How did our species come to develop such a powerful toolbox that not only helps us survive but also enables us to do science—to study particles with accelerators and search for life on other planets with space telescopes? To survive and meet basic needs, all animals—including us—have to learn which types of food are edible and poisonous, how to avoid predators and where to find shelter. This means observing, solving problems, categorising and recognising regularities in nature. Many animals possess these cognitive and social abilities needed to use rudimentary tools and methods—through trial and error, testing hypotheses, and learning and sharing them with others. Take chimpanzees: they use tools for extractive foraging. They leverage stone and wood hammers to crack nuts and long sticks to extend their reach and extract termites and honey. They can use leaf sponges to collect water and levers for different tasks. Harnessing different tools to solve different problems, chimpanzees have a toolkit that they acquire through social learning and experimenting. For chimpanzees to use these tools requires that they have an objective of the tool in mind, predict how the tool can enable them to achieve that objective and understand how to apply the tool. This is not instinct—but relies on understanding the interactions needed between the tool, their hands and the desired outcome. From crows to sea otters and octopuses, many animals possess these cognitive capacities to innovate tools and build shelters by manipulating objects for their purposes and making inferences. If the way we and other smart animals have always gained knowledge was not based on observing, trial and error, and building theories about the world, then we would constantly fall into holes, get burned by fire and even walk off cliffs (something one generally only gets one shot at). Natural selection favours brains that observe, test and remember. Remarkably, these common evolved abilities have been at the centre of how we and other smart animals think, solve problems and leverage tools 202 T HE ENGINE OF SCIENTIFIC DISCOVERY over millions of years and they remain pivotal to science today. This raises crucial questions: how can we best design tools to tackle the diverse bottlenecks of our basic cognitive architecture? How far can AI systems be trained to observe, test, learn and think like clever animals do?
At least about 1.5 to 2 million years ago, early human species like Homo erectus and Neanderthals created complex tools, including the iconic hand axe. While early stone toolmakers used their hands to chip stone, it was their minds that carved the real tools using a broad set of abilities: they needed to carefully observe, experiment (beyond basic trial and error), test hypotheses, infer and predict (Picture 7.1). Making such tools also requires being able to imagine and plan. A remarkable feat was achieved when early humans, about 600,000 to 1 million years ago, learned to control fire—a milestone that made cooking, protection and warmth possible. About 400,000 years ago, they then developed sophisticated fire-hardened spears. These innovations, then and today, demand multi-step reasoning, making interconnected inferences and examining hypotheses. And we rely on these extraordinary methodological skills when we do science today. Picture 7.1 Creating stone hand axes requires causal reasoning, testing a hypothesis, abstraction, planning and refining them, circa 1.5 million years ago. Reproduced from Joy of Museums via Wikimedia Commons. Because we know early human species created these complex tools and because recreating them today requires carefully refining the tools, we know they also used these abilities. There is also no plausible alternative to explain how early humans reasoned and gained knowledge: these cognitive strategies were the only way to produce such technologies.
We are the result of millions of years of evolution, gradually emerging as a unique species in Africa around 250,000 to 300,000 years ago. Our evolution has shaped the way we think and understand the world. It has given rise to our core cognitive abilities to solve problems, experiment, reason about causes and effects, test ideas T and hypotheses and develop methods—abilities used to create stone tools in the Palaeolithic and run high-speed simulations with particle accelerators today. With growing social cooperation, our early ancestors developed more sophisticated language systems. These cognitive bridges enabled us to describe what we observe, share how tools work and better pass along tools and knowledge across generations. Imagination and abstraction also expanded in our early ancestors, not only in toolmaking but also in symbolic cave paintings. Today, across science we also use representational models to simplify complexity—from statistical simulations and mathematical equations, to models of scientific discovery (Chapter 6). Darwin himself sketched for example a model of a branching tree to illustrate evolution, emphasising that all species are related and humans are on just one branch of the same tree. Then came a massive ecological shift: the end of the last ice age around 11,000 years ago. This global climate change led to long-term settlements and catalysed greater cumulative knowledge. With our method-making minds, we then developed remarkable techniques for domesticating animals. This relied on experimentation and understanding biological reproduction, selective breeding to foster specific traits, and animals' nutritional needs. We then did not stumble on agriculture, but created and refined innovative farming techniques over time. To cultivate crops requires us to intricately understand how seeds, rain, soil quality and erosion