tools to accelerate science AI can be a powerful tool in a scientist's toolbox—amplifying how the world can be explored and predicted. AI systems can help scientists streamline the research process by automating repetitive, time-consuming tasks, making research faster and less prone to some human errors. AI-powered search tools can filter through vast scientific literature, extract information and spot trends across studies—from climate science to genetics. While we can delegate our vision to advanced microscopes to make better observations, we can delegate our mind to AI systems to better spot patterns, make better simulations and predictions and even can help design experiments—faster than scientists can alone. How can scientists combine AI with other tools to help tackle our current barriers? Coupling AI tools with high-resolution microscopes vastly improves imaging by using deep learning to sharpen blurry images and detect patterns. This can accelerate real-time disease diagnosis and provide guidance for surgeries. In astronomy, pairing machine learning algorithms with telescopes enables sifting through vast and noisy astronomical data to spot new exoplanets or classify supernovae by detecting subtle patterns in the data. P Climate change is challenging to fully understand given its immense complexity. Machine learning can simulate climate systems—interactions between the ocean, atmosphere, ecosystems and human activity—more precisely by using enormous amounts of recorded data. It can integrate climate models with detailed local weather data to improve predictions about trends in rising sea levels, melting glaciers and extreme weather. This offers policymakers detailed analysis of the local effects of climate change. In drug discovery, AI algorithms can help recognise hidden patterns in vast medical records and help speed up the search for new drugs by predicting how they may work and what side effects they may cause. Advanced AI now helps sharpen MRI and CT scans, making it easier to spot cancers early and detect signs of Alzheimer's. It helps in reducing some human diagnostic errors. By analysing patient data, machine learning models can help detect relationships between diseases, genes and drugs. This can help personalise and tailor medical treatments to individual needs. A remarkable breakthrough has been achieved in understanding proteins. To understand how proteins function in our body, we need to understand their structure. Predicting their structure enables us to identify proteins that affect diseases and so improve diagnosis and treatments. But establishing protein structures is difficult and time-intensive. The barriers of existing methods were broken with the new powerful AI programme AlphaFold developed by Google DeepMind: a programme that predicts protein structures with extremely high accuracy. It solved the decades-old 'protein folding problem'—and fuels drug discovery and advances disease research. So combining AI and self-improving systems with other tools can optimise research: increase speed, efficiency, scale, sensitivity and resolution. In short: AI can improve how our best tools perform. It is becoming more integral to accelerating innovation in different domains—with large language models also helping to mine vast research for hidden relationships. Building AI-enhanced scientific tools could help spur breakthroughs we cannot yet imagine. We could create ever more advanced lab systems that adapt and design their own experiments— learning from their own results. Yet machine learning systems often work like black boxes: making predictions without explaining how. They can also at times amplify biases built into the data they use—their training data. That means human judgement is essential. These systems and algorithms ultimately rely heavily on humans to be designed, understand how they generate results and detect and avoid errors. New approaches to designing methods and tools could emerge: a taxonomy of scientific methods—mapped out in Chapter 6—could be linked to AI to help screen for tools, suggest underused ones and help select new combinations of methods and tools for both general and specific problems. In other words, we could design AI systems in the future that do not just follow instructions but can learn and suggest new types of experimental methods and tools that push us beyond the borders of what we can currently imagine. In the future, we could also train machine learning programmes to analyse data on newly developed methods and tools (such as newly published methods papers) and explore the domains and problems the new methods and tools could best be applied to in spurring breakthroughs. T HE ENGINE OF SCIENTIFIC DISCOVERY
