ADAPTING TO THE CHANGING NATURE OF SCIENCE

The 19th and 20th centuries witnessed a series of institutional innovations that ushered in a fertile era of discovery. The natural philosophy of earlier ages, where a single scholar might range freely across what we now call physics, chemistry, and biology, had given way to specialized disciplines. Each had its own departments, journals, and staff. This transformation, pioneered primarily in German universities during the 19th century, laid the foundation for scientific professionalization.54

America adopted this architecture of discovery and gave it a distinctive twist. Land-grant universities democratized access to scientific education, the college major system restructured undergraduate training, and technical universities focused on industry. These innovations proved remarkably successful. The combination of disciplinary depth with clear career ladders enabled the specialization that carried us into the scientific lead. Yet the disciplinary framework that served us in the 20th century sits uneasily with science in the 21st. The most consequential questions of our era, such as how proteins fold and function, how certain disorders emerge from neural circuits, and how we can harness fusion energy, do not respect departmental boundaries. The protein folding problem that earned the 2024 Nobel Prize in Chemistry belonged to no single academic department; it required deep knowledge of biochemistry alongside advances in AI and engineering at scale.55 Two of the three laureates came from a corporate research lab where team-based science harnessed diverse expertise. The third laureate’s work was seeded by NSF and later heavily supported by private philanthropy, spinning off into a large-scale, university-affiliated center.56 Both efforts depended on decades of infrastructure building and open science, such as the Critical Assessment of Structure Prediction competitions and the Protein Data Bank, built by over 60,000 experimentalists who shared their findings freely.57 The shape of an institution determines the shape of the science it produces.58 The university laboratory, centered on the principal investigator and a rotating cast of graduate students, excels at curiosity-driven research and at training the next generation within established disciplines. The industrial R&D lab, with its permanent staff and tight feedback loops, is suited to engineering projects.59 The national laboratory and large-scale, multinational scientific projects maintain unique capabilities too large for any single actor to sustain. Each form is a container that dictates what research becomes possible and what research never gets attempted.

A growing share of scientific problems now demands containers that do not yet exist at scale in the federal portfolio, and does not yield immediate products that contribute to a company’s bottom line.60 Mapping the wiring of the mammalian brain, for example, requires not a student on a three-year cycle, but a sustained engineering team with the flexibility to scale quickly and hire from industry. It necessitates industrial-scale data collection and analysis, which is infeasible under the fragmented structure of traditional academic grants. It produces a public good whose benefits a single biotechnology company cannot fully internalize.

This pattern extends across other key problems. Achieving practical fusion energy depends on progress in plasma physics, materials science, advanced manufacturing, and systems engineering, paired with the ability to scale up through venture funding. Understanding cognition well enough to address mental disease will benefit from new neural recording probes, machine learning suites that analyze neural activity patterns, and systems that deliver precise, closed-loop therapeutic interventions. Creating new institutional forms, then, plays a central role in bringing new scientific projects to life.

The standard NIH R01 grant, which typically offers a quarter of a million dollars a year for a pre-specified project, is the workhorse of American biomedical research. It excels at supporting hypothesis-driven science by small teams on tractable questions. But over-reliance on this structure creates systematic blind spots. Projects requiring tens of millions of dollars and a team of dozens fall outside the container of what any single investigator can assemble. Academic containers are further shaped by labor availability. Employing postdocs and graduate students remains effective for training the next generation of scientists, but doing so is poorly suited for executing large-scale, mission-driven programs that require continuity, specialization, and long-term institutional memory.

Frequent turnover fragments efforts and slows progress. As science funders have noted, no technology company would entrust its core R&D to a workforce composed primarily of temporary trainees, yet this is the standard model in academic research.61 A more balanced approach would expand stable, well-compensated career paths for staff scientists, engineers, and technicians, roles critical to sustained institutional capability.

NOVEL PERFORMERS

Between the atomized work of the individual investigator and billion-dollar megaprojects like particle colliders lies a vast middle ground of mid-scale science. These are scientific problems requiring tens of millions of dollars, coordinated teams of ten to a hundred people, and timelines of half a decade. They range from the development of minimally invasive brain-computer interfaces that help people with Parkinson’s to the building of new platforms that decode immune memory. Such projects, which are often infrastructure-heavy, engineeringintensive, and cross-disciplinary, are challenging to perform in principal investigator-led academic labs. They are rarely pursued in industry either. Pharmaceutical companies face much stronger incentives to chase the next drug breakthrough than to build platform technologies for decoding basic biology.62 The ARPA model has proven so effective because it fills precisely this gap in mid-scale technology development. By offering grants in the tens of millions of dollars, these agencies can assemble new research teams and help startups tackle ambitious engineering challenges, from robotic satellite servicing to AI-equipped fighter jets.

