global incubators of innovation Small, unexpected connections between method-curious researchers can prove important. Ernst Ruska, who created the groundbreaking electron microscope (Box 1.1), had interactions with Max von Laue, the inventor of x-ray diffraction (Box 1.2). Kary Mullis designed the extraordinary PCR method while working at the biotech company Cetus whose co-founder was Donald Glaser, the pioneer behind the bubble chamber (particle detector). Glaser was intrigued by the potential of molecular biology that was making exciting new findings, so he made the rare shift from physics to establish this biotech company (Box 1.3). Remarkably, Arne Tiselius developed electrophoresis using the ultracentrifuge that his doctoral supervisor Theodor Svedberg recently designed, who allowed Tiselius to pursue his own independent research. And later, Richard Synge—the co-inventor of partition chromatography—spent a year researching with Tiselius (Box6.1). These are not footnotes in the history of science—they are catalytic tools that each enabled dozens of discoveries. These few exceptional inventors did not directly work together, but what unites them is that they took place within the same environment that could inspire focusing on new techniques. These few researchers, with a methodological eye, deviated from conventional research paths—and in doing so, they reshaped the course of science through their new tools. Imagine what would happen if such extraordinary, unexpected interactions were not the rare exception, but systematically cultivated and the norm. There is an enormous untapped potential of triggering innovations in tools: we need to begin intentionally creating dedicated spaces within and beyond universities around the world where method-focused minds from different fields, backgrounds and toolkits can converge to invent, merge, adapt and share new methods—what we call here methods labs and methods hubs. Just being embedded in a methods hub—surrounded by researchers designing the next generation of instruments—can inspire synergies and breakthrough innovations, even among researchers working independently. By examining science's major methods and discoveries, we reveal that Cambridge's Cavendish Lab stands out as the best and closest the world has come to a true methods lab. In just a few decades in the early 20th century, it produced a cascade of world-changing inventions. At the Cavendish Lab, Charles Wilson invented the iconic particle detector (cloud chamber) in 1911. Francis Aston constructed the mass spectrograph in 1919. Patrick Blackett developed the Wilson cloud chamber in 1932. John Cockcroft and Ernest Walton built an improved particle accelerator that same year. Yet it does not stop there. Pyotr Kapitsa then devised a method to produce liquid helium at scale in 1934. And Martin Ryle designed an enormous radio telescope system in 1954. These innovations did not emerge by chance, but were engineered in an exceptional method environment. J.J. Thomson, the Nobel-winning physicist and director of the Cavendish Lab, mentored Aston, whose mass spectrograph was built on Thomson's earlier 1913 prototype. 174 T HE ENGINE OF SCIENTIFIC DISCOVERY Across the Atlantic, Bell Labs later served as another hotbed of innovation. In the span of about four decades, its researchers transformed key parts of science and technology: Shockley, Bardeen and Brattain invented the groundbreaking transistor in 1947. Arthur Schawlow engineered Doppler-free spectroscopy in 1958. Jack Kilby pioneered the microchip in 1959. Willard Boyle and George Smith conceived the CCD sensor in 1970. Steven Chu developed Doppler cooling in 1985. And Arthur Ashkin devised optical tweezers in 1987. Other smaller examples include Scripps Research in California, where Benjamin List invented an environmentally friendly tool for constructing molecules— organocatalysis—in 2000, and where Barry Sharpless developed click chemistry in 2001. And at Berkeley, Ernest Lawrence built the first particle accelerator in 1929, and Luis Alvarez created the hydrogen bubble chamber in 1959. All of these innovative scientists won a Nobel prize for these breakthrough inventions. The Cavendish Lab—peaking from the 1910s to the 1950s—and Bell Labs—from the 1940s to the 1980s—were not just two productive labs, they were the two most prolific methods incubators in the history of science. Remarkably, none of these institutions was founded with the goal of advancing scientific methods. Yet they became, for a period in scientific history, the world's leading methods powerhouses—without a systematic, science-wide understanding of how transformative tools speed up progress. These innovations made in the Cavendish Lab—a part of Cambridge's department of physics—and in the Bell Labs were all made by physicists—except for Jack Kilby, an electrical engineer. But the vision of a network of methods labs worldwide is far broader: it is not confined to physics or limited to any specific field, but spans across science and connects cross-disciplinary toolmaking and infrastructure. Today, no such methods powerhouses exist. The Cavendish Lab, Bell Labs and also Uppsala Lab—described in Box 6.1—are the nearest we have gotten but are field-specific exceptions of the past. Yet they made clear that the invention of better tools can be just as transformative, if not more, than the