Measure for measure on Measure for Measure

Scientific and Artistic Imagination

aghayeva.u@gmail.com (Ulkar) — Thu, 29 Jan 2026 00:00:00 +0000

This essay was first published on the Elicit blog.

Oh, most magnificent and noble Nature!
Have I not worshipped thee with such a love
As never mortal man before displayed?
Adored thee in thy majesty of visible creation,
And searched into thy hidden and mysterious ways
As Poet, as Philosopher, as Sage?

Humphry Davy, as quoted in John Davy’s “Fragmentary Remains”

It may seem that scholars have been arguing about the relative priority of sciences and humanities in education and culture since the earliest days of the Academy.¹ But, as we saw in the previous post, the idea of separating academic disciplines into these groups is in fact no older than the 19th century. In the last quarter of that century, two eminent scholars and public figures of their time, the poet Matthew Arnold (“at once the most formidable and the most temperate champion of the Humanities”²) and naturalist Thomas Huxley (“Darwin’s Bulldog”), had a gentlemanly but pointed public exchange on the matter that encapsulated some of the best arguments on both sides.

Since then, the debate has resurfaced many times in various forms and contexts — from the much more bitter C.P. Snow – F.R. Leavis controversy in 1950s Britain, to the more light-hearted repartees over physicists and lyricists in the Soviet Union and the shape rotator vs wordcel memes in recent internet memory.

That the debate has been recurring and each time emotionally stirring tells us something about the lasting tensions about who gets to define “culture” and what it means to be “educated.” But there is also an unwavering desire for a unified culture. There is a sense that the things an engineer or a biologist cares about and those salient to a novelist or a musician ought to somehow belong to a single conversation rather than to isolated subcultures. The tension is not only about whose work matters more or less but about whether we can make ourselves intelligible to one another.

Physician-writers, linguists turned mathematicians, English majors who went on to become neurologists and history majors who switched to theoretical physics (not to mention scientists who make art and music) are all testaments to the affirmative. Yet the very fact that we single them out suggests that this cultural unity is not to be taken for granted. It may well exist in individual lives but it is hard to come by at the institutional level.

In the 19th century, as Western economies were industrializing, major educational reforms were under way. In Britain, the traditional curriculum of Greek, Latin and English literature which would previously grant one the “stamp of the educated man” was starting to be deemed insufficient to prepare the new professional classes for modern life.³ In a time of rapid technological advances, training in science was becoming as essential for education as Horace’s and Milton’s poetry had once been. But it took time for this disposition to be widely accepted.

Enter Thomas Huxley and Matthew Arnold. The two gentlemen were on friendly terms and had exchanged extensive private correspondence in the 1860s and 1870s before taking their debate on the sciences and humanities into the public space.

Thomas Huxley on the role of the sciences in culture

Huxley delivered his lecture titled Science and Culture at the opening of a science college for working and middle-class students in Birmingham in October 1880. Mason Science College wasn’t the first educational institution in Britain to teach the sciences⁴ but it was first to take a stand against the classics, banishing not only theology but also “mere literary instruction and education” from its curriculum. It is no coincidence that Thomas Huxley chose the opening of this college as an occasion to advance his arguments on the increasingly important role of the sciences in culture.

He starts out with painting a picture of the state of science education of his time. It would seem quite unusual to us, reading the lecture in the 21st century, that in Huxley’s Britain the natural sciences had to fight for their rightful place in university curricula, to be on par with ancient and modern literature. The advocates of science education, Huxley notes, have been opposed both by the practical men of business and by classical scholars of the day.⁵ Yet Huxley comes forth offering his defense of the educational experiment commencing at Mason College:

For I hold very strongly by two convictions—The first is, that neither the discipline nor the subject-matter of classical education is of such direct value to the student of physical science as to justify the expenditure of valuable time upon either; and the second is, that for the purpose of attaining real culture, an exclusively scientific education is at least as effectual as an exclusively literary education. [here and below, bolding added]

For the majority of educated Englishmen at the time, however, culture was to be acquired only through “liberal education,” which meant primarily instruction in literature, particularly of Greek and Roman antiquity.

But what is meant by culture here? Huxley turns to the definition offered by “our chief apostle of culture,” Matthew Arnold, stating that the meaning of culture is “to know the best that has been thought and said in the world.”⁶

Huxley readily agrees with Arnold that culture is “criticism of life contained in the literature,” our way of making sense of it:

For culture certainly means something quite different from learning or technical skill. It implies the possession of an ideal, and the habit of critically estimating the value of things by comparison with a theoretic standard. Perfect culture should supply a complete theory of life, based upon a clear knowledge alike of its possibilities and of its limitations.

But, he says, literature alone is not sufficient to supply this knowledge, and that natural sciences are essential for a complete culture. In what follows, he unfolds his criticism of the contemporary state of humanities and calls for a revision of the British educational system.

Looking back at the history of education, Huxley contrasts the modern study of science with the Medieval scholastic approach to learning, which was “essentially bookish” (as C.S. Lewis put it in The Discarded Image⁷). Fluency in Latin was a necessity for acquiring “all the higher knowledge of the western world,” since it was contained in books written in that language. Through these written works, Medieval people were equipped with a complete and essentially theological criticism of life — their Model of the Universe and the place of humanity in it:

Our ancestors had a living belief in this theory of life, and acted upon it in their dealings with education, as in all other matters. Culture meant saintliness—after the fashion of the saints of those days; the education that led to it was, of necessity, theological; and the way to theology lay through Latin.

This apparent completeness of the Medieval system of the world was challenged by the revival of the Greek studies through Arabic translations (and later originals) of ancient texts from the 12th century onward. Greek philosophy and literature tremendously enriched and expanded Western European culture. With the invention of the printing press in the 15th century, classical learning spread more broadly, and “[t]hose who possessed it prided themselves on having attained the highest culture then within the reach of mankind.” Huxley refers to them as the Humanists. By the Renaissance, ancient Greeks were seen as the paragon of “perfect intellectual freedom – of the unhesitating acceptance of reason as the sole guide to truth and the supreme arbiter of conduct.”

But, says Huxley, in his own 19th century the classics alone no longer represent the only avenue to culture or a complete encapsulation of it, because of the increasing importance of natural sciences. Natural knowledge is what may be called “a scientific criticism of life”:

It appeals not to authority, nor to what anybody may have thought or said, but to nature. It admits that all our interpretations of natural fact are more or less imperfect and symbolic, and bids the learner seek for truth not among words but among things. It warns us that the assertion which outstrips evidence is not only a blunder but a crime.

Moreover, and this is the crux of Huxley’s argument, – what we now know as classics is itself a part of a culture that was deeply engaged with the natural world. In what feels like the culmination of the lecture, Huxley puts forward his main sharp criticism of “the modern Humanists”:

…that they may be learned specialists, but that they possess no such sound foundation for a criticism of life as deserves the name of culture. And, indeed, if we were disposed to be cruel, we might urge that the Humanists have brought this reproach upon themselves, not because they are too full of the spirit of the ancient Greek, but because they lack it.

That is quite a devastating contention – that the humanists have betrayed the Greek spirit whose torch they claim to carry forth. Indeed, for ancient Greeks a complete education meant natural philosophy – knowledge of the natural world — along with metaphysics and rhetoric and poetry and gymnastics. The spirit of the ancient Greek was inquisitive about all that exists in the universe and certainly wasn’t content with written texts only. Huxley continues,

We cannot know all the best thoughts and sayings of the Greeks unless we know what they thought about natural phenomena. We cannot fully apprehend their criticism of life unless we understand the extent to which that criticism was affected by scientific conceptions. We falsely pretend to be the inheritors of their culture, unless we are penetrated, as the best minds among them were, with an unhesitating faith that the free employment of reason, in accordance with scientific method, is the sole method of reaching truth.

The use of the “scientific method” is rather anachronistic here but otherwise this is Huxley’s strongest challenge to classical humanism.⁸ Granted, he acknowledges that the innate capacities and talents of mankind are varied, as accordingly are the paths to culture:

I am the last person to question the importance of genuine literary education, or to suppose that intellectual culture can be complete without it. **An exclusively scientific training will bring about a mental twist as surely as an exclusively literary training. **

Yet at the end of the lecture he reiterates and reinforces his main argument that for those who want to dedicate their lives to the study and practice of science or medicine or business — “for all these … classical education is a mistake.”

Shots fired. It would take a no less eminent and articulate thinker than Huxley himself to present a worthy response.

Matthew Arnold’s Rede lecture in defense of humanities

It wasn’t until August 1882, about two years after Huxley’s lecture, that Matthew Arnold delivered his Rede lecture⁹ titled Literature and Science at the University of Cambridge in response to him. Arnold opens his talk documenting the turning tide in the place of sciences in education, with “friends of physical sciences”¹⁰ becoming “a growing and popular body.” Leading publications of his time were already prophesizing the downfall of humanities in a not so distant future:

…the Times…takes the gloomiest view possible of the future of letters, and thinks that a hundred years hence there will only be a few eccentrics reading letters and almost every one will be studying the natural sciences…¹¹

In response to Huxley’s insinuation that literature alone does not supply sufficient knowledge to represent and make sense of culture, Arnold suggests that some clarification of terms is in order. Namely, what is meant by literature, as a path to culture? Literature that Arnold has in mind is not just belles lettres (“fine letters,” like fiction, poetry and drama¹²), as implied by Huxley. It’s not just “superficial humanism” or “a smattering of Greek and Latin and other ornamental things” which can be seen as the opposite of the exact and veridical scientific knowledge. Instead, in Arnold’s definition:

Literature is a large word; it may mean everything written with letters or printed in a book. Euclid’s Elements and Newton’s Principia are thus literature. All knowledge that reaches us through books is literature.

