Perspectives on Science

Anaximander's cosmos. The Earth is a column drum whose diameter is three times its height, and it stays where it is because it sits at equal distances from the extremes. Around it turn hollow wheels with rims of opaque vapour, filled with fire that shines through openings: the stars nearest, then the Moon, and the Sun farthest away .

Some of the content in this post is taken from a draft of my thesis.

Much has changed in the world of AI since my post on GFlowNets and scientific discovery. AI for Science has now become a mainstream topic. A Nobel Prize was awarded for the development of AlphaFold and LLM-based systems have solved hundreds of open problems in various areas of math including the Navier-Stokes Millennium prize problem Just during the process of writing this post I had to rewrite this line several times due to the constant release of new major results in math.. The scientific process has transformed drastically across every field thanks to the availability of strong consumer AI agents. The promise of accelerating scientific discovery is much closer to reality. This progress has also been accompanied by a reduction of the diverse views on the nature of science to a simplistic utilitarian one. In this post, I will discuss a wide variety of perspectives on the nature of science and the motivations of people who engage in the scientific enterprise. The aim of the post is not to “move the goalposts” for the impacts of AI on science, but rather to take this as an opportunity to look inward and reflect on how we view scientific practice and better understand this fuzzy enterprise to preserve what matters through this transition.

Whence things have their origin,
Thence also their destruction happens,
As is the order of things;
For they execute the sentence upon one another
— The condemnation for the crime —
In conformity with the ordinance of Time.

Anaximander of Miletus

sci·ence /ˈsaɪ.əns/ n. [Middle English < Old French < Latin scientia ‘knowledge’ < scīre ‘to know’]

The word “science” itself first appeared in Middle English around the 14th century, much later than the earliest signs of scientific practice. The word, roughly embodying “the state of knowing”, finds its origin in the Latin word scientia, meaning “knowledge, awareness, understanding”, which can be traced further back to scīre, meaning “to know”. The origin of scientific practice, however, remains more mysterious than the word itself. The generally accepted consensus is that scientific thinking emerged over tens of thousands of years independently across the world, but direct evidence of how that happened remains limited.

The documented history of science has been a delicate balance between the explanatory and predictive roles of science. The oldest records of “scientific practice” paint a predictive picture. Babylonian astronomers compiled centuries of systematic observations and developed elementary methods that could forecast eclipses and planetary positions with reasonably high accuracy, even without specific understanding of the exact dynamics or causes of planetary motion . The explanatory picture can be traced back to Anaximander of Miletus on the Ionian coast in the sixth century BCE Anaximander can be traced as the origin based on known written records, but it is also possible that similar ideas were developed and lost in the preceding centuries.. Anaximander was perhaps the first person on record to propose, among other things, that the Earth is a body floating unsupported in space. Beyond his specific predictions, Anaximander treated the inherited mythic account of the world as something that could be challenged. Unlike many of his peers and ancestors, Anaximander believed the world could be understood, that our understanding could be criticized and improved, and no authority is exempt from the criticism. Aristotle turned this belief into his explanatory theory of knowledge: to know something (scientifically) is to grasp the causes from which it follows .

In eleventh-century Cairo, Ibn al-Haytham tested competing theories of vision with targeted experiments using dark chambers, apertures, and lamps, with the outcomes explaining that light travels to the eye rather than from it . On the other hand, Chinese astronomers and engineers accumulated observational records and built intricate astronomical instruments which allowed them to build increasingly accurate predictive models for astronomical events . When European “natural philosophy” accelerated in the sixteenth and seventeenth centuries, it drew on mathematics, instruments, and texts that had circulated across all of these traditions.

A combination of socio-cultural and technological developments of the 16^th and 17^th centuries culminated in a phase of tremendous scientific progress, often called the “Scientific Revolution”. From Kepler’s laws to Newton’s explanatory theory of gravity, the balance between the predictive and explanatory nature of science continued. The impact of discoveries made during this period also compounded due to the establishment of the social bedrock of the scientific community at the same time. This includes academies, journals, and correspondence networks which collectively aimed to make the results public, criticizable, and cumulative. Advances based on the understanding developed during this period subsequently diffused to the world at large in the industrial revolution that followed, resulting in the organization of science as it operates today. The structures and norms developed during this period, while not perfect, have been critical for compounding knowledge and accelerating both discovery and technological growth.

What is Science?

Science is often described as an iterative and cumulative process, a puzzle solved piece by piece, with each piece contributing a few hazy pixels of a much larger picture.

Siddhartha Mukherjee, The Emperor of All Maladies

The success of the scientific enterprise, broadly construed, over millennia has received significant attention from a growing number of philosophers attempting to discern and codify what makes scientific practice so unique among other aspects of human activity. This collective deliberation led to a plethora of diverse and contrasting perspectives on the nature of science . The logical empiricists of the Vienna Circle in the early 20th century approached scientific discovery through the lens of formal logic, attempting to ground all scientific knowledge in empirical observation and logical deduction. As an alternative to the Vienna Circle perspective, Popper famously posited falsifiability as the defining characteristic of scientific thinking . Kuhn argued for the view that science progresses in long stretches of normal science with puzzle-solving within a shared paradigm, punctuated by paradigm-shifting revolutions . Feyerabend, on the other hand, painted a more anarchic picture of science, suggesting that every proposed rule of method has, at some decisive moment, been fruitfully broken .Notably, Hardt also argues that many of the developments in machine learning bear strong resemblance to Feyerabend's anarchic perspective on science. Anticipating this position more practically, Hanson observed that there is no “objective” eye, because every observation is fundamentally laden with theory , for instance through choices of what quantity is measured and how it is measured. Hacking further noted that experiments have a life of their own, driven by instruments and skill as much as by the testing of high theory .