causally interact. Farming emerged by careful, patient and iterative experimentation using natural controls: testing different seeds, soils, timings and planting techniques, and tracking seasonal cycles and comparing the outcomes to optimise yields. Developing agriculture goes far beyond trial and error by controlling for these variables—reflecting more systematic experimentation that resembles agricultural trials today (Picture 7.2). The payoff of this method-making strategy was enormous: it enabled stable food production, allowed more people to survive, generated a surplus of labour and freed up time for innovation—whether in toolmaking or even astronomy and mathematics that soon followed. Picture 7.2 Early farmers required reasoning about the causal interactions between seeds and rain, and comparatively experimenting with seeds—circa 9000 BCE. Reproduced from archaeology newsroom. The advent of agriculture and animal domestication marked a massive leap in our ability to actively shape and manipulate the environment using ingenious techniques. They were a milestone that enabled population density, more specialised labour and methodological diversity to grow hand in hand—each influencing and amplifying the others. What explains the difference between the pace at which we accumulated knowledge over the last 10,000 years and the ever-increasing pace today is not a 204 T HE ENGINE OF SCIENTIFIC DISCOVERY vast difference in our brainpower. Biologically, our brains have not changed much. Rather, the difference lies in the surge in the variety and sophistication of the tools and techniques we have invented, refined and passed along over generations. Our evolved abilities for method-making and tool-making developed through continual dynamic feedbacks between our physical environment, our cognitive capacities and our social structures (Chapter 8). These have enabled us to build ever more complex knowledge systems, invent tools, organise societies and ultimately lay the groundwork for science long before the 17th century. Our tools are products of our method-making capacities harnessed to solve problems more effectively. These abilities and methods evolve through a cumulative, bootstrapping spiral: each method we create enhances the next, creating a reinforcing cycle of innovation. So a natural question arises: why have our early ancestors largely been forgotten in the history of science? Because we do not have written records describing the methods they developed and the theories they held. They did not leave us behind direct documentation of their systems of language and numeracy—as the Sumerians, Egyptians, Greeks and others did. So we return to the puzzle: if the deep foundations of science—basically all the necessary methodological abilities to be able to reason and do science—existed at least 11,000 years ago, why did we not develop modern science then? Quite simply, these abilities alone are not enough.
As hunter-gatherer clans shifted into agricultural life, growing into villages and eventually cities, we developed tools and knowledge at an unprecedented rate. Our improved methods and tools changed most aspects of our lives: from the crops we cultivated, to the more complex structures we built and lived in, to the early medical and technological advances we made. With growing populations in the first civilisations, we took large leaps towards science: around 6000 years ago, we developed the earliest known systems of written language and mathematics. These were enormous cognitive upgrades. Babylonian mathematics developed fractions, algebra, Pythagorean theorem and quadratic and cubic equations ( Picture 7.3). These remarkable systems marked a transformational shift in our thinking from oral to written knowledge. These systems make recording and organising what we observe far more efficient. They reduce our cognitive constraints in memorising everything, processing information Picture 7.3 Babylonian mathematical tablet, circa 2000 BCE. Reproduced from public domain. T and making mathematical calculations—and even more accurate comparisons across generations. Once ideas could be written down, tested and shared without distortion, we could more efficiently build on our past methods and tools. It marked a leap in how knowledge evolved: human thought and tools became more testable. These advances can then be more easily shared (without forgetting) from generation to generation. Creating systems of writing reflects a pivotal point in cumulatively developing science, but also in understanding how science developed. Around 5000 years ago, developing geometry can be traced back to Mesopotamia and Egypt. It involved principles of areas, lengths, angles and volumes—used to survey land, measure building materials, manage irrigation systems and study the stars. Geometry helped us plan, measure and map the world and the sky. Creating astronomy provided a strategic advance: it enabled us to use stars and the moon as a clock for time keeping, a compass for navigating and a calendar for planning crop cycles, weather and temperature. At least 4500 years ago, ancient Egyptian civilisation and Norte Chico civilisation in modern Peru constructed massive pyramids. To construct a pyramid, we require—both then and today—developing principles in engineering, architecture and geometry that are grounded in systematically measuring, planning and experimenting. It reflects the kind of multi-domain thinking that underpins science today. Early systematic thinking in medicine also emerged. A surviving Egyptian medical textbook from around 1600 BCE outlines surgical procedures and instructions for treating fractures, tumours and wounds. It follows a clear experimental structure: diagnosis, intervention and