Some research starts with an obvious destination: a disease to treat, a test to build, a technology to improve. Other research begins much earlier, with a question about how nature works and no clear application in sight. The difference can make the first kind of research seem more useful and the second more difficult to justify. But by the time a scientific problem has a recognizable clinical or technological solution, much of the knowledge required to solve it may already have been built through work whose eventual usefulness could not have been predicted.
That is the central difficulty in evaluating curiosity-driven research. Its value often lies not in answering a question we already know will matter, but in producing knowledge that changes what questions can be asked at all. The path from discovery to application is rarely visible from the beginning, and some of the most consequential advances in medicine have depended on research undertaken before anyone could plausibly have described the treatment, technology, or even the scientific problem that would eventually emerge from it.
KJ Muldoon’s story is a striking example. The first critical step in his treatment was not a gene editor. It was an ammonia test. When KJ was about thirty-six hours old, a neonatologist recognized that something was wrong and checked his ammonia level. The level was dangerously high. He had severe carbamoyl phosphate synthetase 1 (CPS1) deficiency, an ultra-rare genetic disorder. Without functioning CPS1, the liver cannot properly dispose of waste nitrogen, causing ammonia to build to toxic levels.
What followed had an immediate clinical goal no one could miss. A team at Children’s Hospital of Philadelphia (CHOP) and Penn Medicine identified KJ’s exact genetic variant and created a personalized base-editing treatment designed to correct it. KJ received his first treatment at about seven months old. A year later, CHOP reported that he could tolerate more protein, needed less nitrogen-scavenging medication, and was walking and talking.
Real physician voices, twice a week
Free, and one click to unsubscribe.
The therapy was designed and manufactured in roughly six months. But those six months were possible only because decades of scientific work had already laid the groundwork before KJ was born.
Before anyone knew what it was for
To understand how six months of development could rest on decades of science, start with the tool. KJ’s treatment did not begin with KJ. In 1987, Yoshizumi Ishino and colleagues in Japan, while studying a bacterial gene, noticed something strange nearby: short DNA sequences repeated at intervals, with other sequences between them. They published what they saw but did not know what it did. Similar patterns later appeared in other microbes, and Francisco Mojica encountered them while studying microorganisms adapted to extreme salt. He spent years trying to understand them. By 2002, the strange sequences had a name: clustered regularly interspaced short palindromic repeats, or CRISPR.
In 2005, researchers recognized that some of the DNA between those repeats matched genetic material from viruses and proposed that CRISPR might be part of a microbial defense system. Experiments confirmed the idea in 2007. As scientists discovered that CRISPR used RNA made from stored viral sequences to recognize and target matching viral DNA, its engineering potential became clearer. In 2012, Emmanuelle Charpentier, Jennifer Doudna, and their colleagues showed that CRISPR machinery could be programmed to cut DNA at chosen locations. In 2016, researchers introduced base editing, an approach capable of changing individual DNA letters without making the double-stranded DNA breaks used in standard CRISPR-Cas9 gene editing. Less than nine years later, a personalized base editor was being infused into KJ.
Looking backward, the path can seem almost inevitable. It was anything but. The scientists staring at unusual bacterial DNA in the 1980s could explain why the observation was worth studying. They could not possibly have predicted that the observation would become part of the scientific foundation for a personalized treatment nearly forty years later. That future was not simply difficult to foresee; the knowledge needed to imagine it did not yet exist. A scientific question can be worth pursuing even when no one can yet foresee what its answer may eventually make possible.
The same pattern appears elsewhere. In the 1960s, microbiologist Thomas Brock became fascinated by organisms living in Yellowstone’s hot springs at temperatures scientists had not expected life to tolerate. He and Hudson Freeze described Thermus aquaticus. Years later, scientists developing polymerase chain reaction, or PCR, needed an enzyme that could survive repeated heating while making millions of copies of DNA. Taq polymerase, isolated from Thermus aquaticus, could. It helped make PCR one of the foundational tools of genetics, medical testing, DNA sequencing, and forensic science. Brock was trying to understand life in extreme heat. PCR was still nearly two decades away.