Since our mind's capacities evolved mainly to survive and reproduce, do we face ultimate limits to what we can know? Is there a ceiling on our understanding the biggest puzzles—like the nature of life, consciousness and the universe? As innovative toolmaking enables opening new layers of reality, the boundaries of discovery are not preset. Instead, we have—method by method—pushed them back. New methods of observing, measuring and experimenting—and new method combinations—have continually expanded what was previously beyond our reach. Our biology, our mind's constraints, our scientific institutions and the time and place we live influence the parts of the world we can access. But they do not dictate the types of tools we can come up with to tackle those very constraints. The world has boundaries. Human life is short and finite. The speed of light is fixed. Our planet's natural resources are limited. But does that mean that science will eventually hit a wall too? Our methods and mind do not seem bounded by clearly defined limits—given the seemingly endless method combinations and our human imagination (Chapter 6). Yet science has settled some of our fundamental puzzles. We will not re-discover the basis of biology (evolution and DNA's double helix), the backbone of chemistry (the elements that make up the periodic table) or the foundation of physics (relativity, quantum mechanics and the big bang). But we keep updating, expanding and better measuring them as we improve our methods. While it is true that the neutron, the DNA molecule, electron microscopy and CRISPR can only be discovered once, these landmark breakthroughs opened entire fields of research that we keep expanding today. But some theoretical sciences may eventually slow down. Once major, successful theories become established—as in physics—establishing other grand theories would become less likely. In other words, continuously uncovering grand discoveries that lay a new foundation for a major field would be unlikely. So we are unlikely to continually developing fundamental theories that govern the world. The theory of evolution for example—underlying biology—is established with independent methods across diverse fields. It is unlikely that a new foundational discovery will make the theory of evolution obsolete. And that new methods and tools drive discoveries and push the frontier is also unlikely to be completely replaced by a fundamentally new, unknown driver of scientific progress—a new driver that does not rely on sharper microscopes, more sensitive detectors and more powerful computation methods. Developing vast new fields that parallel in importance to biology, chemistry and physics is unlikely—just like we are not going to find an entirely new continent on our planet. But we will keep making major discoveries and opening new fields—from quantum biology to machine learning. We are learning to look better inward and outward at once unimaginable scales. At the frontier, breaking new ground unlocks more questions and mysteries, not fewer—in a seemingly endless loop. What we know—at any moment—is largely determined by the tools we have built so far. Our future knowledge will remain on those parts of the world that our methods and technologies allow us to study. For some unknowns, we have not yet developed P the right tools to uncover them—from dark matter to the nature of time. There are also imaginables—things we can imagine but do not yet know how to develop the needed tools and test them, like artificial consciousness or the possibility of finding life on other planets (or even entirely other forms of doing science). We humans are fascinated by foundational questions about distant origins, vast complexity or fundamental structures. Yet are answers to such questions often beyond what we can currently observe and test? Indeed, it is here where science often hits walls and stops working—where we run into the boundaries of what we can discover. A methodological border of science emerges: science can study what is subject to observation, scientific methods and often quantification. This is why it is so difficult to get our grips on the nature of knowledge, consciousness, time, morality, subjective experience, freewill, politics, love and much more. Science also runs into barriers when asking 'why' questions: why do we exist? Why did the universe begin? Why are we conscious beings? Even though our toolbox largely explains our current boundaries, it cannot tell us everything about them. Our frontier also partly depends on what research areas and instruments receive more policy attention, resources and public interest—from public health and cancer research to renewable energy (Chapter 10). Ethical issues and government decisions also partly influence some research questions: what research is legal? Should we edit human embryos? How should AI be used in informing or making medical diagnoses? Should we prohibit technologies and research that can foster climate change? Ultimately, the edge of science looks different in each field—and new fields keep emerging. Yet in the end, those aware of how new tools break open mysteries and those who get their hands on them will be the ones most likely to push the frontier forward.