prognosis. It is not modern medicine—but it is methodical, corrective and cumulative. These early systems—writing, geometry, astronomy and medicine—paved the way for science by increasingly freeing our minds from the limits of oral memory and isolated trial-and-error. Surprisingly, our ability to design controlled experiments stretches back at least to the Old Testament. The book of Daniel (1: 12–13) describes an experimental trial with control groups that tests the influence of a vegetarian diet: 'Test your servants for ten days. Give us nothing but vegetables to eat and water to drink. Then compare our appearance with that of the young men who eat the royal food [and drink wine], and treat your servants in accordance with what you see' . This is a classic controlled experiment: two groups, two diets, one observable outcome. The core logic of experimentation is clear: isolate a variable, control the conditions and observe the effect—asking 'what happens if we change just one factor?' To conceive and design such a controlled experiment—whether in ancient Babylon or in a modern hospital—we have to integrate multiple cognitive skills. It demands systematic reasoning to test whether a potential cause (a vegetarian, water-based diet compared to a non-vegetarian, alcohol-based diet) has an observable effect on people's physical appearance. It requires carrying out a trial and recording and comparing the outcomes of the physical appearance between the two groups after 10 days, and then drawing inferences from the trial outcomes to inform people's diets in the future (Picture 7.4 ). Today, we rely on these same method-making skills when designing clinical trials, but we amplify them with advanced tools like randomised controlled trials and statistical modelling. In the 19th century, we then combined controlled experiments 206 T HE ENGINE OF SCIENTIFIC DISCOVERY Treatment group (vegetarian diet) Sample population Control group (non-vegetarian diet) Outcome Baseline (day 1) Endline (day 10) Picture 7.4 Controlled experimentation using a treatment and control group described in the Old Testament, 2nd century BCE with newly developed methodological features like blinding and randomisation to reduce human bias. So how did we become able to do this kind of extraordinary experimental reasoning in the first place? It is the remarkable flexibility—plasticity— of our cognitive architecture that enables us, often through trial and error, to develop techniques to test interventions and isolate causes from correlations. From testing a hunting strategy to testing a crop rotation, our ancestors had to isolate causes, adjust interventions and watch the results—over and over again. What is astonishing is that civilisations across different regions of the world—the ancient Babylonians, Chinese, Mayans and others—developed agricultural, mathematical and technological systems largely independently, by using these methodmaking abilities and tools in increasingly experimental ways. Because they developed them largely independently illustrates that these enabling conditions are robust features of the human mind. These early systems were lenses through which we began to see the world in more structured, analytical ways—seeing patterns and possibilities, not just events and outcomes.
With a remarkably hands-on, experimental mindset, the ancient Chinese pioneered— as the first or independently—an impressive array of experimental advances that outpaced earlier civilisations and even their Mediterranean counterparts, the ancient Greeks. From immunisation techniques, magnetic compasses, negative numbers and the 'Pascal' triangle, to sophisticated astronomical records of supernovae, seismographs, advanced irrigation systems and quantitative cartography, Chinese innovation was driven by a pragmatic commitment to method and measurement. Ancient Chinese also devised papermaking and printing that preserved, spread and accumulated that growing knowledge—while ancient Greeks used slaves to handcopy texts. They created a more complex system of astronomical records than any previous culture—including detailed star catalogues and observations of eclipses and supernovae. Because of these meticulous observations, our records in contemporary science are able to go back millennia and link to observations we make today. These advances allowed the ancient Chinese to predict and control aspects of the natural world with striking accuracy through new methods. Take their smallpox immunisation technique: the groundbreaking method involved making T a powder-based vaccine out of smallpox scabs from infected individuals that was administered through nasal inhalation or small incisions into the skin of healthy individuals ( Picture 7.5 ). This produced a milder, controlled infection that offered immunity—drastically reducing mortality from smallpox by 20–30%. This incredible method mirrors today's immunology and vaccine logic: introducing a weakened form of a pathogen to train the immune system. It reflects the complex outcome of experimental strategies and a causal understanding of disease transmission. This striking depth in methodological thinking—applying methods to tackle chance—was not confined to medicine. The Chinese state actively supported experimentation across domains—from engineering and geology to technology and alchemy/chemistry. Picture 7.5 Ancient Chinese developed a smallpox immunisation technique that required understanding the causes and effects of the disease and how to control it. Reproduced from Wellcome Library, London. Take the invention of the mechanical clock engineered in the 8th century China— and later in the 13th century Europe. What does it take to create such a sophisticated technology—not just then, but