Dear Aventine Readers, 

This week we’re looking at efforts to make the act of scientific research more productive. By many measures, the pace of scientific breakthroughs has slowed in recent decades. And the lengthy review process required to validate new scientific findings is notoriously slow. As a corrective, several governments, including the US, are embracing a practice known as metascience, which basically boils down to studying the scientific process the way you might a business: find inefficiencies and fix them. It has its critics, and is being introduced alongside budget cuts. Whether or not it succeeds will be hard to measure, but supporters of the practice believe it will allow researchers to do more with less, and do it faster. 

Also in the issue: 

Until next week,

Danielle Mattoon
Executive Director, Aventine

Subscribe

Subscribe to our newsletter and be kept up to date on upcoming Aventine projects

In the name of making the process of scientific discovery more efficient, a handful of governments — including the US — are embracing a potentially effective but untested approach: turning the process of scientific inquiry on the process of science itself.

On September 10, the National Science Foundation (NSF) proposed the creation of a new Office of Metascience. Part of a larger effort to double the impact of each federal dollar spent on science by 2036, the new office is designed to help the agency become a “self-improving institution” that “tries new approaches” and “measures their effects.” The National Institutes of Health (NIH) is creating a new division, too — the Office of Research Economics, Planning, and Analysis — to carry out similar work. 

And the US is not alone: The British government established a UK Metascience Unit in 2024, and Japan and Ireland recently committed to similar approaches. Moonshot factories launched in recent years, such as the NSF’s X-Labs initiative and the UK government’s Advanced Research and Invention Agency, are also seen by some experts as an offshoot of metascience.

The initiatives come at a time when, by many measures, our ability to generate scientific breakthroughs has declined dramatically. Sustaining the rate of improvement in computer chip advances now requires 18 times as many semiconductor researchers as it did in the 1970s, for example. The number of new drugs approved per billion dollars spent on R&D has halved roughly every nine years since 1950. It is not helped by increasingly burdensome bureaucracy, which can make securing funding and practicing science itself far more time-consuming than ever before. Metascience is meant to be corrective. It’s about “how [we can] be more effective, more innovative, drive the pace of scientific progress more quickly,” said Stuart Buck, a longtime supporter of metascience who runs the Good Science Project, a nonprofit think tank with the stated mission of improving the funding and practice of science.

If metascience succeeds in its mission, the societal benefits could be significant: more breakthroughs, sooner, for less money. It could supercharge all aspects of science, from drug discovery to energy generation. But the approach is untested at scale, and establishing dedicated offices inside governments, with budgets and the authority to change how billions of dollars are spent, is a radical experiment with no sure outcomes. In other words, there are risks involved, and it will likely take years or even decades to determine whether the endeavors have been worthwhile.

The rise of metascience

The science of studying science began in the 1930s. But while early work was primarily observational, metascience is “both academic and activist,” said David Peterson, a professor of sociology at Purdue University. Roughly defined as the practice applying scientific scrutiny to the work of science in order to make it more reliable and efficient, the approach began to crystallize into its current form in the 2010s partly as a result of the replication crisis in the field of psychology. That’s when it became clear that many peer-reviewed studies could not be reproduced due to small sample sizes, questionable statistical analysis and a publishing culture biased toward surprising results. More recently, metascience has been popularized by what’s become known as the progress studies movement, inspired by a 2019 article in The Atlantic titled “We Need a New Science of Progress.” Co-authored by Patrick Collison, the CEO and co-founder of Stripe, and Tyler Cowen, the economist, it sought to determine whether society was paying a reasonable price for scientific breakthroughs.

The UK Metascience Unit, while young, is an early example of what the approach looks like in practice. One of its early projects has been to change the traditional peer-review process for grant applications. Historically, a small panel of experts evaluates all the applications for a specific grant; the new experiment requires that every applicant for a grant also be involved in reviewing them. So rather than a small team of experts reviewing hundreds of grant applications, each applicant reviews a handful, spreading out the work and leveraging the applicants’ desire for the process to move quickly. 

Early studies of the effort show that the approach shortened the decision-making process by three months, or by around 60 percent. Based on the success of those experiments, UK Research and Innovation, which invests £9 billion of public money into research each year, is adopting the approach across more of the funding councils it oversees. 

It is still too early to know exactly what the NSF and NIH initiatives will look like, though Buck at the Good Science Project co-authored a list of potential reforms they could adopt, including ideas like directing funding toward individuals instead of research and pitting research teams with differing approaches to a similar problem against each other, known as "adversarial collaboration.” We do know that the NSF will investigate new approaches to grant-making, and its announcement memo discussed experimenting with a so-called Golden Ticket approach, by which a single grant reviewer (using his or her metaphorical ticket) can choose to champion a project without panel consensus. The NIH initiative, meanwhile, will include a strong focus on replication and reproducibility.

Success will be hard to measure

There are open questions around how this sort of scientific reform should be undertaken and how its impact is measured.