Exploring science's future can help us better plan, innovate and anticipate it. But the further we look ahead, the blurrier and more uncertain things get—like our methods. The key is that the combinations of methods we can create seem almost limitless, and can open many new paths and fields to explore (Chapter 6). Science is an evolving, flexible system because we do science using our flexible methods and elastic mind. After all, our scientific methods—the ways we study the world—depend on creativity and generate even greater creativity in science that does not exist in the observable world itself. As long as we live, we will keep investigating what affects our health, minds and societies. We will need new cures for evolving viruses, diseases, cancers and pandemics, and new solutions for our changing human and social challenges. So fields like medicine, biology, psychology and economics will keep growing. The same goes for fields like agriculture, engineering, environmental research and computer science—because they improve our lives and tackle the challenges we face. From new forms of agriculture that are better adapted to extreme weather and climate change, to green technology that replaces fossil fuels faster. In short, science that improves our lives—applied science—will likely always grow and not run out of problems to solve. In fact, it makes up most of science. So it does not seem that many parts of science will ever become finite and complete. Just as it was not possible to predict CRISPR, quantum computers, smartphones and the internet a hundred years ago, the same applies to scientific and technological developments in the next hundred years. And we will very likely look back at today's breakthroughs as just the beginning. So will we near the limits of science? The answer depends on the feature of science we focus on: when it comes to method breakthroughs and experimental discoveries, we will very likely keep growing—especially across applied science—while uncovering entirely new fundamental theories and laws of nature will likely slow down. The pace will vary from field to field. Yet it is clear that new methods and problems open further methods and problems—in a cumulative loop. The future of science in any field will fundamentally depend on the tools we come up with. With the past century marked by exponential scientific progress, can we keep growing as fast—or even faster? It seems very likely. But many earlier discoveries were low-hanging fruit because they were relatively easier to uncover. Think of looking through early telescopes to discover new planets and moons, to using the James Webb space telescope to estimate the age of the universe at about 14 billion years. Think of early microscopes uncovering bacteria and blood cells, to using cryo-electron microscopes to map protein structures at extraordinary atomic resolution. Or think of early x-rays methods revealing broken bones in the body, to using x-ray crystallography to determine the detailed structure of DNA and viruses. Many future breakthroughs will come from new method combinations—especially at the crossroads of different fields. Others will need entirely new methods that at times can be somewhat more challenging to invent. At the research frontier, the future growth of science looks very promising for three reasons. First, making new method innovations is still surprisingly not yet coordinated in a strategic way across science—but these scattered breakthroughs consistently trigger new pathbreaking findings. There is an enormous untapped potential of new tool combinations that are almost seemingly endless and within our reach (Chapter 6). Second, the pace of major new discoveries and fields emerging and expanding does not seem to be slowing. From CRISPR gene editing, machine learning, quantum computing and environmental engineering, to nanotechnology, computational chemistry, quantum biology and even the science of science. Third, the past highlights that those in each generation who think we are nearing our limits have been incorrect. The real limit is largely how fast we can upgrade our toolbox. Science is not something we may ever finish, with no more challenges and mysteries to resolve—from mutating viruses and changing ecosystems to evolving language and human systems. It is largely self-renewing. Only since the 20th century did we spark most of science's major discoveries. The future is much longer, holding many more advances through powerful new tool innovations. If we knew all the answers, it would mean we would already know all the questions and have already developed all the needed methods and tools we could ever need. But that is not how discovery works. This methods-powered view offers an alternative, more promising perspective of the future: if scientists commonly only chase the questions they already have, they are held back by what they know and can imagine today. But history shows that developing new methods and tools has unlocked most new questions we could not have predicted across fields, from biology, astronomy and geology to medicine, climate science and neuroscience. Each major tool we invent does that—from the PCR method to x-ray devices. With so many new method combinations and scaffolds easily within our reach today—small and big—we are in a better position than ever to keep pushing the boundaries of the unknown (Chapter 6). That is the essence of frontier research: not just answering today's questions, but strategically extending our methods that opens and answers the questions of tomorrow we have not yet imagined. Because the real frontier of science is not just out there—it is mainly in building the tools that reveal what is out there—often unexpectedly.