If culture is defined as “the best that has been thought and said in the world,” Arnold also includes in it the works of “the great observers and knowers of nature.”¹³ By this token, all serious thought participates in the same enterprise – building a shared recorded culture. And in this sense, there’s no disagreement between Huxley and Arnold. However, some later commentators have argued that “to reduce science to literature by insisting that science is a kind of writing” is misleading.¹⁴ Science is first and foremost a living practice and is far from being fully contained in academic articles and books.¹⁵

Arnold goes on to praise the value of scientific knowledge as rooted in verifiable observation and experiment that everyone should have some experience of. Indeed, from the beginning of his career as Inspector of Schools, Arnold defended the view that while not every student needed to learn the classics through original Greek and Latin, a balanced, complete education would include both humanities and scientific training.

But where Arnold diverges in opinion with “the friends of the physical sciences” is when it comes to universally making science the predominant, or even exclusive, program of education. In the following, Arnold builds a compelling line of argument in defense of humanities.

What do the advocates of science education risk missing by neglecting or excluding humanities from a complete education? The very constitution of human nature, no less. Arnold starts by enumerating the major forces that play out in human life and that build up human nature: “the power of conduct, the power of intellect and knowledge, the power of beauty, and the power of social life and manners.”

In other words, these are the powers that are essential for the answers to eternal questions like “How should I live my life?”, “How should I carry myself in the world?”, “How should I relate to people around me?”. Importantly, these powers are not isolated:

Following our instinct for intellect and knowledge, we acquire pieces of knowledge; and presently, in the generality of men, there arises the desire to relate these pieces of knowledge to our sense for conduct, to our sense for beauty, and there is weariness and dissatisfaction if the desire is balked.** Now in this desire lies, I think, the strength of that hold which letters have upon us**.

Bits of knowledge in any field can be interesting in and of themselves but it is in human nature to seek to piece them together and relate them to general principles. This is a clear tendency within the disciplines of natural science. The periodic table systematizes our knowledge of chemical elements and equips us with the ability to predict their properties. Conservation laws in physics set constraints on the possible behavior of physical systems. Phylogenetics in biology is a reconstructed view of evolutionary history based on genomic relatedness of organisms in the tree of life.

But the same tendency just as well applies outside of systematic natural knowledge:

We feel, as we go on learning and knowing… the need of relating what we have learnt and known to the sense which we have in us for conduct, to the sense which we have in us for beauty.

In Plato’s Symposium, the priestess Diotima of Mantinea speaks of men’s desire that “that which is good should be for ever present to them.” This desire, she says, is the essence of love (eros). Strange as it may sound to the modern ears, Diotima’s definition points not to romantic love but to nothing less than a cosmic striving for unalloyed Good (agathon) that manifests as a drive to “beget in beauty” – through children, through works of thought, deeds of virtue and, at a societal level, through laws and institutions.

Arnold suggests this same primordial desire is at the root of the human instinct for unifying and integrating knowledge, including natural knowledge, with our sense of what it means to live a good life. In following this instinct, “we are following the instinct of self-preservation in humanity.” Eros in Diotima’s sense is an integrative force, it refuses to be content with fragmentary and disconnected facts. Instrumental, factual knowledge is useful inasmuch it can serve as a bridge to something beyond itself:

Knowledges which cannot be directly related to the sense for beauty, to the sense for conduct, are instrument-knowledges; they lead on to other knowledge, which can.

In the words of Saint Bernard,¹⁶ “There are those who seek knowledge for the sake of knowledge; that is Curiosity. There are those who seek knowledge to be known by others; that is Vanity. There are those who seek knowledge in order to serve; that is Love.” Of course, any discipline, scientific or artistic, can be engaged with at any of these levels or their combination. But Arnold argues that it is the unique role of the humanities to cultivate a disposition toward the world which nurtures the third level of knowledge seeking and integration, one built on Love.

The realm of science is that of means, not ends. It tells us how to best do things we have already decided to do, not why we ought to do them. Scientists work with dedication to expand and deepen our knowledge of nature - but take any scientific theory or experimental result, and ask: what does it mean in the grand scheme of things, how does it relate to our sense of beauty and moral conduct?

There are of course practicing scientists who are inspired and imaginative writers communicating new ideas in science to the general public, and it can be argued that they were proportionately even more numerous earlier in science history. But those that take on the daunting task of addressing Arnold’s questions are few and far between. That is a task more fit for the humane letters, at least in Arnold’s view. But the question remains, whether modern humanities are indeed up for the task. Do humanities in practice deeply engage with new scientific knowledge and integrate it into the collective humanistic sensemaking? We will return to this question later.

Someone who is a “born naturalist” can perhaps be so deeply absorbed by his research that he can “pass his life happily in collecting natural knowledge and reasoning upon it,” and be content with it. Arnold relates a sentiment expressed by Darwin, who was said to have once admitted to a friend that he lacked the need for two things “which most men find so necessary to them – poetry and religion; science and the domestic affections, he thought, were enough.”¹⁷ Though this anecdote must have taken place later in his life – in his 20s, while on the board of the Beagle, Darwin had one constant companion book with him, and that was Milton’s Paradise Lost, which he re-read multiple times.

Arnold thought “Darwins are very rare”, but perhaps (late) Darwin was in fact a harbinger of things to come. In Arnold’s own time a more typical man of science was more like Michael Faraday, who still considered himself a natural philosopher with a deep interest in theology, rather than a “mere” scientist.

While Huxley was largely critical of Medieval education as static and limited, Arnold calls attention to the fact that scholastic knowledge

…delivered by Scripture and the Church so deeply engaged men’s hearts, and so simply, easily, and powerfully related itself to the desire for conduct, the desire for beauty <…>. All other knowledge was dominated by this supposed knowledge and was subordinated to it <…>.

And even if, as Huxley warned, the new knowledge brought forth by modern physical sciences threatens to shatter the harmonious worldview that our predecessors lived by, the more there is a need for the humanities to reintegrate and relate this knowledge to our sense of how to live a good life:

…if we find by experience that humane letters have an undeniable power of engaging the emotions, the importance of humane letters in man’s training becomes not less, but greater, in proportion to the success of science in extirpating … “mediaeval thinking.”

That the letters indeed have this power over us, says Arnold, is ubiquitously attested by experience. As an immediate example, Arnold invites us to ponder the difference between two statements: a common saying like “Patience is a virtue” and this weighty quote from Homer, “for an enduring heart have the destinies appointed to the children of men.” The former, more plain saying conveys its message clearly, to be sure, – but it is the latter that stirs something in us, that elevates us above mundanity. This is not merely a matter of stylistic or aesthetic differences. Great literature doesn’t merely express thoughts more beautifully; it makes conceivable kinds of thoughts that plain speech cannot reach. It creates a larger expanse for thoughts that can be thought within it.

There’s no straightforward answer or a recipe as to how exactly the humane letters can relate new scientific knowledge to our moral and aesthetic values. But Arnold is convinced that the art and poetry of those who lived before us, who even had a limited or erroneous knowledge of nature, have in fact

…not only the power of refreshing and delighting us, they have also the power, – such is the strength and worth, in essentials, of their authors’ criticism of life, – they have a fortifying and elevating and quickening and suggestive power capable of wonderfully helping us to relate the results of modern science to our need for conduct, our need for beauty.

Arnold thus reinforces his belief that the appeal of humanities will remain irresistible and essential in our cosmic quest to unify all knowledge in service of a good life, so long as the human nature remains what it is.

There is undoubtedly an untold wealth of this “elevating power” contained in the writings of those who came before us, regardless of the accuracy of their scientific picture of the world. But it seems that the modern humanities have been taking this point to implicitly mean that there’s little to be added to our understanding of human nature, besides the already existing literature, or other literary works building on it. This may explain the general resistance of the modern humanities to absorbing new scientific knowledge into a more complete criticism of life. Cognitive literary studies, environmental humanities and a suite of new humanities disciplines with the neuro- prefix (neurophilosophy, neuroaesthetics, neurohistory, neuroarthistory; usually based on the work of a single innovative author) are all steps in this unifying direction; yet they are marginal movements that command little engagement both within the broader humanities and outside of them.

Contrary to Arnold’s expectation, it seems that historically it has in fact been scientists who have made more earnest and rigorous attempts to reach across the growing divide between “the two cultures”, as C.P. Snow called the estranged institutions and practitioners of literature and of the sciences, nearly 80 years after the Arnold-Huxley debate. And while initially the scientists’ role as cultural translators was, in a way, a social necessity imposed by the then existing cultural conventions, in more recent times such attempts to synthesize and work towards a consilience of all knowledge have come from maverick undertakings of individual scientists, going against institutional pressures for specialization.

