do not appear in the Science and Engineering Indicators, which measure what is produced in laboratories. Building equivalent indicators to measure what occurs in receiving institutions is itself a political project. No one has seriously launched it yet.
The Distinction Missing from the Debate
The American debate over science is conducted almost entirely in the register of production: how much do we spend, how much do we publish, how many Nobel Prizes, how many patents. This is a useful register. But it leaves in shadow the question of use, which is a question of institutions as much as knowledge.
A country can remain at the global scientific frontier while simultaneously losing the collective capacity to use what that frontier produces for its most important decisions. These two degradations do not occur at the same pace, are not measured by the same instruments, and are not corrected by the same policies. To confuse them is to believe that an R&D budget suffices to guarantee an informed democracy.
The distinction between financing discoveries and financing the institutions that render them credible and usable deserves to enter the ordinary vocabulary of science policy. It is not there yet.