Another example came from an unlikely source. Glucagon-like peptide-1 (GLP-1) is a hormone released after eating that helps regulate blood sugar, in part by stimulating insulin secretion. But native GLP-1 is broken down rapidly in the body, limiting its usefulness as a drug. In 1992, John Eng reported exendin-4, a peptide isolated from Gila monster venom that was later shown to activate the GLP-1 receptor and remain active much longer than the native hormone. A synthetic version, exenatide, entered clinical use in 2005 as the first FDA-approved GLP-1 receptor agonist. Modern GLP-1 medicines did not all descend directly from exendin-4, but exendin-4 provided an early demonstration that this biological pathway could be turned into a practical therapy. One branch of that story began with a scientist asking what was in a desert lizard’s venom.
When medicine borrows from other fields
Some of medicine’s essential tools came from fields that were never trying to solve a medical problem. Around the turn of the twentieth century, physicists were asking some of the most fundamental questions: What is matter made of? What is an atom? What are the strange forms of radiation emerging from experiments with electricity and matter? There was no semiconductor industry, no laser, no modern computer, and no medical-imaging industry waiting at the end of those questions. Yet the effort to understand atoms, electrons, and radiation would become foundational to much of modern technology.
One of the first medical applications appeared almost immediately. In 1895, Wilhelm Röntgen discovered X-rays while experimenting with cathode rays. Within weeks, physicians were using them to see bones and foreign objects inside the body without surgery. Within months, doctors were experimenting with X-rays as a treatment for cancer. A line of research into unfamiliar radiation had produced both a new way to see inside the human body and the beginnings of radiation therapy.
In 1917, mathematician Johann Radon solved an abstract problem: how to reconstruct the interior of something from measurements taken through it. Decades later, physicist Allan Cormack independently confronted essentially the same reconstruction problem in the context of radiation passing through matter. Today, a computed tomography (CT) scan takes X-ray measurements from many different angles and uses mathematical reconstruction to create detailed cross-sectional images of the body. That abstract mathematics would eventually help make CT possible, turning measurements taken from many angles into images physicians use every day.
The path to PET was less direct. Paul Dirac was trying to reconcile quantum mechanics with relativity when his equation implied the existence of an antiparticle of the electron. Carl Anderson observed the positron in 1932. A positron emission tomography (PET) scan uses a radioactive tracer that emits positrons inside the body. When a positron encounters an electron, the two annihilate and produce photons. PET scanners detect those photons and use them to map metabolic and other biological activity within tissues. A particle first predicted through fundamental physics had become the basis of a medical-imaging technique. Like CT, PET emerged from work that had little to do with medicine when it began.
When value is hard to recognize
Unpredictability does not always mean waiting decades for an application to appear. Sometimes the destination is already imaginable, but the importance of the science that might lead there is still difficult to recognize. Katalin Karikó spent years trying to make therapeutic messenger RNA (mRNA) work despite limited enthusiasm for the approach. In 2005, she and Drew Weissman published a key discovery showing that modified nucleosides could reduce the unwanted inflammatory response triggered by laboratory-made mRNA. Its significance was not immediately apparent. Fifteen years later, that discovery helped provide the foundation for the mRNA COVID-19 vaccines.
Here, the destination was somewhere in view, but it remained difficult to know which scientific advance would prove decisive. The significance of a result is much easier to recognize in retrospect.
When the answer changes the question
There is another form of unpredictability: Sometimes answering one scientific question reveals a larger one that was not visible at the outset. I have seen that process in my own research.
We began with a narrow biological question: why do some DNA regulatory elements turn genes on or off while many other similar sites do not? Combining experimental models with large-scale genomic data, we found that inversion symmetry determined which DNA elements transcription factors bound and, together with DNA sequence constraints, whether those interactions could initiate or maintain transcription. At the nucleotide level, these effects depended on where variants occurred within a DNA element and at the corresponding reverse-complement positions in that element. The pattern held across all members of the steroid nuclear receptor family in hundreds of experiments from independent laboratories, spanning a wide variety of mouse tissues and human cell lines, and again when we tested p53, an unrelated transcription factor. Together, those findings revealed a broader grammar for how transcription factors control gene activity.