In the 19th century, a practicing scientist, like any educated person, had to be closely acquainted with the Western literary canon to be admitted to the “cultured class.” Breadth of reading was highly prized, and scientists in their writings made refined references to contemporary and older fiction and poetry to win the confidence of their readers. And indeed, if you read through the essays by scientists and literary writers of the 19th century collected in this anthology, you’ll see that scientists of the day were no less skilled and imaginative writers than their contemporary men of letters. Some even wrote poetry reflecting their scientific work or criticism of scientific practice (see also the epigraph to this post).¹⁸

Today, if we hear about a professor of molecular biology who also reads lectures on Faust and theology, we are rather surprised (and delighted). People like Dr. Stephen Meredith of the University of Chicago are quite rare. Among their small but notable number are Iain McGilchrist, Terrence Deacon, E.O. Wilson, Antonio Damasio, Eric Kandel and Amos Funkenstein. Each of them has seriously attempted to weave new scientific knowledge into a unified picture of human nature and human culture.¹⁹

But perhaps another source of the still predominant recalcitrance of the contemporary humanities to deep engagement with the contemporary sciences is the implicit fear that science is capable of “explaining away” human nature with crude reductionism. A wonderful account of this tension is developed by C.S. Lewis in Meditation in a Toolshed where he relates his experience of standing in a dark shed with a shaft of sunlight falling through a crack in the door. One can look at the beam and see a bright column of dust motes, or one can step into its path and look along the beam, out of the shed altogether, seeing the sun and the green world beyond.

Looking at and looking along, Lewis suggests, are not rival methods competing for the one true picture, but two irreducible modes of our access to reality. To look only at is to risk mistaking external detached analysis for the whole of experience, and to look only along is to risk never seeing how our experience might in fact itself be open to probing and explanation. The best of two worlds is to learn how to move between them without losing sight of one while inhabiting the other.

The faculty that allows us to do so is imagination. Scientific imagination lets us build abstractions of reality and model that which we cannot observe directly – worlds too small or too large to access with our senses. Artistic and literary imagination lets us inhabit other minds, other times and other lives at the immediate human level. At its best, a complete education is the cultivation of a unified imagination – The Educated Imagination – capable of both looking at and looking along, which would allow us to recognize the sciences, the humanities and the arts all as different manifestations of the same underlying human project.

Classical Academy is the name of the public garden where Plato taught his school. Greek Akadēmeia refers to “the grove of Akadēmos," a legendary Athenian of the Trojan War. ↩︎
The Times (London), October 2 1880. ↩︎
Literature and Science in the Nineteenth Century: An Anthology, Ed. Laura Otis (2009). ↩︎
For example, The Royal College of Chemistry in London was established in 1845 and Owens College in Manchester in 1851, but neither excluded the classics from their curricula. ↩︎
In Huxley’s words, literary men would retort that “the study of physical science … touches none of the higher problems of life; and, what is worse, … the continual devotion to scientific studies tends to generate a narrow and bigoted belief in the applicability of scientific methods to the search after truth of all kinds.” ↩︎
Matthew Arnold, Culture and Anarchy (1869). The full quote defines culture as the “pursuit of our total perfection by means of getting to know, on all the matters which most concern us, the best which has been thought and said in the world, and, through this knowledge, turning a stream of fresh and free thought upon our stock notions and habits, which we now follow staunchly but mechanically.” ↩︎
C.S. Lewis writes in The Discarded Image that in the Middle Ages the “single, complex, harmonious mental Model of the Universe” was conditioned on two factors: “the essentially bookish character of their culture, and their intense love of system. <…> At his most characteristic, medieval man was not a dreamer nor a wanderer. He was an organiser, a codifier, a builder of systems. He wanted ‘a place for everything and everything in the right place’. Distinction, definition, tabulation were his delight.” ↩︎
Though arguably truth can be reached not only by reason alone, taking as but one example numerous cases of scientists finding answers to their research questions in a dream. ↩︎
Sir Robert Rede’s Lectures, or Rede lectures, are annual public lectures at the University of Cambridge established in 1668. ↩︎
By “physical sciences” Arnold most likely meant mathematically and experimentally oriented natural sciences like physics, chemistry, astronomy, geology, not just physics alone. “Friends” here refers to not only practicing scientists but also educators and advocates of science education. ↩︎
Admittedly, this prophecy has come to pass to some extent, with the share of humanities degrees among college graduates and with the support for humanities departments at universities declining over the course of the past century. Offsetting the institutional decay of humanities, however, are the grassroots initiatives and a flourishing of online literary communities that have been springing up in the past decade. ↩︎
Though there’s also a pejorative modern use of the term that is almost the opposite of the literal (artistic and imaginative forms of writing) – writing that prioritizes aestheticism over substance. In yet another sense, belles lettres include essays, “published collections of speeches and letters, satirical and humorous writings, and other miscellaneous works.” ↩︎
As well as arts: in a letter to Huxley dated 17 October 1880, Arnold clarified, “[the quote] was meant to include knowing what has been said in science and art as well as letters.” (emphasis added) ↩︎
And so is, for that matter, “to reduce literature to science by insisting that its codes also give a higher or privileged access to the real.” ↩︎
And so are the humanities and arts themselves – the living practice of performance and active engagement beyond written texts and scores. ↩︎
As quoted in The Three Fausts, by Stephen Meredith. ↩︎
To this example we can add Richard Feynman who famously had no time for all things culture in his scientifically very prolific life (though he was an enthusiastic bongo player). ↩︎
There’s an irony here, in that, while in Arnold’s day science deferred to letters to be respectable, in our times it is humanities that may feel pressured to borrow from scientific legitimacy and authority (with the neuro-, evolutionary and data-driven approaches mentioned above). ↩︎
We may also observe that many prominent scientific ideas of our time like evolution, relativity, information theory and complexity theory have to some extent already become part of our literary and artistic imagination. Conversely, some tools of the humanities like narrative analysis have been integrated into the sciences and medicine (though the prioritization of “storytelling” in science is to be taken as a cautionary tale). ↩︎

Scientific and Artistic Imagination

aghayeva.u@gmail.com (Ulkar) — Sat, 29 Nov 2025 00:00:00 +0000

This essay was first published on the Elicit blog.

In the 1880s, there was a famous debate between the English poet and cultural critic Matthew Arnold and the biologist and anthropologist Thomas Henry Huxley (also known as “Darwin’s Bulldog”¹). The debate concerned the place of natural sciences vs humanities and arts in contemporary education and culture.

In Arnold’s definition, the meaning of culture is “to know the best that has been thought and said in the world.” Fittingly, the Arnold-Huxley debate is itself in that category when it comes to understanding the still existing tension between the practice and practitioners of science, on the one hand, and humanities and arts, on the other.

In this series, we will explore this tension, starting with the very notions of ‘scientist’ and ‘artist,’ so common in everyday language that we forget about their thorny origins. Next, we will follow the Arnold-Huxley debate itself and how it relates to the current state of affairs. We will then look into the workings of scientific and artistic imagination and representation in the arts and sciences.

Scientists, like artists, are praised for their ingenuity, imaginative thinking and aesthetic judgement. It is a sign of high appreciation to refer to a scientific theory or experimental design as an artwork, as did ‘the father of nuclear physics,’ Ernest Rutherford, in his speech at the Royal Academy of Arts in 1932:

A strong claim can be made that the process of scientific discovery may be regarded as a form of art… A well constructed theory is in some respects undoubtedly an artistic production.

The French theoretical physicist and historian of science Pierre Duhem also spoke eloquently to this effect:

It is impossible to follow the march of one of the great theories of physics, to see it unroll majestically its regular deductions starting from initial hypotheses, to see its consequences represent a multitude of experimental laws down to the small detail, without being charmed by the beauty of such a construction, without feeling keenly that such a creation of the human mind is truly a work of art.²

On the other hand, a work of art is rarely compared to a scientific discovery as a matter of praise. For example, T.S. Eliot wrote that Joyce’s Ulysses “had the importance of a scientific discovery,” and Émile Zola in his essay The experimental novel (1880) likened working on a novel to experimenting, inspired by the physiologist Claude Bernard’s Introduction to Experimental Medicine (1865). But these cases stand out precisely because they are rare.

Today’s practicing scientists and artists work at separate institutions, rarely collaborate, and communicate their work to the public in very different ways. But has this always been the case?

It may come as a surprise that the very notions of ‘scientist’ and ‘artist’ are modern inventions. Before scientists, there were natural and experimental philosophers, scholars, theologians, and practitioners of various liberal and mechanical arts³ which some of the modern sciences sprang from (astronomy, agriculture, and medicine, to name a few). The term ‘artist’ in our modern understanding of a creator of works of fine art is itself less than 300 years old. As a matter of fact, ‘artist’ and ‘artisan’ were used interchangeably before the 18th century, when the split between ‘fine arts’ and ‘crafts’ crystallized.

Debates over ‘scientist’

The word ‘science’ itself is a French import from the Middle Ages, initially used as a synonym of knowledge in general. It then acquired the connotation of systematized knowledge arrived at by logical demonstration (deduction) from first principles. This is the Aristotelian conception of knowledge, in distinction to the less reliable ‘common knowledge.’ In 1620, Francis Bacon in his treatise Novum Organum proposed a different conception of science as knowledge arrived at gradually, by means of observation and experiment – what we now refer to as empirical science.

The English polymath William Whewell coined the term ‘scientist’ in 1833, in the course of his debates with the Romantic poet and philosopher Samuel Coleridge on the nature of sciences and their place in the tree of knowledge.⁴ In his later book, Philosophy of the Inductive Sciences (1840), Whewell further advocated the use of the term:

We need very much a name to describe a cultivator of science in general. I should incline to call him a scientist. Thus we might say, that as an artist is a musician, painter, or poet, a scientist is a mathematician, physicist, or naturalist.