Funding at this scale, which has worked well in incubating new technological capabilities, could be extended to basic science as well, where the government is not merely procuring a weapons system, but pursuing scientific advancement for the national interest.

Among private funders, a new class of focused research organizations, or “FROs,” has started to fill the gap. FROs are time-bound, nonprofit research startups, engineered to break specific scientific bottlenecks. They hire professional engineers and career scientists, building institutional memory instead of just cycling graduate students through their training process. They produce public goods like open datasets, platforms, and tools, rather than the proprietary intellectual property that defines the commercial startup. And unlike the national laboratory, which is built to last indefinitely, the FRO is built to dissolve, pursuing a welldefined technical milestone and winding down once the mission is complete.

These organizations can set long time horizon milestones to target specific bottlenecks and provide full salary support, removing the grant-writing treadmill.63 The time-limited nature also gives scientists who complete the project a chance to return to academia, or to spin off a startup and raise venture capital. 25

The FRO model acts as an open call for new kinds of science, and for ideas our researchers have rarely dared to pursue thus far because there are no avenues for them. Academic incentives filter out team-based execution; commercial incentives filter out public goods; national laboratories filter out the agility to hire flexibly and execute rapidly. The FRO occupies the new ground of problems too large for the standard federal grant, too non-commercial for venture capital, and too risky and fast-moving for government facilities. But the FRO is just one point in a broader design space (Table 1). As the examples of Cold Spring Harbor Laboratory and the Institute for Advanced

Study illustrate, many other approaches are possible. One proposal taxonomizes a range of novel institutional structures, including minimally constrained homes for basic science, FRO-style teams that execute against specific bottlenecks, and specific formats focused on the scouting and seed-funding of non-consensus ideas.64 Others have written about the variables that together map out the design space: the timeline over which projects are expected to pay off, the revenue strategy, the intellectual property policy, the size of the team, and the use of clear market signals to drive problem selection, among others.65 Table 1: Established institutional forms, such as universities, corporate laboratories, and federal laboratories, each carry their own relative advantages. Future modes of organization should be designed to fill the scientific gaps our existing institutions miss. Activity University Corporate Lab Federal Lab New Institutions Curiosity-driven, investigator-led research + × ≠ By design Larger-scale, engineeringintensive science × ≠ ≠ By design Long-horizon platform and tool development × ≠ + By design Public-goods data and infrastructure development ≠ × ≠ By design Mission-driven, public-good science ≠ × ≠ By design Proprietary product development × + × By design Workforce training and apprenticeship + ≠ × By design

Legend: + Well-Suited ≠ Partially-Suited × Less-Suited 26

Betting exclusively on the existing funding model is like building a military composed entirely of infantry, effective for one kind of warfare, inadequate for others. Fortunately, there are ways to expand the scope of federal grantmaking to support these innovations. NIH and NSF possess OTA that allows them to bypass traditional grant constraints. Our national laboratories can also create pathways to stand up flexible, federally supported scientific teams on time-bound missions. Recently, NSF’s TIP Directorate launched the X-Labs, the first federal program explicitly designed to fund independent research organizations outside of traditional academic institutions. X-Labs will provide full-time teams of researchers, scientists, and engineers with operational autonomy and milestone-based funding as they pursue technical breakthroughs. These teams will not only produce traditional research outputs like publications and datasets, but also command the resources and financial runway to develop revolutionary platform technologies that unlock new fields of scientific inquiry. NSF’s X-Labs represents a proof of concept for what federal science funding can become. Consider the mammalian brain mapping example again: one of neuroscience’s grand challenges. A federally supported initiative could draw from extensive public-private partnerships, leverage matching grants from America’s vibrant philanthropic sector, and bring the best researchers from academia together to develop moonshot infrastructure that scales connectome mapping, much as the Human Genome Project commoditized genetic sequencing. A hypothetical X-Lab could help map the reward and motivation circuits across various small mammals, producing one-of-a-kind datasets. For medicine, these circuit diagrams would offer a way to map the circuitry implicated in depression, addiction, and autism. For AI, they would provide a biological reference architecture for building more robust systems, drawn from natural structures that keep impulses in check and align short-term behavior with long-term goals.66

NEW MECHANISMS

Reformed and new scientific institutions should also be matched with a broader menu of improved selection mechanisms for determining who and what type of organization receives scientific funding. Just as some organizations are better suited to certain kinds of scientific projects than others, so too are some selection processes better than others at identifying and motivating promising talent and programs.