theories they later uncover. Nearly all labs today are problem-driven or data-driven—but tool-driven labs offer a very different kind of power: they multiply and extend what scientists can do in tackling problems and analysing data. Just as we build research centres around scientific topics—like climate change or cancer—it is time we equally build methods centres: incubators of tool innovation designed to accelerate discovery. The goal of methods labs would be to speed up the creation of cutting-edge tools by fostering a flexible environment of innovation—by merging, adapting and testing methods and tools in new ways to solve today's challenges and anticipate tomorrow's. Think of methods labs as method accelerators where researchers can freely explore hybrid approaches, prototypes and experiment with new tools across disciplines more strategically and rapidly. These labs would deviate from traditional, disciplinary structures, where scientists are often constrained by existing methods and norms within their fields. These toolmaking labs would be where breakthroughs often begin— before the experiments and findings can even start. Here we envision how such labs could be organised—places designed not just to answer scientific questions but to invent the tools that open entirely new questions and answers in the first place (see Figure 6.5). Core goals Invent, recombine, adapt and reinvent methods and tools to unlock unanswered questions Tackle current method bottlenecks that slow or stall discovery Design tools for problems not yet fully conceptualised or understood Place tool-making at centre stage as a scientific pursuit in itself Inside a methods lab Research focus Innovative method-and tooldevelopment and rapid prototyping Tool-centred problem-solving as a scientific method in itself – creating tools enabling discoveries Merging tools – for example digital and physical tools (like AI and optics) Cross-disciplinary problem-solving (like biology with engineering tools) Anticipating unmet scientific needs through new technologies and methodologies Researchers Toolmakers and methodologists Statisticians and data scientists Computer scientists, AI and machine learning specialists Field-specific researchers, engineers, innovators Metascientists studying possible novel pathways for discovery Dynamic environment Highly flexible workspace – culture of open-ended experimentation Setting of iterative trial and error, and refinement –with fail-fast cycles Embedded impact evaluation of new tools Interaction with topic-specific labs and platforms to ensure high usefulness of tools 'Tool programmes' and 'tool fellowships' Central elements Streamline methodchains and toolchains spanning end-to-end method design – from concept to application Experimental sandboxes for high-risk, high-reward projects and prototypes Approach problems from all angles – integrate experimental, computational, data-driven perspectives Design 'recombination tools' –built to plug into and extend other techniques across fields Figure 6.5 Creating methods labs as powerhouses of innovation and discovery The concept of methods labs is developed here by synthesising—and expanding on—patterns observed in the methods and tools that have been created to trigger science's over 750 major discoveries. Ultimately, it is difficult to even imagine trying to do research or make a discovery in genetics without methods like CRISPR or PCR, in archaeology without methods like carbon-14 dating, in neuroscience without tools like MRI, in particle physics without particle accelerators and detectors, in epidemiology, public health and applied economics without statistical modelling and controlled experimental methods. This logic applies across all fields of science. This makes the necessity for methods labs that expand our tools of discovery clear—and long overdue. In short, establishing global networks of methods labs would not just help speed up the pace of discovery— it would redefine how discovery can happen strategically by turning today's method limitations into tomorrow's possibilities of discovery. Box 6.1 A small lab in Sweden enabling many discoveries by developing the ultracentrifuge and electrophoresis—one of the world's first methods labs It is no coincidence that two of science's ten great method advances were invented in the same lab—a modest lab at Uppsala University in Sweden. We now zoom in on this small lab that made many groundbreaking discoveries possible—before zooming back out on the broad patterns across science. The story begins with 176 T HE ENGINE OF SCIENTIFIC DISCOVERY T heodor Svedberg, a chemist who studied at Uppsala and stayed there for his entire career. What inspired Svedberg to build his new instrument was the recently created ultramicroscope used to study colloids—leading him to call his instrument the ultracentrifuge. The ultramicroscope, designed by Richard Zsigmondy, enabled studying particle sizes in colloids beyond the reach of light microscopes. Zsigmondy won the Nobel prize in 1925 for developing the ultramicroscope and his research on colloids applying it. Yet the ultramicroscope too had limitations and Svedberg set out to tackle them. In 1924, he constructed the ultracentrifuge, a powerful tool designed to investigate and measure such microscopic phenomena in far greater detail. He published this breakthrough in a seminal