That same year, the word ‘scientist’ was added to the Oxford English Dictionary. Still, it wasn’t readily adopted into everyday use, partly because many (as we would call them) scientists of the day like Michael Faraday and Thomas Huxley⁵ strongly preferred the term ‘natural philosopher’, which encompassed broader philosophical and theological areas of inquiry. They saw themselves as men of broad liberal education, who chose to engage in scientific investigations both as a personal vocation and service for the benefit of mankind. They found it distasteful to pursue science for money. “To them the word scientist implied making a business of science; it degraded their labours of love to a drudgery for profits or salary.”⁶

As an interim solution, the phrase ‘man of science’ was used in formal discourse and writing for much of the 19th century. But fierce debates over the grammatical and semantic legitimacy of the word ‘scientist’ unfolded over the course of the second half of the century.

Even as late as the 1920s, the debates about ‘scientist’ continued on the pages of Nature, with the objections ranging from the pedantic, related to the word formation rules, all the way to the vocational, like an ‘antipathy to science as a calling or profession.’ What eventually sealed the term’s acceptance were educational reforms that placed science education on the same footing as education in medicine, law and theology. Science became merely another profession to choose among others, and to be a scientist now meant to be a professional, like a physician, a lawyer or a clergyman.

By the 1960s, ‘scientist’ fully took root in the language, with a definitive history of the term opening thusly:

The appellation scientist is considered a title of honour, hotly contended for by economists, engineers, physicians, psychologists, and others.

That is indeed a 180° turn from the initial antipathy with which the word was met. But in most other languages the story of the equivalent of ‘scientist’ wasn’t nearly as dramatic as it had been in English. In many cases there was a gradual transition from a native ‘scholar’ or ‘sage’ or ‘learned man’ to ‘scientist,’ by way of narrowing the meaning of the original term.

In German, for example, the current term ‘Wissenschaftler’ literally means ‘practitioner of (systematic) knowledge’, though earlier terms included ‘Gelehrter’ (‘learned one’) and ‘Naturforscher’ (‘investigator of nature’). In Russian, ‘ученый’ is a ‘learned one,’ as is the Hungarian ‘tudós’ and the French ‘le savant.’ In Japanese (科学者, ‘science-person’) and Chinese (科学家, ‘science-expert’), however, the introduction of the equivalent of ‘scientist’ was a matter of deliberate linguistic modernization.

The union of arts and sciences

The history of the notion of ‘artist’ unfolded in Europe over several centuries. The Latin ‘artem’ (plural ‘ars’) and Greek ‘techne’ originally referred to (technical) skill and mental agility in general. For over two thousand years, art signified “any human activity performed with skill and grace,”⁷ from shoemaking to architecture to rhetoric. And, for what it’s worth, even as late as 1755, one of the meanings of science itself was given in Samuel Johnson’s Dictionary of the English Language as ‘any art or species of knowledge’ or, alternatively, as ‘one of the seven liberal arts.’ Arts and sciences went hand in hand and were practiced by the same people.

Artists-artisans of the Renaissance made extensive use of the technical knowledge from the liberal arts like geometry and mechanical arts like medicine in their work:

In the course of the fifteenth century, the accelerating spread of the knowledge of perspective and modeling, along with the revival of ancient models, led to the conviction that painting and sculpture now required not only apprenticeship but also some knowledge of geometry, anatomy, and ancient mythology. Alberti and Leonardo projected an image of the artist as a “craftsman-scientist.”⁸

A similar sentiment was expressed by David Hume in An Enquiry Concerning Human Understanding (1748):

An artist must be better qualified to succeed in this undertaking; who, besides a delicate taste and a quick apprehension, possesses an accurate knowledge of the internal fabric, the operations of the understanding, the workings of the passions, and the various species of sentiment which discriminate vice and virtue. How painful soever this inward search or enquiry may appear, it becomes, in some measure, requisite to those, who would describe with success the obvious and outward appearances of life and manners. The anatomist presents to the eye the most hideous and disagreeable objects; but his science is useful to the painter in delineating even a VENUS or an HELEN. <…> Accuracy is, in every case, advantageous to beauty, and just reasoning to delicate sentiment.⁹ [bolding added]

In the older art system, what we separate today as arts, crafts, and sciences, was joined even at the institutional level:

Instead of the modern art museum, for example, the sixteenth and seventeenth century had the “cabinet of curiosities,” which displayed seashells, clocks, sculptures, and precious stones as a visual table of knowledge.

Besides the cabinets of curiosities, there were anatomical theaters where dissections were open to the public, and botanical gardens attached to universities, which also served as study contexts for painters and illustrators. In Florence, painters belonged to the same guild as apothecaries (Arte dei Medici e Speziali, Doctors and Apothecaries) who supplied them with pigments and varnishes. And some of the natural philosophers of the Enlightenment, like Robert Hooke, were also talented illustrators themselves and accompanied their philosophical treatises with their own meticulous drawings of living things.

Separating the ‘artist’ from the ‘artisan’

Throughout the Renaissance, making art usually was a cooperative affair, with many minds and hands at work, in masters’ workshops, as exemplified by Raphael’s frescoes, Botticelli’s paintings and countless lesser known artifacts. Composers freely borrowed melodies and harmonies from each other, and plays were often adapted and remade in various theater productions. The rise of the singular ‘artist’ whose work is an expression of an individual creative spirit only became a pattern in the 18th century.

Originally, the artist, also known as ‘artificer,’ was a maker, not a creator. In Renaissance terms, while God creates nature out of nothing, nature makes what is potential into the actual. The artificer simply modifies the actuality present in nature. Poets, painters and composers were viewed not as ‘creating’ beauty but as discovering what was already there, by imitating it.

Until the 18th century, there was no sharp distinction between artists and artisans, or between ‘high art’ and ‘low art.’ Art was a part of life, and it was made with the purpose of serving a useful function — in churches, households, courts or town squares. Still, what distinguished exceptional works of art from merely good ones was the grace, invention and imagination of their makers.

But after the 18th-century split, all of those graceful aspects of the older image of the artisan-artist were attributed solely to the artist, whereas the artisan or craftsperson “was said to possess only skill, to work by rule, and to care primarily for money.”¹⁰ What happened?

Diverging paths

Two major societal changes catalyzed the separation of arts from crafts, on the one hand, and arts and humanities from the sciences, on the other, - a picture familiar to us today. The Industrial Revolution, which spanned the 1760s to 1840s, brought about mechanization of many of the processes that were previously the provenance of skilled artists-artisans. And the Romantic movement that emerged towards the end of the 18th century represented a fundamental shift in the way people saw themselves and their relations with the human society and the natural world.

Concomitant with the Industrial Revolution, the artist patronage by wealthy nobility was replaced with an emerging art market and a middle-class art public. Where art was once made for specific purposes (to serve ritual and worship, glorify rulers, embellish furniture and garments), it now had to deliver to the demands of the public. This catalyzed the splitting apart of the old, unified art system into ‘fine arts’ and ‘crafts.’ In a society undergoing industrialization and commercialization, the (fine) arts retreated from the world:

Although Schiller and Goethe never completely abandoned the hope that art might improve society, by the late 1790s, both had given up the eighteenth-century view of art as a means of public enlightenment. Schiller could even write in 1803 that Art should “totally shut itself off from the real world.” Many saw art as fundamentally alien to a society propelled by commerce and industry. Moreover, following the collapse of the Napoleonic empire, the various absolute monarchies of Central and Eastern Europe vigilantly exiled or imprisoned most social and political dissenters, including those who championed a politically engaged art, such as Heine. This made it easier for the idea of a completely autonomous art to gain early acceptance.¹¹ [bolding added]

Schiller and other Romantic thinkers were developing a new idea of the self-contained, standalone work of art that demanded a refined and inward-looking aesthetic response. And in doing so, these thinkers were in fact reacting to their own frustrations with the art market and the new public.¹² Art was to remain free from the concerns of money and class, and it is in this regard that fine arts and crafts started diverging. The fine arts — poetry, music, painting, sculpture — were thus conceived of as a matter of inspiration and genius, standing apart from the world, and were meant to be enjoyed for themselves, privately, in moments of refined pleasure. Both the creation and contemplation of fine art became a private and personal matter. In contrast, the crafts and popular arts like embroidery, pottery and weaving were seen as only requiring skill and rule-following and were meant for wide public use or entertainment.

Interestingly, as artists moved away from guilds and workshops to private studios, scientists moved from private investigations to laboratories and research universities and became more cooperative. The founding of the Royal Society in London in 1660 and the Académie Royale des Sciences in France in 1663 were only the most visible signs of the growing institutionalization of science as a distinct realm of activity. In the 18th-19th centuries, modern institutions of science beyond royal academies included public scientific lectures, laboratory demonstrations and public debates — the hallmarks of Romantic science.¹³ In parallel, separate new institutions were emerging also in the arts: art museums, secular concerts and literary criticism. Romanticism in many ways contrasted itself with the Enlightenment rationalism:

The scientific revolution of the late seventeenth century had promulgated an essentially private, elitist, specialist form of knowledge. Its lingua franca was Latin, and its common currency mathematics. Its audience was a small (if international) circle of scholars and savants. Romantic science, on the other hand, had a new commitment to explain, to educate, to communicate to a general public.¹⁴

Romanticism birthed a new, Faustian, image of a scientist who was a “solitary scientific ‘genius’, thirsting and reckless for knowledge, for its own sake and perhaps at any cost.”¹⁵ It was the time of Alexander von Humboldt, Humphry Davy and the Herschel dynasty of astronomers who became the successors to the natural philosophers of the Enlightenment like Newton, Hooke, Locke and Descartes. This notion of a scientific genius possessing an intense imagination and obsession with discovery echoes the inspired image of the Romantic poet and artist.