But that result raised a larger question: why was the symmetry there at all? When we widened the analysis from regulatory elements to the genome itself, the pattern appeared on a much larger scale. When we analyzed one strand of the genome at a time, DNA words and their reverse complements occurred at essentially the same frequency, a relationship maintained across individual nuclear chromosomes. In other words, the same mathematical relationship we had first encountered in transcription-factor binding and functionality was also reflected in the larger organization of the genome. From an information-theory perspective, the genome is organized not only by the sequences it contains, but also by relationships among those sequences.
That line of inquiry now extends to the genetic code, where we found a different symmetry in the organization of codon degeneracy, the fact that multiple codons can specify the same amino acid. Here, the molecular system reading the sequence is different. Rather than transcription factors recognizing regulatory DNA, the ribosome and associated translation machinery read codons during protein synthesis. We found that this symmetry acts as an exact reflection of a position-specific decoding-stability score rooted in the chemistry of the ribosomal decoding center. The molecular mechanisms differ from those in our earlier work on genomic reverse-complement symmetry, but the deeper question is similar. In each system, informative structure emerges from the relationship between sequence symmetry and the position-specific variables sensed by the molecular machinery that reads it.
What began as a question about why some DNA elements turn genes on or off while others do not led to much larger questions about symmetry in the genome and the genetic code, questions we could not have known to ask when the work began. That is a different form of unpredictability from not knowing whether a discovery will someday become a drug or technology. Sometimes research changes the scientific question itself. That is why “curiosity-driven” can sound almost too modest a description. It is not simply scientists pursuing whatever interests them. Often it is the discipline of following an answer far enough to discover that the question you started with was only part of a much larger one.
Can usefulness actually be predicted?
These stories also invite an objection. CRISPR worked. Taq worked. Radon’s mathematics found an application. Dirac’s particle became medically useful. We remember these stories because history has already identified the winners. But most scientific questions will never produce a medicine or technology that ordinary people recognize. Research funding is finite, and not every interesting question deserves a grant. If the defense of curiosity-driven research is simply that anything might someday prove useful, it is not much of an argument.
The problem is not just hindsight. Scientists have recognized this uncertainty before the eventual applications were known. In 1940, mathematician G. H. Hardy described relativity and quantum mechanics as, at the time, almost as “useless” as the theory of numbers, then added, “Time may change all this.” He was right. Quantum mechanics would become foundational to technologies including transistors, lasers, and microchips. Hardy recognized that time might change the usefulness of abstract ideas, but even he could not have foreseen how consequential some of them would become. The lesson is not that every abstract idea will eventually prove useful. It is that future usefulness can be extraordinarily difficult to judge from the circumstances in which a question is first asked.
Modern data allow that question to be tested more directly. A 2024 study examined more than 1.2 million scientific papers resulting from more than 90,000 National Institutes of Health (NIH) grants. Seventy percent of the papers fell into at least one scientific or disease category that had not appeared among the original grant’s categories. To make the test more conservative, the researchers then grouped closely related categories together. Even then, 58 percent of the papers contained at least one category not represented among the research categories NIH had assigned to the original grant. Research, in other words, frequently outgrows the scientific categories in which a project began.
But scientific direction is only one form of unpredictability. Another question is whether a project’s eventual technological impact can be foreseen from how basic or applied it appears at the outset. A study following NIH grants over 27 years used patent citations to trace whether research funded by those grants later contributed to technological inventions. It found no systematic relationship between a grant’s basic-versus-applied research focus and its likelihood of being cited by a later patent. Even by that narrow measure, looking more applied at the beginning did not make a grant more likely to contribute to a later patented invention. The same study found that about 10 percent of NIH grants generated a patent directly, while roughly 30 percent produced scientific papers cited by patents. Much of that measured technological influence traveled indirectly, through knowledge that others eventually used. Even within the limited window provided by patent citations, proximity to an application at the outset did not reliably identify the work that ultimately fed technological invention.
The same pattern appears when the outcome is medicine rather than patents. Researchers examined 356 drugs approved by the U.S. Food and Drug Administration (FDA) from 2010 through 2019 and found NIH-funded research connected to 354 of them. They identified about $187 billion in NIH funding associated with those drugs, measured in inflation-adjusted 2018 dollars. Importantly, 83 percent of that funding supported basic research on the biological targets the drugs would eventually act on, rather than research on the drugs themselves. NIH did not develop those medicines; companies and other developers still carried out the essential work of turning scientific knowledge into approved treatments. But the finding shows how often the foundation for drug development is laid by basic research on the underlying biology and the targets that later become therapeutically useful.