article in the Journal of the American Chemical Society, earning him the Nobel two years later, in 1926. How the ultracentrifuge works is by spinning samples at over 40,000 revolutions per minute. By generating enormous centrifugal force, this ingenious invention enabled, for the first time, to separate and study for example blood, proteins and cells and calculate their molecular weight with unprecedented accuracy. To visualise how the ultracentrifuge works, think of a milk separator: as milk spins, the heavier components (skimmed milk) are pushed outward and separated by centrifugal force while the lighter components (cream) remain in the centre. This same principle is also how plasma is separated from blood. The potential of this new extraordinary tool to advance biochemistry was quickly recognised. Svedberg then obtained one of few grants available at the time to help further expand his research, receiving 25,000 Swedish crowns in 1924 (about 123,700 US$ in 2025 prices). In Svedberg's lab, researchers did not devote their work to one branch of chemistry or biology, but rather developed methods and tools to separate molecules—a true pioneering lab for method advances far ahead of its time. The lab was extremely progressive, building the tools that would unlock the secrets of life's most complex molecules. It uniquely treated methodology itself as a central research goal. Just one year after Svedberg crafted his groundbreaking ultracentrifuge, Arne Tiselius—at only 23—began working in Svedberg's lab in 1925 (Picture 6.1). The new ultracentrifuge inspired Tiselius, his doctoral student, who devised a transformative new method to separate and study proteins using electricity: electrophoresis. The principle is deceptively simple: place a charged molecule in a fluid and apply an electric field; the molecule moves. He developed this ingenious and simple device at a low cost and consisted of a U-tube, a thermostat and a microphotometer (Picture 6.2). He published this invention in his 1930 PhD thesis, The moving boundary method of studying the electrophoresis of proteins. These extraordinary tools sparked discoveries from lysosomes using the centrifuge, to DNA sequencing using electrophoresis. This small Swedish lab has had an enormous impact on scientific progress. This method-driven research environment attracted the attention of the Nobelist Richard Synge—the inventor of partition chromatography, another one of science's top ten tools. And Synge T s pent a research year with Tiselius at Uppsala in 1946 (Box 5.2). The early pioneering methods lab—where tools, not theories took centre stage—has unfortunately remained an exceptional rarity in science. Yet in 1958, another nobelprize-winning instrument, electron spectroscopy for chemical analysis, was also born at Uppsala by Kai Siegbahn. Though they did not collaborate, he was possibly inspired by the legacies of Svedberg and Tiselius in developing cutting-edge instruments at this institution. Picture 6.1 Tiselius in the lab. Reproduced from Uppsala University Library via Alvin. Picture 6.2 The device Tiselius created for his electrophoretic research. Reproduced from Hjertén 1988.
and tools—the making of discovery-triggering tools We now turn to what is the most overlooked force behind scientific progress: methodmaking and tool-making. With this new method-to-discovery principle, we can begin to strategically plan and target ways to accelerate science: rather than waiting for opportunities to arise, we can actively develop new tools and adopt them from other fields. Innovation in methods follows two primary approaches. The most common strategy is to extend our tools, recombine them in novel ways or invent entirely new tools—with capacities never before achieved. The other strategy is to scan other fields for tools to tap into to solve problems in one's own field. Yet because science still does not have the infrastructure set up to systematically foster tool-building (methods labs 178 T HE ENGINE OF SCIENTIFIC DISCOVERY and hubs around the world) and rapidly spread knowledge about new tools across fields when they emerge, most breakthroughs continue to be made in an ad-hoc way. There is more chance and less design in discoveries than necessary. (Those readers more interested in discoveries than how we design methods can skip this section.) So what general steps increase our chances of breaking new ground? To answer this, we scan science's over 750 major discovery-making studies to identify patterns in how method innovations behind them came into being. We uncover eight key recurring pathways of how we create new methods and tools. To bring these to life, we illustrate these general pathways using what is often seen as the leading method across the medical, behavioural and economic sciences: randomised controlled trials (RCTs). This powerful experimental method randomly divides individuals into a treatment or control group to isolate and test the causal impact of for example a treatment, drug or policy intervention. To date, more than two million RCTs have been conducted worldwide—on topics from vaccine safety to poverty alleviation policies. So what are the eight pathways we can take to design future breakthroughs? One, we can search other scientific domains for toolsthat we can leverage and repurpose in our own field. Strikingly, the RCT method, first conceived in