But as a movement of contradictions, Romanticism also engendered a suspicion toward science as a method of inquiry that was draining the beauty and mystery and our sense of the transcendent from the universe, by replacing them with a purely mechanistic image. The English Romantic poet Samuel Coleridge, whom we met earlier, spoke in 1816 of the contemporary state of science in France: “We have purchased a few brilliant inventions at the loss of all communion with life and the spirit of nature.”¹⁶ And the Scottish historian and philosopher Thomas Carlyle wrote in 1833, “The progress of science is to destroy wonder, and in its stead substitute mensuration [measurement] and numeration.” This ongoing ‘naturalization’ as a process of transformation of our understanding of the world and ourselves was believed to lead to a disenchantment of the world. This is what Edgar Poe’s Sonnet — To Science (1829) is about:

Science true daughter of Old Time thou art
Who alterest all things with thy peering eyes.
Why preyest thou thus upon the poet’s heart,
Vulture, whose wings are dull realities?
How should he love thee? or how deem thee wise,
Who wouldst not leave him in his wandering
To seek for treasure in the jewelled skies.
Albeit he soared with an undaunted wing?
Hast thou not dragged Diana from her car?
And driven the Hamadryad from the wood
To seek a shelter in some happier star?
Hast thou not torn the Naiad from her flood,
The Elfin from the green grass, and from me
The summer dream beneath the tamarind tree?

In his lecture series on The Roots of Romanticism, Isaiah Berlin highlights this Romantic resistance to “scientists, bureaucrats, persons who made things tidy, smooth Lutheran clergymen, deists, everybody who wanted to put things in boxes.” Coming mainly from the men of letters, such an attitude seems quite in contrast to how the men of science themselves saw their craft at the time — a quest to uncover the secrets of nature through thoughtful and skillful experiments whose design demanded flights of imagination no less audacious than those of the poets and artists of the day.

Thus, from the early decades of the 19th century, the arts/humanities and sciences went their separate ways. ‘Science’ came to be used to describe only empirical sciences (primarily physics, but also chemistry and biology) and had to carve out a space in university curriculums still dominated by humanities.

It is in this atmosphere of separation and growing alienation of the sciences and humanities that the famous debate between the two gentlemen and good friends, the poet Matthew Arnold and the naturalist Thomas Huxley, took place. They articulated some of the most poignant arguments about the place of natural sciences and humanities in human culture that are as apt and relevant today as they were in the late 19th century. In the next post, we will explore these arguments and the subsequent cultural commentary that the debate inspired.

So named for his fierce advocacy of Darwin’s theory of evolution. ↩︎
Pierre Duhem, The Aim and Structure of Physical Theory (1954) ↩︎
Liberal arts, originating in the Greco-Roman tradition, included the trivium of logic, grammar, rhetoric and the quadrivium of arithmetic, geometry, astronomy, and music theory. Mechanical arts were codified in the 12th century and consisted of weaving, armament (including architecture), commerce, agriculture, hunting, medicine, and theatrics (including wrestling, racing and dance). ↩︎
Francis Bacon’s tree of knowledge is based on the three faculties of memory, reason and imagination. He placed history under memory, philosophy under reason and poetry under imagination. The French Enlightenment thinkers (Diderot) added science to the division of reason. ↩︎
In fact, Huxley hated the term, so it is all the more ironic (though in all likelihood unintentionally so) that his biography by Cyril Bibby, published in 1959, is titled T. H. Huxley: Scientist, Humanist, and Educator. ↩︎
Here and below, the quotes are from the wonderful article by Sydney Ross, Scientist: The story of a word (1964). ↩︎
Larry Shiner, The Invention of Art, p. 5. ↩︎
Larry Shiner, The Invention of Art, p. 46. Though the use of ‘scientist’ is of course anachronistic here. ↩︎
David Hume, An Enquiry Concerning Human Understanding, p. 4. ↩︎
Larry Shiner, The Invention of Art, p. 13. ↩︎
Larry Shiner, The Invention of Art, p. 222. ↩︎
Larry Shiner, The Invention of Art, p. 7. ↩︎
Defined in The Age of Wonder by Richard Holmes as the time “between two celebrated voyages of exploration – Captain James Cook’s first round-the-world expedition aboard the Endeavour, begun in 1768, and Charles Darwin’s voyage to the Galapagos islands aboard the Beagle, begun in 1831.” ↩︎
Richard Holmes, The Age of Wonder: The Romantic Generation and the Discovery of the Beauty and Terror of Science, p. 9. ↩︎
Richard Holmes, The Age of Wonder, p. 8. ↩︎
Though Coleridge himself had a keen interest in chemistry, studied it seriously and was close friends with the chemist Humphry Davy, whom Coleridge called “the Man who born first a poet first converted Poetry into Science.” ↩︎

Looking for Hidden Gems in Scientific Literature

aghayeva.u@gmail.com (Ulkar) — Sun, 02 Nov 2025 00:00:00 +0000

This essay was first published on the Elicit blog.

But it really makes no difference whether the unknown lies in the lap of Nature or, instead, is buried among the pages of worthless manuscripts read by no one; because an idea that has not entered the bloodstream of science, and does not circulate seminally in it, in practice does not exist for us.

Stanisław Lem, His Master’s Voice

In the expanding universe of human knowledge, there is intellectual dark matter. Some of it is lost in time without trace,¹ some is only accessible within private institutions (proprietary knowledge) and some can only be learned through direct experience but cannot be legibly recorded (tacit knowledge). Lost knowledge is an unknown unknown. Proprietary and tacit knowledge are a known unknown. But there is also the unknown known: not lost, but forgotten and recoverable knowledge contained in obscure or neglected published research papers. Don Swanson, the founder of the field of literature-based discovery, referred to it as ‘undiscovered public knowledge.’

Why do research papers remain in obscurity, does it matter if they do, and what can be done to redeem them? They may be hard to find to begin with, if they’re published in poorly indexed or low visibility journals or in a foreign language. Alternatively, they may be relatively easy to find but still be neglected if they are ahead of their time and not recognized as important at the time of publication (such papers are called sleeping beauties). It could also be that obscure papers are truly of low quality and are thus justifiably neglected.

But undiscovered public knowledge can also be scattered across even well-read papers whose findings complement each other but are not explicitly linked. A classic example was pioneered by Don Swanson himself in the 1980s, in biology — a field new to him as a physicist by training. In the course of literature review, Swanson noticed that magnesium deprivation causes symptoms similar to those of migraine but that papers investigating either migraine or physiological effects of magnesium had never been jointly cited or connected before. He formulated a hypothesis that dietary magnesium supplementation can alleviate migraine, which was subsequently supported by clinical trials. Swanson also noted that, given the sheer volume of research literature, such connections are bound to exist but remain truly unknown until someone discovers them:

Undocumented connections arise neither by chance nor by design but as a result of the inherent connectedness within the physical or biological world; they are of particular interest because of their potential for being discovered by bringing together the relevant noninteractive literatures, like assembling pieces of a puzzle to reveal an unnoticed, intended, but not unintelligible pattern. The fragmentation of science into specialties makes it likely that there exist innumerable pairs of logically related, mutually isolated literatures.

Scientific literature is vast. There are on the order of 100 million papers² published to date. It is physically impossible for a single researcher or a team of researchers to be aware of, let alone study thoroughly, all existing literature even within their own discipline, let alone across disciplines. In light of this, and given the ‘inherent connectedness’ of natural phenomena that Swanson emphasized, there are bound to exist islands of knowledge in the published corpus that harbor latent connections between each other.

This is the promise of literature-based discovery (LBD) — to search for and reveal already existing, but still hidden, links between concepts, findings, questions and answers within scientific literature that would otherwise take much longer to stumble into. It is an attempt to build a whole elephant from the parts of it that are only known to researchers in different, disconnected fields and subfields.

LBD can be viewed as a part of the ecosystem of ‘assisted’ scientific discovery — computational tools and approaches that aid human researchers at various stages of the discovery process: literature review, hypothesis generation, experiment performance, and data analysis and interpretation. LBD comprises the first two stages: mining scientific literature for novel hypotheses. Other tools and models can be trained directly on raw data, scaling up and speeding up the manual work of unaided human researchers. For example, deep learning models trained on DNA sequences, protein sequences and 3D structures, neural spike trains, imaging data or mass spectrometry data learn to predict novel structures and biological function. On the hardware side, robotic labs such as Emerald Cloud Labs and University of Liverpool’s ‘mobile robotic chemist’ are being developed to automate wet lab work and run programmable experiments.

The first instances of literature-based discovery were accomplished by cumbersome manual screening of papers. But the field has moved much further since. With the primary medium of LBD being natural language, expanding the scope of the field required handling large amounts of text in an automated way. In response to this challenge, an array of methods has been developed in the fields of natural language processing, machine learning, and artificial intelligence.

These methods range from simple statistical operations of lexical statistics to automated reasoning over scientific literature using large language models (LLMs) of today (see Appendix). Along the way, standardized databases, ontologies and knowledge graphs were built to organize the expanding domain-specific vocabularies (including gene interaction networks, biochemical pathways, and chemical reactions). Scaling to increasingly large bodies of knowledge, LBD methods saw a steady increase in algorithmic complexity over time.