Even after a molecule has an intended use, the path can turn again. Researchers examining 26 drugs or drug classes that physicians had judged transformative found that nine, or 35 percent, had been repurposed from products developed for other indications. Ten, or 38 percent, had first been developed for rare diseases before broader uses were discovered.
Elacestrant shows how an unexpected result can redirect a drug entirely. The compound, then called RAD1901, was being tested for menopausal hot flashes. The lowest dose reduced hot flashes, while higher doses were ineffective and could even worsen them. That unusual dose response caught the attention of researchers in Donald McDonnell’s laboratory at Duke University. They found that the drug’s behavior changed as the dose increased: At lower concentrations it could produce estrogen-like effects, while at higher concentrations its ability to block the estrogen receptor and drive its degradation became dominant. That was precisely the kind of activity that could be useful in estrogen-receptor-positive breast cancer. What looked like a confusing result in one indication revealed a mechanism that pointed toward another. In 2023, the FDA approved elacestrant for postmenopausal women and adult men with advanced or metastatic estrogen-receptor-positive, HER2-negative breast cancer carrying ESR1 mutations whose disease had progressed after at least one line of endocrine therapy. Sometimes science does not just fail to predict the answer; it predicts the wrong patient.
That uncertainty does not mean research decisions have to be made blindly. It means the science itself and predictions about its eventual usefulness are two different things. Scientific proposals can still be judged on whether they ask a meaningful scientific question, use a rigorous approach, and provide a credible path to an answer. The distinction is between judging the quality of the science in front of us and assuming we can also know everything its answer may make possible decades later.
What uncertainty requires
If future usefulness is hard to predict, that does not mean every project should be funded. It means that expected practical payoff is an imperfect guide to which research will ultimately prove most consequential. KJ’s story makes the problem concrete. By 2025, the value of developing a treatment for a critically ill infant was obvious. What had not been obvious decades earlier was which studies of bacterial DNA, enzymes, gene repair, delivery systems, or liver biology would eventually make that treatment possible. Those scientific foundations could not be assembled after KJ became ill. They had to exist already.
A broad research portfolio is one response to that uncertainty. The point is not to fund everything, but to avoid building a system in which only research with an obvious near-term application can survive. What is much harder to know is what that research may ultimately make possible. A healthy research system therefore has to preserve room for basic research alongside work with clearer translational goals, rather than requiring every project to justify itself by predicting an application years or decades in advance. The goal is not to identify which grant will become the next CRISPR. It is to build a system broad enough that discoveries no one could have predicted still have a chance to occur.
The difficulty is that the consequences of getting that balance wrong are not equally visible. The cost of supporting rigorous research that never leads to a transformative application is finite and measurable. The cost of failing to support a foundational discovery is much harder to see, because the treatments, technologies, and later questions it might have enabled never come into existence. A discovery that is never made leaves no clinical trial, no company, and no citation trail by which its absence can later be counted.
Maintaining that breadth is not the responsibility of any one sector. National Center for Science and Engineering Statistics estimates for 2024 indicate that the federal government funded about 40 percent of basic research performed in the United States, while the business sector across all industries funded about 34 percent, higher education institutions about 14 percent, nonprofit organizations about 9 percent, and state and local governments about 3 percent. The federal government remained the largest single source of basic-research funding, while universities, businesses, and nonprofits also devoted substantial resources to work whose practical applications may not emerge for years or decades.
That federal role is especially consequential in a highly competitive research-funding environment. At NIH, for example, the success rate for R01-equivalent grants, one of the agency’s main categories of research-project funding, fell from 21.6 percent in 2023 to 13.0 percent in 2025. During those two years, applications rose from 35,072 to 42,022, while the number of new or renewed awards fell from 7,592 to 5,471. Overall NIH spending on R01-equivalent grants, including grants already underway from previous years, rose from about $19.8 billion to $20.8 billion. NIH noted that some of the higher spending per award in 2025 likely reflected, in part, a greater use of up-front funding, with more money for certain multiyear grants committed earlier rather than spread across later years. Even with that accounting context, more applications were competing for fewer new or renewed awards, and the success rate fell substantially.
