Science, Heuretica, and Funding the Public Good
If you speak to anyone involved in publicly funded research (except a university provost) you will hear a ready opinion on its dysfunction. To the first of two groups, this dysfunction can be solved with some form of more: our laboratories need more money, the government needs to write more grants, the country needs more PhDs, and any number of other prescriptions that call to mind turning up so many dials and knobs.
To the second group, the problem is systemic. There’s a growing body of evidence that scientific progress has slowed down, as shown for example by Ben Southwood and Tyler Cowen in their 2019 paper. Spending goes up, but productivity per scientist keeps falling, and no one knows exactly why.
There's a pessimistic explanation and an optimistic one. The pessimistic account is that the problem is inherent. After the explosion of discovery and invention over the last three centuries, we’ve now picked all the low-hanging fruit; by its very nature, the scientific work that comes next will be more difficult and expensive.
The optimistic account is that the problem is institutional. As the scientific enterprise has grown, research universities and grant-making bodies alike have ossified, laying bureaucracy atop bureaucracy. Incentives in now-crowded fields push researchers toward incremental, milquetoast work that’s staid enough to survive peer review, and have led to p-hacking and replicability crises. We’ve arrived at a place where most published research findings are false.
Why is this account the optimistic one? Because it’s something that can be fixed.
I find it likely that both the pessimistic and optimistic accounts are true to an extent. Gone are the days when we can add to the intellectual commonwealth by tying a house key and Leyden jar to some kite string. At the dawn of the 20th century, David Hilbert knew all of mathematics in his day, while experts in the 21st century need to specialize in sub-sub-fields to master what’s already known. Yet despite such observations, our well-documented institutional failures suggest that we’re far away from the optimal frontier of scientific pursuit.
It only makes sense to ignore the first cause and attack the second (except to the extent that the former affects the latter). There is nothing to be done about the nature of science. There is everything to be done about reordering the institutions of scientific progress.
And what, in all of this, is the proper role of government?
Thomas Jefferson wrote that the peculiar character of an idea “is that no one possesses the less, because every other possesses the whole of it. He who receives an idea from me, receives instruction himself without lessening mine; as he who lights his taper at mine, receives light without darkening me.”
In economic terms, an idea is a public good, which means it is neither rivalrous nor excludable. Because an idea is non-rivalrous (my use doesn’t preclude yours) and non-excludable (I don’t have a way to make you pay for using it), markets often fail to yield efficient outcomes in the way they do for bread or smartphones. The classic example of a public good is a lighthouse.
When recounting the early Industrial Revolution and the sudden explosion of human flourishing, we often hear about the importance of the steam engine, the power loom, and machine tools, but rarely mentioned is arguably the most vital invention of the era: intellectual property.
Modern intellectual property grew out of reforms to the English king’s prerogative to grant trade monopolies, and it’s still best understood as a form of monopoly. A patent isn’t a legal right to your invention—it’s a legal right to exclude others from copying or selling or otherwise profiting from it.
We rightly regard such powers with suspicion, but most economists agree that a simple, balanced set of intellectual property laws spur technological progress, to the great benefit of society. The data seem to bear this out: a paper by Nobel laureate William Nordhaus finds that innovators capture only 2% of the value of their work, and the remaining 98% spills over to society as a whole. (Of course, bloated or overbroad intellectual property laws curb technological progress in their own way, with patent trolls and Mickey Mouse laws.)
Yet when it comes to technological progress in its rawest form—the conjectures, discoveries, and experimental knowledge that constitute science—ideas aren’t always amenable to traditional patents in the way that inventions are. (What would it even mean to grant a monopoly over, say, the law of gravity?)
This is why there are, broadly speaking, two institutional roles our government can and does play in cultivating science, technology, and “useful arts”: the indirect legal role of establishing an intellectual property regime, and the direct administrative role of funding and organizing research and development.