But do we observe a corresponding increase in the number of literature-based discoveries made after Swanson’s seminal studies? LBD’s most successful recent application has been in the field of drug repurposing. At the beginning of the COVID-19 pandemic, in late January 2020, BenevolentAI conducted an exhaustive search for potential drugs with combined anti-viral and anti-inflammatory potency for treating acute cases of the disease. Within two days, using their in-house literature-based knowledge graphs and a suite of ML methods, researchers at BenevolentAI screened in silico through 378 drug candidates, narrowing the list down to six. Of these, baricitinib was selected based on its safety profile and high affinity to its target, the AP2-associated protein kinase 1 (AAK1; an endocytosis regulator expressed in cells that are susceptible to the virus). This drug was previously approved for treating rheumatoid arthritis. A promptly conducted clinical trial showed a large positive effect (faster recovery and lower mortality) on the patients, and baricitinib received FDA emergency authorization as soon as November 2020.

Outside of biomedicine, LBD has been applied in some basic science domains. In materials science, it can make retrospective discoveries of functional materials like thermoelectrics when trained on the literature corpus (millions of abstracts) up to a certain cut-off year. That is, LBD can recover findings made after the training cut-off year, and we can verify the top-ranked candidates by checking the papers published in the following years. This is a testament to the core assumption of LBD: that scientific literature alone contains recoverable latent knowledge. But literature-based prospective discoveries in materials science, from the most up-to-date state of the literature into the future, have been lacking.

All in all, as a stand-alone approach, LBD still remains mostly a theoretical undertaking, despite the rich arsenal of computational methods it has at its disposal. Real-world discoveries that make it to clinical trials and approved drugs or basic discoveries in any discipline remain few and far between.³ Why is that?

First off, LBD has an evaluation problem. In computational, prediction-oriented fields like LBD, the output of a method is often a ranked list of candidates (hypotheses in this case). In order to be able to compare different methods within the field to each other and measure the field’s progress as a whole, we need shared scalable and reproducible benchmarks and evaluation methods. Considering that the real-world, experimental validation of LBD hypotheses, especially in biomedical sciences, is slow and costly, scalable in silico evaluation becomes even more important. Evaluation methods in LBD also formally define the target task: if a research community can agree on how to measure success, it can precisely define what LBD is trying to do.

This poses the question of what exactly counts as a literature-based discovery and what its boundaries are. What does establishing a hidden link between scientific concepts mean? According to one definition, ‘a discovery represents the linking of two or more concepts that had never previously been linked in order to produce novel, interesting, plausible, and intelligible knowledge’ (emphasis added). Such a definition leaves room for interpretation, making it challenging to apply analytical methods to it. Additionally, the definition of discovery may vary by domain (e.g. in social sciences vs biology).

For more than three decades since its founding, LBD evaluation has been mostly relying on the same small set of discoveries as a benchmark: Swanson’s magnesium – migraine and Raynaud disease – fish oil links, along with a few others. This is the so-called replication evaluation: to test whether a new LBD method can re-discover known cases that were previously found manually (in other words, whether it can rank those cases highly).

Are these discoveries an appropriate gold standard for LBD? Their number is small to begin with, and even when a replication works, it doesn’t differentiate whether one method is quantitatively better than another. The test set covers only a tiny portion of biomedical sciences, and there is no representation of other disciplines. More fundamentally, evaluation in knowledge-based discovery is intrinsically hard, since LBD methods surface putatively new knowledge that has not yet been empirically tested. Validation against a known set of hand-picked discoveries is insufficient and provides a weak theoretical foundation for the field.

These limitations are partially addressed in an alternative evaluation method, time-sliced evaluation. In this case, pairwise associations between concepts are not limited by the famous manual set but are searched for in the entire scientific corpus after a certain cut-off time. Every pair that did not co-occur before the cut-off time but starts to co-occur after it is treated as a positive. Pairs that don’t co-occur provide negatives. Highly ranked positive pairs are candidates for real discoveries.

But this method has been beset by its own shortcomings. For one, mere co-mentions do not in and of themselves constitute a discovery – there’s no way around human curation. Most co-occurrences are trivial or noisy, like pairings with generic terms (‘diabetes’ and ‘blood sugar’ or many diseases and ‘inflammation’). A high score assigned by an LBD method thus doesn’t necessarily point to a novel, valuable and generative connection.

The inherent difficulty of establishing evaluation metrics in LBD is what accounts for the fact that historically it has been easier to make a technological contribution to LBD, by developing a new algorithmic method, than to put together a high-quality annotated dataset for evaluation across the field.

Separately from the evaluation problem, it may be argued that language is ‘a lossy abstraction over reality,’ compared to raw data, so LBD may be inherently unreliable. Swaths of tacit knowledge in scientific practice as well as private explicit knowledge (the ‘known unknown’) remain beyond the provenance of recorded literature and can only be acquired through in-person training. And the scientific literature is plagued by non-replicable findings and journal paywalls, further complicating its usefulness as a jumping-off point for generating new knowledge.

Despite its original promise, as LBD researchers themselves admit, there appears to have been relatively little uptake of the LBD methods by their intended end users, active scientists, or few if any large-scale collaborations between LBD researchers and scientists in other disciplines. Even within LBD itself as a field, there are many small and medium-sized groups with little collaboration among them. Ironically, LBD seems to suffer from the same problem it was originally meant to solve: to bridge islands of knowledge and isolated research communities.

Still, the scientific literature is a repository (however noisy and incomplete) of cumulative knowledge of humanity. We would be leaving a lot of value on the table by not digging this knowledge up and looking for latent connections and ideas that are waiting for their due appraisal.

What do large language models add to literature-based discovery?

Pre-LLM methods of literature-based discovery were developed to detect pairwise links (like drug → disease) or linear chains of concepts in the scientific literature (like drug → gene → pathway → disease in knowledge graphs). But such simple relations limit the range and expressivity of hypotheses that could be generated. A major novelty that LLMs bring in is their ability to follow chains of thought in natural language, turning LBD into a reasoning task. LLMs can take in more inputs besides a starting concept, including problem context and constraints, and formulate hypotheses the way a human scientist would express them, in principle.

But there are obstacles on the way to reliable and valuable LLM-generated hypotheses. Hallucination is a major concern because LLMs are trained to predict likely tokens based on prompts, and optimizing for plausibility does not guarantee truthful statements. One way to ground LLM outputs in existing literature is retrieval-augmented generation (RAG). Another approach is to implement self-critique where the model first drafts an answer to a prompt and then revises it, based on its own verification process (Chain-of-Verification, CRITIC, Self-RAG, Test-Time Diffusion Deep Researcher). But ultimately, LLM hypotheses have to be quality-controlled by human expert curation and empirical validation in the lab.

In some research fields, earlier LBD methods are still preferable to LLMs, due to the idiosyncrasies of their datasets. For example, most sociologists still rely on static semantic embeddings when they study cultural shifts using historical texts, for a few reasons. Computations involved in constructing such embeddings are easy to interpret — they are based on pointwise mutual information in word associations. LLM embeddings are largely opaque and can’t be reliably tracked to the statistical properties of the underlying training data.

Most importantly for the study design, pre-trained LLMs are prone to the ‘look-ahead’ bias where the model’s analysis of a historical text is affected by its pre-training exposure to the information from the future relative to that historical period. This leakage of future information renders LLM analysis nearly useless, and there are currently no reliable methods to make pre-trained models forget parts of their training data. While simpler semantic models can be relatively cheaply trained on separate historical corpora, compute costs make this infeasible for LLMs, at least in the context of academic research.

Could LLMs generate truly novel hypotheses?

In biomedical sciences to date, there have been few examples of genuine discoveries made by LLMs. Among them, Google’s AI Co-Scientist suggested an ingenious hypothesis (highest-ranked among five it generated) about a mechanism of a horizontal gene transfer among bacteriophages, after being exposed to publicly available research context on the problem. It turned out that this was the same hypothesis that had already been experimentally validated as a result of years-long, as yet unpublished work by the research group that tested the AI Co-Scientist. Arguably, this is an impressive feat and a significant step toward AI-assisted scientific reasoning.

Other, more modest, emerging studies also document an LLM-assisted discovery process where LLMs supply new hypotheses, followed by laboratory validation. Novel metabolomic findings in the yeast Saccharomyces cerevisiae and synergistic drug interactions in breast cancer cell lines count among such studies. But so far they have been limited to incremental findings in simple systems like cell cultures or microorganisms that allow for high-speed experimental testing of LLM-generated hypotheses.

But all in all, we haven’t witnessed a massive surge of breakthroughs that could be expected from models trained on all of human knowledge.⁴ Maybe there simply hasn’t been enough time to empirically validate all the potentially valuable LLM-generated hypotheses yet. Alternatively, could it be that LLMs lack some fundamental features of human thought that enable scientific creativity?

In her book The Creative Mind: Myths and Mechanisms, Margaret Boden considers three types of human creativity, which she defines as the ability to come up with ideas or artefacts that are ‘new, surprising and valuable.’ Combinatorial creativity generates unfamiliar pairings of known ideas. Such combinations can be seen in poetic imagery, humor, metaphor — and classic literature-based discovery. Exploratory creativity involves searching new regions within an existing conceptual space like painting, cuisine or a research area. In such a structured style of thought, finding a previously unnoticed route enriches that space and adds more detail to its map. Tracing new multi-hop chains of concepts along a knowledge graph is an example of exploratory thinking in LBD (from drug to disease or from a regulatory gene to its target).