The economic field of mechanism design, recognized with the 2007 Nobel Prize, provides the theoretical foundation for understanding how rules shape behavior and outcomes. Mechanism design can be thought of as asking the inverse of traditional economics. Given a desired outcome, what incentives and institutions will produce it? The field has already transformed how we allocate spectrum licenses, match medical residents to hospitals, and price internet advertising. Science funding is an equally rich domain for applying the field’s tools. At its core, our challenge is that scientific research involves private information that funders cannot directly observe. Peer review emerged as one solution to this information problem, but while peer review has a long and time-honored history of distinguishing good science from bad, it struggles to distinguish the exceptional from the merely good. Furthermore, reliability is low, and multiple reviewers rating the same NIH proposal frequently reach contradictory conclusions about the credibility of the science.67 These structural flaws only get worse as the number of proposals rises. When a funding agency can support only one proposal in ten, the noise begins to drown out the signal. All too often, consensusdriven panels fund the least divisive ideas rather than the most promising.

While these problems are widely recognized, reform has been slow because the incentives are asymmetric. A failed experiment invites Congressional scrutiny, whereas continuing the mediocre status quo draws little attention. The review panel thus serves as a convenient liability shield, allowing decisions to be attributed to “the scientific community” rather than to any individual who might be held accountable. A mechanism designed to hedge risk ends up precluding the risk-taking that breakthrough science requires. But evidence from recent experiments suggests that modifications to funding mechanisms can yield substantial gains in both efficiency and scientific output.

BETTER GRANTMAKING

One approach is to improve how peer review functions. In Denmark, a private foundation has experimented with a “golden ticket” system that allows individual reviewers to champion unconventional proposals lacking consensus support.68 This model helps rescue high-risk breakthroughs that colleagues might reject, and it is supported by a double-blind process that removes career history, cutting against elitism and leveling the playing field for younger or less prominent researchers.69 The approach also tends to attract higher-quality reviewers, who are individually empowered to make bold scientific bets.

NSF has begun piloting golden tickets under its TIP Directorate, and opportunities exist for broader adoption across new federal extramural funding agencies.70

Another approach is to change what we fund, supporting people over projects. Using philanthropic money, the Howard Hughes Medical Institute (HHMI) has provided long-horizon support in roughly $10 million over seven years with minimal reporting requirements, while tolerating early failure and betting on people rather than on project proposals.71 When researchers compared these grantees against similarly accomplished federally funded scientists, they found that the privately supported investigators, who have been granted more academic freedom, produced high-impact publications at nearly double the rate of their peers and were far more likely to explore genuinely novel lines of inquiry.72 Similar approaches have been tried in federal agencies, but remain too small a share of the current portfolio. NIH’s own Director’s Pioneer Award, designed to emulate the HHMI program, shows comparable results,73 and NSF’s CAREER award, though smaller in grant size, has also produced countless breakthroughs.

The same philosophy of betting on individuals can be used to support younger researchers as well. Since 1952, NSF GRFP has directly funded some of America’s most promising incoming doctoral students. While more work remains to improve the selection mechanism and further empower students to choose their universities and principal investigators, such an approach has shown significant promise. The fellowship provides three years of support with full portability across institutions, freeing recipients to follow intellectual opportunity. The results speak for themselves: more than forty GRFP alumni have gone on to become Nobel laureates.74

These long-time-horizon grants can be matched with fast grants that provide flexibility on shorter timescales. NSF has mechanisms for fast decision-making that bypass external review panels, but they remain underutilized and often behind schedule. Meanwhile, a privately funded American program has demonstrated that, without any significant sacrifice to scientific quality, funding decisions can be made effectively in 48 hours rather than 6 to 9 months, with applications that take 30 minutes, rather than months, to prepare.75 Scaling these just-in-time grants up within federal grantmaking agencies could encourage more risk-taking on novel ideas, all while reducing administrative burdens.

PRIZE CHALLENGES

Pull mechanisms offer another powerful and underutilized alternative to traditional funding, aligning incentives around outcomes rather than inputs.

Traditional grants pay for effort, such as researcher time, equipment, and supplies, regardless of whether the project succeeds. Pull mechanisms invert this logic by paying for results. The case for pull mechanisms is strongest when the goal is clear, but the path to it is not. The most famous example is the DARPA Grand Challenge for autonomous vehicles, which catalyzed an entire industry. DOE and ARPA-E have also used similar prize authorities to accelerate breakthroughs in energy storage and grid technology.76 A related mechanism is the advanced market commitment, which guarantees a market for a scientific or technical capability before a product exists.