now? We require a deeply methodical, cumulative process: observing mechanisms at play, experimenting to test different designs, reasoning about causes and effects to understand how gears and springs interact, and testing hypotheses to ensure each piece fits and functions precisely. Inventing a mechanical clock also demands abstract thinking to plan and imagine it, predictive reasoning to foresee its use for timekeeping, mathematical reasoning to divide time into repeatable units like 24 hours, 60 minutes and 60 seconds, and deep understanding of the complex interactions of the different parts as a whole. It is about building a remarkable system that precisely measures and regulates the world around us. It even suggested that nature could be thought of as operating like a mechanical machine. The same kinds of reasoning and methodical thinking are at the heart of scientific reasoning today. And the future of AI largely depends on designing systems that can flexibly do and combine such diverse reasoning. In fact, developing mechanical clocks can require as much complex reasoning, experimenting and knowledge as needed for example for Galileo to discover the moons of Jupiter using a telescope or establish that objects of different weights reach the ground at the same time by dropping and testing them. A powerful insight emerges: what matters most is not the object, but the method. Take also the ancient Greeks. Erasistratus for example tested weight loss by repeatedly weighing a captive bird while controlling its food intake—another one of history's many early controlled experiments. Archimedes uncovered the principles of 208 T HE ENGINE OF SCIENTIFIC DISCOVERY levers and buoyancy—fundamental principles in the fields of physics and mechanical engineering. The Pythagorean theorem, developed around 540 BCE, was applied as a cornerstone for developing Euclidean geometry around 300 BCE. Euclidean geometry then enabled Eratosthenes to make the remarkable calculation of the Earth's circumference around 240 BCE at 252,000 stadia—astonishingly close to the actual measurement of 40,075 km. In the Islamic Golden Age, the 11th-century polymath Ibn Al-Haytham made a vast methodological leap: collecting rich experimental evidence to pioneer elaborate theories of vision, light and colour. His groundbreaking work not only covered optics, but also inertia and mechanics—centuries before Newton. One of his most influential works, the seven-volume Book of Optics (Kitab al-Manathir), was completed in 1021, becoming one of the most important scientific texts of the medieval period. After its translation into Latin in the late 12th century, it circulated widely across Europe and influenced the work of scholars including Newton himself. This legacy earned Al-Haytham the title of the father of optics. He also described and applied the classic scientific method, namely doing controlled experiments to prove a hypothesis applying verifiable procedures and mathematical logic. He is often seen as the greatest physicist of the medieval era. Yet today, despite his monumental influence in past centuries, most have likely never heard of Al-Haytham. There is a striking pattern here. The key condition for innovation across these diverse contexts was not just geography or culture, but the patient, deliberate investment of attention and intellectual effort: method has emerged where we dedicate the time to observe and experiment systematically. That is how, over 1.5 millennia up to the 1500s, Chinese thinkers were able to design mechanical clocks and engineer the chain drive and long iron-chain suspension bridges—through systematic experimentation and ever more precise tools. This sustained institutional memory of existing technologies over time had a spiral effect. It enabled creating branches of physics—optics, magnetism and acoustics—and hydraulic engineering. It fed into pharmacology, publishing the first official book of medicinal drugs that outlined techniques for how to use them and their effects. It triggered developing one of the most ambitious intellectual projects in history: the world's largest physical encyclopaedia, the Yongle Encyclopedia—a state-sponsored project to codify and preserve the vast corpus of accumulated knowledge. By the 15th century, the Chinese had built the most sophisticated scientific and engineering system that the world had ever seen—fuelled not just by innovation, but by a foundation of long-term social stability. Importantly, the Chinese architects of these remarkable methods and technologies did not necessarily rely on mythology, religion or the supernatural. But they had to rely on our general abilities of systematic observation, iterative problem-solving and methodical experimentation. So we return to a puzzle raised in Chapter 3: who then actually invented the classic scientific method? We cannot meaningfully talk about the founders of the classic scientific method in any general sense. Yet if we were to, then the first huntergatherer groups who developed methods and tools may be the best candidates as the founders—because they were the first to systematically observe, reason causally, experiment and imagine to be able to develop the methods and technologies that science relies on. As long as we have existed, we have used these methodological T abilities in ever more systematic ways. We see them in the work of Archimedes and Democritus in ancient Greece, laid out in systematic detail in the 11th century by Al-Haytham, and reiterated again in the 13th century by Grosseteste, Roger Bacon and Aquinas; and later in the 17th century by Francis Bacon, Galileo and Newton, alongside others. The term the founder of the classic scientific method has been attributed to each of these scholars.