In contrast, transformative creativity alters the concept-space itself and expands it beyond its pre-existing boundaries. Conceptual leaps are thoughts that ‘are now possible which previously (within the untransformed space) were literally inconceivable.’ These leaps require new representational primitives (‘bootstrapping’), or new representations of observed or imagined entities, that are absent from the language and need to be invented first. The bar here is very high and is reserved for peaks of human achievement: invention of calculus, Fourier analysis, linear perspective in painting, the periodic table, quantum mechanics, chemiosmotic theory (understanding cellular energetics), information theory, cracking the genetic code.⁵

LLMs can excel in combinatorial and exploratory creativity which are possible within the existing vocabulary that they are trained on.⁶ The classic memorization vs. generalization/grokking trade-off (or phase transition) in LLMs maps onto these two types of creativity. Equipped with a rich repository of memorized knowledge, a model can produce potentially interesting concept combinations. Generalization, in contrast, implies that the model has developed a map of the conceptual space and can explore it beyond the known paths, though still remaining within it.

LLMs’ combinatorial and exploratory creativity shines in their ability to solve complex (IMO-level) math problems and assist human researchers in mathematical research. Terence Tao notes that there’s ‘a role for these tools in drawing out a user’s latent knowledge in a problem’, by ‘proposing reasonably relevant ideas that the user is expert enough to evaluate.’ In line with this, Scott Aaronson recently published a significant result in quantum complexity theory where a key technical step came from GPT5-Thinking — though arriving at the correct solution took several iterations, not unlike a conversation with a graduate student or a colleague. Undoubtedly, cases like this can considerably speed up the discovery process. As Aaronson puts it:

Right now, it [an LLM] almost certainly can’t write the whole research paper (at least if you want it to be correct and good), but it can help you get unstuck if you otherwise know what you’re doing, which you might call a sweet spot.

We are yet to witness truly transformational creativity in LLMs, however — scientific, mathematical or otherwise. But all three types of creativity include not only generating new links or thought-paths or representations but also evaluating them, recognizing their importance. These two functions, of a generator and a critic, may need to be separated if they are to be implemented in an LLM-based system.

Is it possible to approach LLM creativity in a more systematic way?

As reasoning entities, LLMs ‘are frozen, unable to learn from experience, and … have no “default mode” for background processing, a source of spontaneous human insight’.⁷ Gwern’s solution to remedy these shortcomings is very similar to the general logic of classic literature-based discovery:

a day-dreaming loop (DDL): a background process that continuously samples pairs of concepts from memory. A generator model explores non-obvious links between them, and a critic model filters the results for genuinely valuable ideas. These discoveries are fed back into the system’s memory, creating a compounding feedback loop where new ideas themselves become seeds for future combinations.

With both a generator and a critic module in place, such a ‘daydreaming’ model would in principle be able to fully automate LBD. This model could be asked to regularly retrieve random sets of concepts from the continually updated scientific literature, then execute a ‘brainstorm’ prompt (evaluate which pairs are worth pursuing further), followed by a judgment of the critic module.

But implementing a daydreaming algorithm would be significantly more expensive than training the LLM itself (what Gwern calls ‘a daydreaming tax’). This is because background sampling of pairs of concepts is inherently wasteful – most pairs won’t be interesting. And there are no shortcuts around this, since if we already knew how to predict the value and interestingness of associations, there wouldn’t have been a need to traverse the entire space of possibilities to begin with. It may be that ‘the most far-flung and low-prior connections are the important ones,’ so none can be discarded a priori.

Donald Swanson’s classic magnesium–migraine link discovery required an unaided scholar to sift through research papers for weeks and months. A daydreaming loop in an LLM can potentially perform billions of such pairings in a much shorter time (hours?). The DDL algorithm could therefore realize the original promise of literature-based discovery – surfacing high‑value ‘unknown knowns,’ at scale.

Appendix

Evolution of computational methods used in literature-based discovery before LLMs

The first implementations of literature-based discovery, like Swanson’s magnesium–migraine association, are known as the ABC model. This model is based on the inference that if concept A is associated with concept B in one set of research papers, and B is associated with concept C in another, disjoint set of papers,⁸ then A is associated with C through the B-concept as an intermediate link. In Swanson’s example, migraine is concept A, magnesium is concept C, and the mechanisms connecting these terms – such as prostaglandins, vascular health, calcium channel blockers – are the B-concepts.

In the evolution of LBD methods, we can observe two broad trends: computational models used in LBD have grown dramatically in size (number of parameters) and algorithmic complexity, while the need for expensive human oversight has been slowly diminishing.

Yet this evolution has been far from linear. Each new wave brought its own requirements for human expertise, and even today’s most sophisticated systems, large language model (LLM)-based research agents, demand substantial supervision. In the limit, LBD tools should be able to take in a question from a human researcher and do all search, synthesis, and plausible hypothesis generation on their own. We are still in the early days of such implements.

The earliest LBD approaches from the 1990s were remarkably simple – they merely counted how often words appeared together in papers (lexical statistics). In doing so, they treated research articles as ‘bags of words,’ without regard for meaning or context. This approach isn’t able to distinguish between sentences like ‘a dog bites a man’ and ‘a man bites a dog,’ since order and syntax are not taken into account in mere word counting. Neither does it distinguish between a statement and its negation – ‘X affects Y’ and ‘X doesn’t affect Y’ look the same, as X and Y co-occur in both cases. And since words may co-occur by chance rather than meaningful association, metrics used in lexical statistics can catch many spurious connections. Human judgement here is necessary to separate the wheat from the chaff.

In general, lexical statistics hasn’t been found powerful and selective enough to be used as a stand-alone method, though it has been incorporated into a suite of methods developed later. But a few studies from the 1990s have relied exclusively on lexical statistics to rediscover Swanson’s Raynaud disease – fish oil and magnesium – migraine links, demonstrating that even such a noisy and simple method can detect high-signal connections in the literature, though not without human curation.

Semantic methods

A transformative breakthrough in LBD came with distributional semantics in the 2000s-2010s, particularly through word2vec. The underlying insight here is the distributional hypothesis, best summarized by the English linguist J.R. Firth in his 1957 paper as ‘a word is characterized by the company it keeps.’ Instead of raw word counts, these methods operate at the level of word meanings (semantics), by placing every term in a high-dimensional vector space based on its neighboring words. For example, a word like insulin is meaningfully associated with other terms like glucose, glycaemia, diabetes, pancreas, pancreatic β-cells and so on. In word2vec, these semantically related words translate into similar vectors. This in turn makes it possible to generalize beyond exact word matches and handle synonyms and analogies in papers computationally. Word2vec can also disambiguate syntactic relations like ‘a dog bites a man’ and ‘a man bites a dog,’ thanks to its contextual sensitivity.

Word2vec thus represents a major inflection point in LBD – a transition from word counting to semantic reasoning. From then on, semantics became an indispensable part of the LBD process. With the training corpus vocabularies counting millions of words and vector space dimensions of 100-300, the model size (measured as their product) jumped to hundreds of millions of parameters, a massive increase compared to the lexical statistics methods.

Ontologies and Knowledge Databases

Around the same time when distributional semantics was entering the LBD toolkit, new ways of organizing and systematizing scientific knowledge enabled novel approaches in the field. In the late 1990s – early 2000s, researchers in biomedicine began organizing knowledge into structured databases and graph networks, to help navigate the expanding biological vocabulary. A few characteristic problems had previously made this task challenging: a single gene or protein often went by multiple names (for example, the epidermal growth factor receptor in humans is known as EGFR, ErbB‑1 or HER1), and naming conventions varied across species. Without a shared reference system, comparing results between species was difficult.

Systems like the Gene Ontology (GO) solved this by assigning stable identifiers – ontology terms – to biological processes (e.g. digestion, glycolysis, photosynthesis), molecular functions (e.g. DNA replication, splicing, ubiquitination) and cellular components (e.g. ribosome, nucleus, cell membrane) to which the contents of biomedical research papers could be mapped. For example, a glycolytic enzyme like hexokinase, in any species, maps onto GO:0006006, glucose metabolic process. Building this shared reference system required a lot of manual labor upfront for annotating papers and coming up with nested classification categories making up an ontology. With these systems in place, literature-based discovery systems could now index papers by concrete ontology terms and compare findings across studies and species more easily.

Also in the 2000s–2010s, emerging high-throughput technologies like sequencing, microarrays and proteomics generated massive datasets that needed to be integrated and interpreted across labs. Databases like KEGG (Kyoto Encyclopedia of Genes and Genomes), UniProt (protein sequence and function database) and Reactome (biochemical pathway library) began mapping genes and proteins onto curated pathways. Beyond a mere cataloguing of biological entities, these databases were designed to highlight interactions between them – which molecule activates or inhibits which other ones or which gene regulates which targets.

Graph-based methods

In light of this, graph-based methods of LBD became indispensable. Graphs represent knowledge as a network of connected concepts (nodes), with edges corresponding to relations between them. The main innovation here is the ability to trace longer chains of reasoning rather than just looking at pairs of related terms.⁹ LBD is abstracted as a link prediction (or missing edge) problem within the graph: given only the edges observed so far, which pairs of nodes are most likely to be connected? If the nodes already share a large number of common neighbors, it is more likely that there are still remaining hidden links that connect these nodes directly, pointing to a previously missed discovery. Parameter counts in knowledge graphs are in the tens to hundreds of millions, and they still require a lot of human curation.

Another innovation of knowledge graphs is that in addition to technical terms (of different types), they can include metadata like papers, authors, their affiliations and citations. Such graphs are called heterogeneous, or hetnets. They can detect cross-field patterns and directions of knowledge flow that are impossible to infer from the methods relying on the text of research papers alone. An example of a hetnet is Hetionet, which is assembled from 29 different databases of genes, bioactive compounds, diseases, drug side effects and more, consisting of 47,031 nodes (11 types) and 2,250,197 relationships (24 types).