One problem is that many metascientific experiments have only been conducted at a small scale. “[Demonstrations] are still at the micro level,” said Brian Nosek, the co-founder and executive director of the Center for Open Science, a non-profit supporting transparency in scientific research that is a strong advocate of metascience. That makes it almost impossible to know whether interventions to improve efficiency will scale up, because the norms in, say, experimental psychology are very different from those in computer science. “Standards of evidence, standards for justification, data collection, replication — everything differs widely from field to field," said Peterson. 

It’s also hard to measure the ultimate impact of an intervention at scale. Once a funder backs one line of research rather than another, there's no way to know where the road not taken would have led. And the outcomes that matter most — consequential breakthroughs — are rare and unpredictable. The result is that the overall impact of metascience is "really hard to measure," said Matt Clancy, who runs the abundance and growth fund at Coefficient Giving, a nonpartisan philanthropy that donates heavily to metascience projects. "We're not going to be able to have a clean [randomized controlled trial]-type result to assess whether [an initiative is] working," he said. 

A Trojan horse for budget cuts or needed disruption? 

The adoption of metascience at the National Science Foundation is part of a broader attempt to reform US science, described by the Office of Science and Technology Policy Director Michael Kratsios in a July report called “Science: A New Golden Age.” Zoe Lofgren, the ranking member of the House Committee on Science, Space, and Technology reflected critics’ concerns when she called it a way to “reallocate funding from congressionally approved programs to White House pet projects” in a public statement.

Skeptics also point out that the same budget request proposing NSF's new metascience office (it has not yet been approved by Congress) also cuts the agency's funding by over 50 percent. If the requested cuts are approved in the 2027 budget, discretionary spending for the agency — which currently provides roughly a quarter of all federal backing for nonmedical fundamental science and engineering research at US colleges and universities — will have gone from $9.06 billion in fiscal year 2025 to $3.96 billion next year. 

Defenders of the metascience approach argue that it will allow organizations to do less with more. Clancy said that both the NIH and NSF serve an important role, and that we should not aim to upend every process inside the agencies. “But like any large organization, there's no reason they can't get better over time,” he said, arguing that metascience would be an important part of that improvement. “There's an infinite frontier to fund and limited budgets,” said Ben Steyn, co-head of the UK Metascience Unit. “Metascience is about making those budgets more efficient and productive, so it's good in the good times or the bad times.” 

Boosters also think metascience could help people think outside the box. Science agency staff are under so much pressure to stay on schedule and avoid mistakes that they often can’t stop to think about how to do things differently, one former federal science official told Aventine. A unit specifically tasked with experimentation sitting close to the executive leadership of the NSF or NIH signals to everyone else that experimentation is desirable, the former official said.

Many of the experts Aventine spoke to are hopeful that the new units at the NIH and NSF would have a positive impact. Buck is "cautiously optimistic" about the initiatives at the NIH and NSF. "There are a few people who are influential and whose agenda is just cutting science," he said, but there are others in the administration who he believes have a "good faith intent to improve science." His hope, he added, is that "the folks with that view can prevail in the end." 

Rendering of a Revoy-enhanced semitruck Credit: Manufacturer

How to turn a diesel semitruck into a hybrid EV. There are an estimated 4 million heavy-duty freight trucks in the US, each covering on average about 60,000 miles a year. If some of them could be fueled by electricity rather than diesel, it could save operators money and cut some of the associated 300 million tons of CO₂ emitted annually. But buying a new all-electric semi, such as Tesla's, is expensive and could mean retiring a still-working diesel unit. So several companies have a new idea: Turn existing semitrucks into hybrids. One approach, developed by San Francisco startup Revoy and reported by IEEE Spectrum, is a detachable unit, packed with batteries and its own electric-powered axle, that sits between the cab and trailer. This, according to the company, would provide vehicles with enough electric power to travel 250 miles or so, potentially reducing diesel consumption by 85 percent. The limitation is that drained units must be swapped for charged ones at specific roadside hubs, restricting trucks to routes where Revoy can roll out the required infrastructure. Another approach, being developed by startups including Range Energy in Mountain View and Trailer Dynamics in Germany, is fitting trailers with batteries and electric motors so that they partially propel the semitruck as they are being pulled. Another IEEE Spectrum story from earlier this year explained that this approach, which is more flexible because the trailer can be recharged like a conventional EV, could cut diesel use by between 15 and 40 percent, depending on battery size. There are caveats. The extra weight of batteries and motors eats into payload, ruling out these systems for the heaviest loads. And they are still either in testing or early commercial deployment, meaning their profitability is yet to be proven. But with US diesel prices running more than 60 percent above what they were prior to the Iran war, fleet operators may be more willing to experiment.