medicine in 1948, was only adopted decades later in neighbouring disciplines like economics and psychology. When the method was finally imported in the early 2000s, it sparked a methodological revolution—shifting these fields away from looser observational correlations towards tighter causal precision. This method transplant forced entire disciplines to recalibrate their standards of evidence. Two, we can extend our tools by combining new features from tools within the same scientific domain. In RCTs, we link blinding techniques with placebos and other controls—design features that drastically reduce bias. Three, we can expand our tools by combining new features from tools in other scientific domains . In RCTs, we can randomise the entire sample of participants before the intervention even begins—a common technique in economics but not yet in medicine although often feasible. Pre-randomising participants ensures more reliable results by balancing background influencers between the trial groups. Four, we can extend tools by refining a feature . In RCTs, we can adopt not only common double-blinding (where participants and clinicians are unaware of who the treatment is assigned to), but also employ triple blinding or even full blinding. This ensures that all individuals involved in a study are blinded, including the data analysts and administrators. Nobody then knows which group the participants are assigned to as each individual can unconsciously influence and bias results at each level. Five, we can expand tools by developing an entirely new feature. Rather than testing just one intervention at a time, RCTs can compare multiple treatments side by side. Imagine a trial comparing increased exercise, improved diet, not smoking and a medical treatment simultaneously. This allows researchers to evaluate not just whether interventions work, but also reveals which combinations are most effective in improving our lives. Six, we can invent a completely new tool through major innovations (at times in pathways two to five). Remarkably, the RCT method itself is such an invention: an extraordinary convergence of design features into a powerful, aggregate methodology. Imagine the AI-driven RCTs we can conduct: patients could be T randomly assigned for example to either receive a diagnosis from a traditional physician using conventional tools, or from a physician using an AI-assisted diagnostic system—trained to detect likely disease patterns using data from thousands of similar cases. The trial would not just test the treatment, it would test the tools of diagnosis themselves. Seven, we can critically identify methodological constraints, assumptions and biases of our tools—and then design strategies to reduce them. These strategies in RCTs include testing a new treatment against both a placebo and a conventional treatment to improve the relevance of the findings. They involve fully reporting both the treatment's positive and negative effects and clearly assessing how generalisable the findings are—how the results apply beyond the trial context. Eight, we can detect cognitive, sensory and social constraints and biases we face—and then devise new techniques to minimise them. In RCTs, we apply experimental controls, randomisation, intention-to-treat analysis, and blinded peer review—each helps tackle deeper human error and bias. Ultimately, through continual methodological innovations, the RCT method has become one of science's most remarkable methods for protecting our lives and improving our health—by assessing medications, cancer therapies and countless other interventions. These strategies reveal a powerful insight: method innovation is not a rare cognitive moment, it is a systematic process. When we treat method-making as a central research goal, we can actively design discovery itself. There are countless opportunities to stretch and upgrade nearly all existing tools through vast combinations of features across disciplines. The sheer number of unexplored method configurations is enormous—especially when we think of the vast range of techniques across the experimental, statistical and computational sciences. Entire domains are unlocked by rethinking how we configure the tools we already possess—and new tools we design. Yet any path we take requires shifting our focus to leveraging tools in new ways that open new avenues of exploration (Figure 6.6). New discoveries and fields 1) Cross-domain adoption: scan other scientific fields for new tools to adopt in one's own field. 2) Intra-disciplinary combination: merge new features from tool within the same scientific domain. 3) Cross-disciplinary combination: integrate new features from tools in other scientific domains. 4) Feature upgrade: refine a tool's feature. 5)Feature development: design an entirely new feature for a tool. 6)Tool invention: create a completely new tool through major innovations (sometimes in measures 2–5 and other times independent of them). 7)Method constraint reduction: identify blind spots, assumptions and biases built into our tools – and create new features to minimise them. 8)Cognitive constraint reduction: detect our cognitive, sensory and social constraints and biases – and develop new features to decrease them. Ways we drive science: Figure 6.6 Practical guidelines for how we extend our methods and tools: Eight pathways The eight pathways were identified by screening the over 750 major discoveries—and they can overlap in practice. For pathway six—tool creation—we commonly invent a new tool with an objective in mind. But in exceptional cases, tools are built with no direct aim or necessity at first, created as ends in themselves. The maser is such an example but this 'purpose-free' tool soon found wide application in science. 