Such approaches can generate massive investment leverage. A privately funded prize for suborbital spaceflight offered $10 million but triggered hundreds of millions in combined research and development spending across competing teams.77 Similarly, the open structure of another prize competition attracted solvers from unconventional backgrounds to read the unopenable Herculaneum scrolls, a feat eventually accomplished not by seasoned archaeologists but by a trio of computer science and robotics students.78 While these mechanisms are illsuited for open-ended, curiosity-driven research, they can be powerful tools for incentivizing use-inspired research and supporting technology commercialization. An optimal innovation portfolio requires both push mechanisms to explore unknown territory and pull mechanisms to close identified gaps.

FUTURE IDEAS

The list of examples go on. Some of these mechanisms already exist in our federal portfolio and should be used more, others should be experimented with, and still others have yet to be invented. Each addresses different aspects of the same underlying challenge. Each represents a hypothesis about how to elicit honest signals, reward productive risk-taking, and allocate resources where they will generate the greatest return. One emerging idea, for instance, is to decentralize decisions. Doing so can leverage the wisdom of crowds to identify good science. Scouts, financially rewarded to find promising projects and individuals, could help identify scientific research for grantmakers. At a larger scale, the “regranting” model rests on the observation that the people best positioned to spot breakthrough opportunities often lack the authority to fund them, while those with the authority lack information to spot them. Regranting bridges this gap, delegating funding allocation to researchers or experts who possess the specific judgment to identify promising work before consensus forms. Existing intermediaries already perform this function with philanthropic funding. Such a model could be extended by funding portfolio-based regranting organizations through federal agencies, or by giving a broad range of scientists the ability to regrant a small check to anyone other than those in their own academic institutions. More speculative mechanisms, such as quadratic funding, remain in early testing.79 This approach weights the breadth of support more heavily than depth. A proposal backed by many small contributions receives larger matching funds than one backed by a few large donors. Quadratic funding reveals community preferences rather than gatekeeper preferences, and has shown promise in opensource software, though evidence of its application to science remains pending. There must ultimately be a menu of options from which those who exercise federal funding authority can choose. The current selection system concentrates decisions among too few people using mechanisms that cannot support the weight placed on them. We stand at the beginning of a renaissance in grantmaking, and the Federal Government should welcome this experimentation. A PORTFOLIO-BASED APPROACH Private capital allocators must deliver results or risk losing their investors. Philanthropies compete for donor confidence. But federal program officers receive little corrective feedback when their grant portfolios systematically underperform, and agencies rarely compare outcomes across funding mechanisms or allocation strategies. Just as investment funds in the private sector balance their portfolios and match mechanisms to the nature of the work, we need to move toward a far more intentional approach to grantmaking. The preceding pages cataloged a diverse arsenal of mechanisms: golden tickets, which move us beyond false consensus; individual-based funding, which bets on researchers rather than proposals; pull mechanisms, which pay for outcomes rather than inputs; and regranting, which delegates decisions to those closest to the frontier. Each works for certain problems, operates well within certain institutional constraints, and produces returns with a particular risk profile.

We can also be intentional about where we place various programs on the exploration-exploitation trade-off. Bold scientific bets can pay off in big ways; the biggest breakthroughs of the past decades have more often than not been driven by a relentless pursuit of tools and frameworks to answer practical questions. Science is not a pure random walk; it often helps to have an inductive bias. This is Pasteur’s quadrant, the domain of use-inspired basic research, which we discussed in Chapter I. But pure curiosity-driven research can also deliver immense value to society. Riemann’s abstract study of differential geometry eventually enabled Einstein’s formulation of general relativity; the field of group theory eventually enabled cryptographic codes, computer graphics, and our understanding of elementary particle physics. The key lies in distinguishing between cases where strategic direction can accelerate progress, and cases where the fog is too thick for anything but an exploratory search. Intentional grantmaking therefore requires deliberate portfolio construction: a mix of high-risk and low-risk bets; a balance of person-based, project-based, and institution-based funding; and explicit strategies for allocating across fields and capability areas. Federal agencies should construct their portfolios the way sophisticated allocators do, with thesis-driven conviction about where breakthroughs are most likely to emerge, while preserving space for the serendipity that no thesis can anticipate. We should aim to engineer a large, well-constructed portfolio that allows us to win in the long run.