The significance of that competition lies partly in what the numbers cannot show. They tell us how many applications were funded, but not what might eventually have come from the projects that were not. Funding decisions are made in the present, but some of their most important scientific consequences may not become visible for decades. A future CRISPR or Taq does not announce itself at the moment the decision is made.
What curiosity makes possible
The argument for curiosity-driven research is broader than the discoveries it produces. Research also builds the scientific capacity needed for the next discovery. Scientists learn to question assumptions, design experiments, recognize artifacts, pursue anomalies, and follow unexpected evidence by doing research in which the answer is not known in advance. MIT President Sally Kornbluth has made the point with a simple analogy: “Would you fly on a plane with a pilot who had never flown?” Scientists learn how to do research by doing it.
A PhD is often mistaken for simply more education in a subject. Doctoral research is an apprenticeship in creating knowledge where no answer key exists. A scientist must learn how to identify what is genuinely unknown, turn an observation into a testable question, devise a way to answer it, distinguish a real signal from an artifact, recognize when an assumption has failed, and decide which unexpected result deserves to be followed.
That training matters far beyond universities. PhD scientists identify therapeutic targets, design and develop potential treatments, determine how those treatments work, establish their safety and biological effects, and help carry them through preclinical development, clinical testing, manufacturing, and regulatory review. They are often involved throughout the process that turns a biological discovery into a medicine that can actually reach patients.
That expertise is easy to overlook because the public usually encounters science at the end of the process. We see the drug when it is prescribed, the scan when it is ordered, the genetic test when a result comes back. What is less visible is the scientific work that preceded those products and continued through their development, testing, and refinement. By the time the public encounters the result, much of the scientific process that made it possible has already happened out of view.
This is why losing research capacity means more than losing individual papers or experiments. It can mean graduate students who are never trained, techniques that are not passed to the next generation, expertise that never has time to mature, and future investigators who choose another path before their scientific careers begin. A scientific workforce takes years to build. It cannot be recreated on demand after the need for its knowledge becomes obvious.
A research system therefore produces more than papers, patents, and technologies. It produces people who know how to work at the boundary between what is known and what is not. Those scientists are not merely a byproduct of discovery. They are part of the infrastructure that allows discovery to continue.
That capacity may become even more important as medicine moves toward therapies designed for increasingly small groups of patients, sometimes even a single patient. KJ’s case points toward that future. In February 2026, the FDA issued nonbinding draft guidance for individualized therapies under what it calls the Plausible Mechanism Framework. When a genetic condition affects so few people that conventional randomized trials may not be feasible, the framework describes how developers may build evidence for safety and effectiveness using a well-supported biological mechanism together with clinical, nonclinical, and manufacturing evidence. The framework came a year after KJ received his first treatment and did not govern his case, but it addresses the same challenge of evaluating individualized therapies for conditions too rare for conventional development pathways.
Individualized therapies do not emerge from a single breakthrough. They become possible because the necessary scientific pieces (and the people capable of assembling them) are already in place before a particular patient needs them. In KJ’s case, those pieces included decades of work on genes, enzymes, disease mechanisms, editing, delivery, pharmacology, and manufacturing.
The next breakthrough may already have a clear goal. Or it may look like an odd pattern in bacterial DNA, a microbe living in a hot spring, an equation with no obvious medical use, a strange form of radiation, a drug behaving opposite to the way researchers expected, or a mathematical symmetry in DNA whose larger meaning is only beginning to come into view. Looking backward, it is tempting to imagine that a more prescient research system could have supported only the discoveries that eventually proved transformative and skipped the rest. History does not offer that menu. We know which strange questions proved important only because someone was allowed to ask them before anyone could say what they were worth. The challenge is not simply to fund the research whose value we can already see. It is to preserve the people, institutions, and freedom of inquiry that allow important questions to be pursued before their value becomes obvious. By the time the cure is visible, the science that made it possible may already be decades old.
Laurel A. Coons is a scientist with a background in genomics and biomedical research. She completed her doctoral training in pharmacology and cancer biology at Duke University and conducted research at the National Institute of Environmental Health Sciences. Her work has focused on genomic regulation, endocrine signaling, and translating complex scientific data into insights relevant to medicine and patient care.
She shares professional updates on X at laurelcoons.