180 T HE ENGINE OF SCIENTIFIC DISCOVERY These eight practical strategies are not just a map of how past discoveries have been sparked—they serve as a roadmap for future breakthroughs. Any young student or established researcher can use these strategies to drive innovation in their work. Gaps in our tools do not just stifle progress, we find that they are its core inspiration. Because the limits of our tools largely determine what we can achieve experimentally and theoretically, it is crucial to tackle our tool constraints—and do so more efficiently. There is vast potential here: if researchers begin adopting this mindset, tool-making may emerge as the most powerful meta-method in science and discovery. It would shift how we view scientific innovation: from sudden to systematic inspiration—from intuitive to intentional design. Nearly all fields can benefit from integrating general-purpose methods—for example new machine learning techniques, advanced statistical methods or big data analysis. These methods can generally be accessed freely by researchers anywhere in the world with a computer and training. Some breakthroughs hinge on coupling cutting-edge computational and machine learning methods with established tools, from microscopes and spectrometers to x-ray scanners. These recombinations hold vast potential to generate new perspectives. Yet this requires computer scientists, data scientists, experimentalists, microscopists and the like at times to collaborate not to analyse findings, but to expand our scientific toolbox that generates new findings. Scientists need to think of themselves not just as problem solvers, but as method architects—designers of the very tools that shape their ability to know. Yet some new tools can face some resistance. Some encounter ethical constraints, others involve much technical complexity and still others are entangled in economic or political hurdles. A striking example is CRISPR, the extraordinary gene-editing method that some scientists are hesitant to adopt for certain purposes—raising ethical fears over possible 'designer babies' and irreversible ecological impacts. Other powerful tools demand greater training that can affect the pace of discoveries. Advanced computational methods and super-resolution microscopy promise greater efficiency, but their complexity can slow their adoption. Tools are transformative when more than a handful of scientists can use them. Some tools can be further influenced by commercial interests. New imaging instruments, CRISPR and other tools tied to patents or licensed by private companies can limit who can leverage them. Still other tools depend on cross-disciplinary collaborations, like machine learning and AI methods that bridge the worlds of computer scientists and statisticians with biologists, chemists and climate scientists. Take the remarkable AI programme AlphaFold that required extensive interdisciplinary collaboration and transformed our ability to predict protein structures in biology and biochemistry with unprecedented accuracy. It has also accelerated research in drug discovery and genomics. Creating advanced large language models (like ChatGPT) also involves collaborating across multiple domains and enables rapid processing and synthesising vast amounts of information. And of course, while scientific discoveries and tools have vastly benefitted our lives, some have unintended risks—like contributing to environmental degradation and climate change. Others also raise fears about some unforeseeable consequences of using AI tools in science. These may contribute to automation and job loss—not just in manufacturing, but in research and education—and to some feeling more socially T isolated as machines can replace human interaction and even human judgement. To navigate these challenges, regulation and ethical oversight are essential. Governments and scientists must ensure the tools we build serve the public good. We later unpack the actual constraints holding back the top ten tools of science— and how tackling them can vastly expand the research frontier (Chapter 11). It goes without saying that designing methods—like doing science itself—is often a process of trial and error. Many initial methods, when first developed, may not be useful enough or too limited in scope to survive and spur advances. Just like experimental results and theories, we only see methods that make it into publications. So, when are new methods most useful and have the biggest impact? It is when they solve a critical methodological bottleneck—visualising or measuring something we could not before. When they significantly improve the efficiency or precision of our current methods, unlocking completely new perspectives, questions and even entire research areas. And when we can apply and scale them across disciplines, catalysing insights at the intersection of fields. New machine learning methods, electron microscopes, advanced statistical methods, etc meet all of these conditions.