Accepting the abstract argument—science is a public good, patents can fail to incentivize it, and therefore the government should intervene directly by funding R&D—is insufficient. Government ceaselessly shows itself to be a terrible investor, and bureaucrats and politicians are no less driven by self-interest than scientists and entrepreneurs. Science can be warped for political causes. Government science has in different times given us Lysenkoism, involuntary sterilization, and the food pyramid.
Even if we could ignore the moral dimension, in that all state action necessarily involves an element of force, at the very least we must demonstrate that our particular program for funding science, amidst bad political actors and perverse incentives, is better than the program’s not being there in the first place. And if we’re responsible, we should demonstrate that our program not only outperforms the null case, but alternative programs as well. The U.S. government currently does not attempt this.
The most recent example of how we misunderstand the R&D dilemma is the proposed Endless Frontier Act, unveiled in May, which commentators argue will supercharge national R&D. Its name comes from a famous 1945 report by Vannevar Bush that shaped our modern research complex.
Lamentably, and perhaps expectedly, the bipartisan bill is a product of that Panglossian first group who believe our problems can be fixed with more. Entrepreneur and engineer Benjamin Reinhardt, host of the Idea Machines podcast, has written about the features of the bill in detail. The bill would increase funding of the National Science Foundation (NSF)—the primary source of non-defense R&D grants alongside the National Institutes of Health (NIH)—by $100 billion over the next five years. The current NSF budget is $8.3 billion, the NIH $42 billion.
The NSF would be renamed the National Science and Technology Foundation, and its mandate expanded accordingly. The new money would be allocated to today’s most fashionable technologies like artificial intelligence and quantum computing. New grants would be awarded in much the same manner as now.
But we do not want for money. Total U.S R&D expenditures are close to $600 billion today. Since the end of the Apollo program in 1972, federal R&D expenditures as a share of total discretionary spending have been fairly stable.
One worry I’ve seen raised is that hidden amongst the growth of R&D spending over time is a neglect of basic, long-term research, as businesses tend to invest in work that can be quickly commercialized. The NSF helpfully divides R&D into three categories:
Basic Research: Experimental or theoretical work undertaken primarily to acquire new knowledge of the underlying foundations of phenomena and observable facts, without any particular application or use in view.
Applied Research: Original investigation undertaken to acquire new knowledge; directed primarily, however, toward a specific, practical aim or objective.
Experimental Development: Systematic work, drawing on knowledge gained from research and practical experience and producing additional knowledge, which is directed to producing new products or processes or to improving existing products or processes.
While businesses heavily invest in experimental development, as one would expect, basic and applied research still see healthy investment from all sectors. The federal government today spends over $80 billion per year on basic and applied research.
The Endless Frontier Act actually shifts the distribution of federal R&D funding away from basic research, as most of the $100 billion will be spent on the development of existing, nascent technologies. The closer R&D is to a commercial product, the better intellectual property rights and market incentives work. In that light, one might expect the government distribution to look something like the inverse of business, with roughly 80% of funds allocated to basic and applied research. Pushing our federal R&D toward a dirigiste industrial policy is perhaps the last thing American science and technology need.
Another popular worry is national security. If we fail to invest in scientific research, the argument goes, the best minds will go elsewhere, threatening our position in the world, especially in the face of a rising, truculent China. But in 2017, U.S. spending on basic and applied research ($91.5 billion and 108.8 billion, respectively), dwarfed that of the closest countries, China ($27.5 billion, $52.1 billion) and Japan ($22.4 billion, $31.9 billion).
While I’m sympathetic to new public research spending, we shouldn’t deceive ourselves into thinking that more money will necessarily be helpful. There are barnacles to scrape and bilgewater to pump. The Endless Frontier Act does little to redress the systemic problems in American research and may well exacerbate them, disbursing new money by the same old degenerate set of incentives.
Universities typically spend roughly a third of grant money on administrative overhead. Professors spend about 40% of their time working on grant proposals. In 2015, the NSF merit review process consumed 360 person-years of scientists’ time. In such an environment, where the effort devoted to the grant-chasing machine can outweigh the quality of new research, the value of spending more money may actually be a net negative.