Given the long and rich history of scientific practice, it is not surprising that no single perspective manages to capture all its nuances . Even falsifiability, a much celebrated and widely accepted perspective, sits uneasily with practice. The sharpest critiques come from Duhem and Quine who observed that hypotheses never meet observation alone but rather in bundles with auxiliary assumptions. A failed prediction cannot always identify whether it was the assumptions or the hypothesis that failed . Another commonly held picture of science is the standard “scientific method”, the famous loop of observation, hypothesizing, prediction, measurement, conclusion. While the scientific method is an appealing model for scientific practice, it is worth emphasizing that it is ultimately just a model. Scientific progress rarely follows a neat structured process. Hypotheses can be developed after some exploratory experiments and measurements with instruments and data rather than before them. Even anomalous measurements can result in fixing an apparatus issue or the development of new instruments rather than rejection of a theory. Reducing science to this structured loop also leaves out the critical piece of deciding what questions are worth asking in the first place.

Scientists themselves often describe science in ways that are less systematic but far more revealing than any attempt at a formal definition. Bridgman rejected the notion of a structured process altogether, remarking that the true scientific method is simply “doing one’s damnedest with one’s mind, no holds barred” . Medawar described science as “the art of the soluble” , the craft of turning vague wonder into questions that can actually be studied. Hopfield considers physics not just as a subject matter but “a point of view about the world” , an expectation that the right abstractions, approximations, and measurements can make phenomena intelligible. Feynman believed that science is a discipline of intellectual honesty, warning that “you must not fool yourself, and you are the easiest person to fool” .

Another important aspect, often lost in views that focus solely on the nature of scientific practice with respect to an individual, is the deeply social nature of the scientific process. Science is fundamentally a collective activity carried out in decentralized communities. “Objectivity” itself is less a property of individual scientists, who inevitably carry their own biases and blind spots, and more of an outcome of independent scrutiny from diverse perspectives in a community. Merton also credits institutional norms, such as communalism and organized skepticism, for keeping a research community self-correcting . Zooming out further, Alexander Bogdanov described science as the historical process of a successful collective organization of human experience . In this view, the social process of science also acts as a long-term medium of communication across societies. Theories, mathematical abstractions, and experimental records serve as compressed artifacts through which scientists communicate their understanding of the world for future generations of scientists, allowing people separated by centuries and cultures to argue with, test, and build on each other’s ideas. This naturally leads us to the next question: what motivates people to engage in this enterprise?

Why do people do Science?

Tiger got to hunt, bird got to fly;
Man got to sit and wonder, “Why, why, why?”
Tiger got to sleep, bird got to land;
Man got to tell himself he understand.

Kurt Vonnegut, Cat's Cradle, ch. 81

The diversity of perspectives on the nature of science also points to diversity in the motivations of people doing science. The oldest and most common answer is the sense of wonder that comes from understanding. Aristotle held that it was through wonder that people first began and still continue to philosophize . Firestein remarked that research is driven by a cultivated, “high-quality ignorance” that results in questions made sharp enough to pursue systematically . Put differently, science is a way for individuals to become systematically less confused about the world around them.

Bacon’s Novum Organum was instrumental in advocating for a utilitarian perspective of science, stating “human knowledge and human power meet in one; for where the cause is not known the effect cannot be produced. Nature, to be commanded, must be obeyed” . Vannevar Bush modernized this utilitarian motivation for the scientific enterprise, arguing in his seminal report towards the end of the Manhattan Project that basic research “creates the fund from which the practical applications of knowledge must be drawn” and therefore merits public support . This utilitarian perspective has been the bedrock of the modern organization of scientific communities.

On the other end of the spectrum is the view of science as a quest for beauty. Poincaré held that a scientist does not study nature for its utility, but because it is beautiful. This notion of beauty, Poincaré argued, comes from the “harmonious order of the parts, and which a pure intelligence can grasp” . Echoing this, Wigner also noted that concepts that mathematicians develop largely for their formal beauty end up forming the right abstractions for physical sciences, a coincidence he called “a wonderful gift which we neither understand nor deserve” . Last but not least, a less flattering but equally strong motivation for scientific inquiry is often recognition. Merton argues that recognition from peers is the principal reward in science, and that its forms (e.g. eponymous laws, prizes, citations) turn private ambition into public knowledge, since a discovery earns credit only when it is shared with the broader community .

In scientific practice, as with many human endeavours, the motivations of individual scientists are rarely singular and often evolve over time. The aims and motivations of the scientific enterprise as a whole are correspondingly plural. All motivations contribute in proportions that vary across fields and decades, and many of these motivations may not always agree.

As AI systems grow more capable, there is a natural push to use them to accelerate discovery. Most fields involving any computational aspect are already seeing rapid progress with the current generation of AI systems. Yet, the process of delegating parts of the scientific process to an autonomous system involves stating some desired objective (a score or a loosely defined task). Any such objective is always a lossy projection of the underlying aims and motivations that are far richer, partly tacit, partly contested, and underneath all of that, an expression of the underlying human desire to understand and communicate. It is hard to capture these aspects when we rely on a pass-rate on a proxy of predictive capabilities as a measure of success. Nevertheless, as autonomous systems get more capable, questions about the nature and motivations for science become increasingly important. We should strive to carry forward the deeply human aspects in this new era for scientific discovery.

Epilogue

In August 1938, speaking to a congress of anthropologists at Kronborg Castle in Helsingør, Niels Bohr argued that the lesson of modern physics, that every description of nature depends on the conditions of observation, also counselled humility in comparing human cultures. He described the common aim of all science as “the gradual removal of prejudices” . Rhodes writes that Bohr saw science as “a profoundly political force in the world” . Its purpose, he believed, “was to set men free.”