A mouse with a brain cortex made from human cells. A team of researchers at Stanford University created mice in which nearly half of their brain volume has been replaced with human cells. The hope of such work, as Nature reports, is to build living models of human brain tissue that can be used to investigate neurological conditions. The researchers genetically engineered the rodents so that cells that would have formed the cerebral cortex did not survive, leaving a large portion of brain cavity empty. Then, days after the mouse was born, human brain tissue was introduced into its brain, where it thrived. Over the course of two to three months, the mass of human cells grew almost fivefold, filling more than 90 per cent of the empty space and sending neurons down into the spinal cord. In tests, the presence of human brain cells restored some cognitive function that had been lost. Mice with human brain cells performed as well as healthy mice on a short-term memory test, for instance.The approach shows early promise for simulating brain related disease in humans; when researchers exposed the mice to low-oxygen conditions linked to cerebral palsy in premature babies, the mice carrying human tissue developed motor coordination problems similar to those of humans with the condition. The hope is that down the line, researchers will be able to use this method to investigate other neurological conditions that plague humans and test the impact of drugs more easily in the lab.

Can we tackle rat infestations through birth control? Poisoning rats may be effective, but it's considered inhumane (they typically die from painful internal bleeding) and can have knock-on consequences for other wildlife, as cats and birds can eat dead rodents and ingest the poison. The Atlantic reports that an alternative is emerging: a liquid contraceptive called ContraPest, laid in bait stations instead of poison. Chicago and Boston have tested the approach, and Washington, DC, is now trying it for a second time. Tests conducted by SenesTech, the maker of ContraPest, claim it can bring about a 94 percent reduction in rat activity, though that test was conducted on a poultry farm, not in a city. So far the results from real-world deployment in cities are mixed. The Chicago tests went well, with rat activity down 88 percent by some measures, and are being expanded. The Boston tests, on the other hand, have been shuttered. Washington, DC, first tried ContraPast in 2019 and the results were inconclusive. There are plenty of reasons why the approach might have mixed results. Rats typically congregate around food sources and in places where it’s plentiful they might not be as motivated to drink liquids. Plus, ContraPest requires regular doses to be effective, something that's hard to ensure in city rat populations. The new test in DC, meanwhile, is part of a multipronged approach to rat reduction that also includes poison, so while it will not lend definitive proof of the contraceptive approach, it could point to an alternative hybrid solution. 

Future forests, from Science
3,200 words, or about 13 minutes

Extreme wildfires are increasingly ravaging forests across the western US, often burning so hot that few trees survive to reseed the land. On top of that, global warming may reduce the chance that the same types of trees will thrive on the same land. This can mean that without intervention some forests may turn permanently into shrubland, decreasing biodiversity and carbon storage. A technique known as assisted migration, whereby seedlings from trees adapted to warmer conditions are planted on burned land, could be deployed to avoid such outcomes. This piece explores efforts by researchers from the Colorado Forest Restoration Institute using this approach in the wake of past fires, planting thousands of seedlings from a variety of trees to determine which are best suited to different locations. So far, the results are mixed, depending on the elevation at which trees are planted and the severity of the preceding fire, with seedlings faring better where burns were most intense. In the meantime, the process prompts some questions: Since the intervention costs more than leaving the land as is, what areas should be targeted and who will make that decision? And what are the downstream effects of overriding nature instead of letting it take its course? 

The World’s Most Reliable AI Detector Has a Human Problem, from Bloomberg Businessweek
4,300 words, or about 17 minutes

Pangram has built software that, according to its own testing, can identify text written by AI in 99.98 percent of documents with a false positive rate of roughly 1 in 24,000. Earlier AI detectors flagged suspicious text based on how varied and unpredictable it was. Pangram’s technology uses an AI model trained on documents known to be written by humans alongside paraphrased versions known to be generated by AI. (The model has, over time, become more nuanced, but broadly speaking its core trick remains the same.) Independent researchers have largely backed up Pangram’s results, but the company has nevertheless received sharp criticism. Even with low false positive rates, a single false positive error could unfairly end a writer's career. It probably doesn't help that one of the company’s founders, Max Spero, is combative online, sometimes scanning his critics' writing and publicizing how much of it Pangram flags as AI-generated. New AI writing tools do sometimes get past Pangram, though the company argues it can adapt to them as they gain traction. Ultimately, though, it seems like AI detection will be a constant game of cat and mouse, with both sides wielding the same technology. 

Mark, from Colossus
14,900 words, or about an hour

In a news cycle dominated by the frontier AI labs, it’s easy to forget about Mark Zuckerberg. But this profile of the Meta CEO serves to remind us that he should not be overlooked. Despite scandals, court cases and no shortage of ridicule, Zuckerberg has remained determined to lead the company he founded in 2004. And through interviews with three dozen people — including Zuckerberg's parents, wife and mixed martial arts coach — this captivating (and very long) profile makes one thing clear: He is not going anywhere. About 18 months ago, Meta was a laughingstock due to its Llama 4 AI model, which woefully underperformed its competitors. As a result Zuckerberg overhauled his company’s AI operations and came back with Muse, a model released in early September that has pulled Meta back into the frontier pack. The app of the same name, which allows users to delegate tasks such as shopping and travel booking, has quickly become a consumer success, at one point replacing ChatGPT as the most popular free app in the Apple App Store. The profile reveals Zuckerberg as someone who avoids the apocalyptic framing of AI favored by the CEOs of OpenAI and Anthropic, framing it instead as a quotidian tool billions of people will use.