So how should government fund R&D? There are many adventurous policies we can try, like experimenting with iterations of the DARPA model (which the bill purports to do), starting new Manhattan Projects, or decoupling from universities. I haven’t addressed the thorny issue of how federal funding affects intellectual property rights (which Gerald Barnett’s blog discusses at length). These are ideas worth returning to.
But government being what it is, we cannot simply tear down the edifice and start anew. And political courage being what it is, we need to ask not for bold changes, but for the most modest changes that will yield the greatest improvements. Fortunately, research by Pierre Azoulay, Joshua Graff Zivin, and Gustavo Manso on the Howard Hughes Medical Institute shows that changes of this nature can indeed boost scientific output.
Here are what I see as valuable proposals, which can be realized with only minor changes to the current order of things. They’re far from clever, but that’s by design; the point is that they’re easy to try. Most have already been suggested by others.
- Fund prizes, not just grants. Competitive prizes are given for new accomplishments, whereas grants are given for the promise of new accomplishments. Prizes create a different set of incentives, and can encourage more risk-taking. A famous example is the British longitude rewards, offered in the 18th century to find a way for ships to measure longitude. Modern examples are XPRIZE contests and the Millennium Prize Problems.
- Fund people, not just proposals. “People, not projects” is the funding philosophy used by Howard Hughes, to great success. As studied by Azoulay and his coauthors, Howard Hughes investigators tend to produce more impactful work than their NIH peers.
- Grant long. Many professors lament the short-sightedness of industry, but university research today is even more short-sighted, built around the yearly grant-chasing ritual. Imagine how that might change with, say, seven-year grants. Long-term funding for scientists and laboratories allows for long-term research.
- Grant fast. One of the most successful projects in response to the COVID pandemic has been Fast Grants, launched by Cowen. Fast Grants are $10,000 to $500,000, and decisions are made in under 14 days. By comparison, NSF grant decisions can take six months.
- Grant by partial lottery. A first-stage lottery—you enter a lottery to win permission to apply for the grant—would save hundreds of person-years of scientists’ time wasted on fruitless applications. A second-stage lottery—proposals of sufficient merit are entered into a lottery, and the lottery then awards the grants—would hedge against groupthink and institutional bias in the merit review process.
- Grant without open applications. The Gates Foundation, the largest private philanthropy in the U.S., seeks out proposals from candidates that are first identified internally. The MacArthur Foundation awards its “genius grants” with no prior input from the recipients. This can save time and money, and forces grantors to refine their strategic priorities.
- Fund independent grantors. Instead of deciding grants by committees of professors with ties to the NSF, give an independent scientist a tranche of, say, $5 million and, excepting requirements to avoid conflicts of interest, total autonomy to make idiosyncratic bets on other scientists. Awarding grants by committee means building consensus, which necessarily filters out projects seen as too risky or ambitious.
- Use broader criteria for evaluating scientific impact. Many criteria used by reviewers for allocating funds, like citation count and impact factor, favor conservative research that can be reliably published. New criteria like edge factor have been suggested to better gauge novelty and unconventional thinking.
- A/B test it all. Any of the above suggestions may be pregnant with unintended consequences. (You've probably imagined several.) If we were rational about improving the state of scientific progress, we would try not just one of the proposed changes, we would try all of them in a controlled manner and compare the outcomes.
When it comes to funding R&D, and building the institutions that can successfully organize for technological progress, we’re discussing not only innovation itself, but the art of bringing it about. What do we call this? I’ve heard “innovation policy” and “science of science.” I’m partial to a term suggested by historian Anton Howes: heuretics. Heuretics, or heuretica, comes from the seventeenth century and is defined as the “art of invention, teaching how to find the new, and to judge the old.”
We need to be open to the idea that our current R&D complex, the postwar Endless Frontier of Vannevar Bush—grant-driven, flush with government money, managed and overseen and peer-reviewed—has actually failed to outperform the emergent order it succeeded—the cash-poor, informal, independent research universities of yore, funded by a constellation of private largesse. The limiting constraint on American science isn’t money, it’s heuretics.