Machine learning methods

In the 2010s, machine learning entered the picture. Broadly, ML models are those that learn from data without explicit programming and can identify patterns and make predictions about unseen data. Applied to Swanson’s ABC model, classifier ML algorithms like support vector machines and random forests can be trained to rank intermediate B terms between A and C terms, and can filter candidate B lists much faster than human researchers. But since they are trained on examples of already known links, these algorithms come up with more conservative and incremental rather than truly novel hypotheses.

Next wave of LBD methods, from the late 2010s, consisted of deep embedding methods like DeepWalk and node2vec. These ML models convert an entire biomedical knowledge graph into a geometric space where every concept is a point, and the distance between them indicates how strongly two ideas are implicitly related. Such mapping lets the algorithm filter through millions of papers at once and find pairs of concepts that don’t co‑occur in the existing literature but that nonetheless share rich structural context within the graph (for example, a metabolic pathway and a drug that hasn’t been previously tested for its effects on this pathway).

By acting on the graph rather than raw text, as word2vec does, deep embeddings can trace multi‑step patterns, starting from a gene all the way to a disease, and find cross‑disciplinary connections. Their use cases range from disease etiology to drug-disease associations to protein-protein interactions. Still, a deep embedding map is a frozen snapshot of a graph, and it has to be rebuilt each time to integrate new literature. And of course, the quality of the insight derived from it still depends on how carefully the underlying graph was curated.

Graph neural networks are an improvement over the static maps of deep embedding, as they are able to dynamically update node representations in knowledge graphs and capture contextual nuances better. This contextual awareness allows GNNs to trace more sophisticated reasoning paths and, crucially, to scale to graphs with millions of nodes. GNNs have been shown to improve LBD performance 2-4 fold compared to co-occurrence methods. Still, they have similar limitations to deep embedding methods, as they can be biased by highly connected nodes, and adding new nodes after the initial training requires retraining of the model.

Including most Greek tragedies, the Voynich Manuscript (unknown script), and the recipe for making Damascus steel, among many others. ↩︎
And more than 200 million items with DOIs, including journal articles, book chapters, conference abstracts, preprints etc. ↩︎
There are a number of studies that use LBD to find candidate disease-associated genes (for example, for polymicrogyria, a brain malformation), candidate treatments (for example, for Parkinson’s and Alzheimer’s) or predicted adverse drug effects but they have so far not translated into novel empirically confirmed discoveries or real-world treatments. ↩︎
For the purposes of this essay, we won’t consider foundational models trained on raw data and discoveries they have enabled but instead focus on LLMs in relation to scientific literature. ↩︎
There are parallels between Boden’s combinatorial, exploratory and transformative creativity and the concepts of interpolation, extrapolation and hyperpolation in machine learning. ↩︎
Such creativity has also been demonstrated by earlier, much simpler algorithms like AARON (autonomous painter), EMI (Experiments in Musical Intelligence) and BACON (data-driven rediscovery of quantitative laws), among others. ↩︎
In human brains, ‘default mode’ is the neural activity pattern that is observed when one doesn’t actively engage with the external world and is instead given to mind-wandering and daydreaming. ↩︎
That is, papers that don’t cite them and are not cited by them. ↩︎
This is known as the AnC discovery model, where the starting term A is connected to the target term C through a number of intermediate nodes on the graph. ↩︎

Multiple discoveries and the nature of scientific knowledge

aghayeva.u@gmail.com (Ulkar) — Tue, 30 Apr 2024 00:00:00 +0000

“Time is the greatest innovator.” - Francis Bacon

In the late 1980s-early 2000s, there was a famous debate between Stephen Jay Gould and Simon Conway Morris about the nature of the evolutionary process. Gould thought that evolution is radically contingent: if “the tape of life” were to be rerun, even a minute change would result in widely divergent outcomes, and the likelihood that intelligent life would evolve again in such a rerun is vanishingly small. Indeed, according to Gould, “no important and sufficiently specific evolutionary outcomes are robustly replicable”¹.

Conway Morris agreed with Gould that historical contingencies are pervasive in evolution but:

Put simply, contingency is inevitable, but unremarkable. It need not provoke discussion, because it matters not. There are not an unlimited number of ways of doing something. For all its exuberance, the forms of life are restricted and channeled.

In other words, if the life forms are solving problems in a global search space, there’s a limited number of good solutions to various design problems (limited by laws of physics and chemistry), and by moving towards those optimal solutions, species will eventually acquire characteristics of these options, irrespective of their starting position.

This debate has far-reaching implications for complex search processes not unlike the scientific enterprise.

What, if anything, does the pervasiveness of multiple discoveries tell us about the nature of scientific knowledge? It seems intuitive to think that if a result is obtained independently by multiple researchers, it must be in some sense inevitable, almost forcing itself upon us, and it is likely to reflect something true about the world². In that sense, multiple discoveries are an instantiation of the “inevitability thesis” that lies at the heart of the contingency vs inevitability debate in philosophy of science: “If the result of the scientific investigation of a certain subject matter is correct, then any investigation of the same subject matter, if successful, will yield basically the same result.”

The implication here is that the results of successful science are essentially contained within a final, complete state of scientific knowledge that our collective inquiry is approaching, and in that sense are inevitable. This is a stronger claim than to say that, given enough time and resources, it is inevitable that correct results will be obtained — though the distinction is arguably fine. And one could say that the two statements entail each other. But it is eventually about tying inevitability to a final state of scientific truth (if such a thing exists) and not the research process itself.

In any case, upon a closer look, multiple discoveries may not be ironclad evidence for the inevitability side of the debate.

Contingency in the context of scientific research can mean many things. Experimental results, methods, theories, instruments and devices used to make experimental measurements, identities and boundaries of whole disciplines, scientific frameworks and paradigms — all of these are a result of historical contingencies, a result of choices made, collectively agreed and built upon by the scientific community³.

The existence and abundance of multiple discoveries can be easily accounted for in this view, just as in the inevitability thesis. Since scientists working in a given domain researching the same question have been exposed to the same background knowledge in that domain (Thomas Kuhn would say, the same paradigm), they operate in the same branch of the possible theory space, and it is no wonder that they arrive at the same results. This does not eliminate the possibility of existence of parallel unexplored branches that may be as successful in explaining and predicting the behavior of systems under study. There could be physics without quarks and biology without genes, which is hard to imagine for us who have bought into the current branch — but not unimaginable.

real science fiction (alternative histories of science) has never been tried. or has it?
— Ulkar (@ulkar_aghayeva) April 19, 2024

Researchers may work independently from each other but not really be independent from the state of contemporary scientific knowledge in a given field. This background knowledge includes all accessible knowledge that a scientist implicitly or explicitly relies on and eventually uses to interpret a new finding. When a new result (or, for that matter, a multiple discovery) is added to this repository of background knowledge, that does not in itself say anything conclusive about whether the result should be interpreted in realist or antirealist terms. That is because, to an antirealist, the mere acceptance and coherence of new knowledge with the existing body of knowledge does not mean that it necessarily reflects a truth about the world. So it appears that multiple discoveries are neutral with respect to both the contingency/inevitability and realist/antirealist philosophical divides.

With that said, science education and practice are firmly, if implicitly, based on realist and inevitabilist views of scientific knowledge. Contingency and antirealism seem to be relegated to philosophy of science departments and are rarely discussed by scientists themselves. All the more I think it’s worth diving deeper into them, and that’s what the rest of this post is about.

The Self-Vindication of Laboratory Sciences

‘A rule is amended if it yields an inference we are unwilling to accept; an inference is rejected if it violates a rule that we are unwilling to amend.’— Nelson Goodman

One of the major researchers of the contingency vs inevitability debate is Ian Hacking. The title of this section links to his paper which looks into how components of the research process — theories, measurement instruments and practices, and data processing practices — can adjust to each other over time, resulting in a stabilization of science. This stable, internally consistent structure of laboratory sciences may be quite independent from the external world it is supposed to describe.

Despite our recent enthusiasm for refutation and revolution, these sciences lead to an extraordinary amount of rather permanent knowledge, devices, and practice. It has been too little noted of late how much of a science, once in place, stays with us, modified but not refuted, reworked but persistent, seldom acknowledged but taken for granted.

First of all, what does Hacking mean by laboratory sciences? In his definition,

the “prototype” laboratory sciences are those whose claims to truth answer primarily to work done in the laboratory⁴. They study phenomena that seldom or never occur in a pure state before people have brought them under surveillance. Exaggerating a little, I say that the phenomena under study are created in the laboratory.

To be fair, this definition is contestable, as anything from cell divisions to animals’ foraging behavior happens, of course, all the time outside of the lab. Trees fall in forests even when people aren’t watching. But a fair share of phenomena is indeed being created in the laboratory — from photoelectric effect to lasers to chemical synthesis to grafting tumors in mice for testing anti-cancer drugs.

Disciplines that are mainly observational, taxonomical or historical — like botany and paleontology — are not included among laboratory sciences, even though they may involve methodologies that are carried out in a lab setting. The status of astronomy, astrophysics and cosmology is ambiguous since they don’t generally interfere with the phenomena that they study.

So what does self-vindication mean in this context? It means that any test of theory happens through the use of instruments and devices that evolved in conjunction with it and in conjunction with practices of data analysis. And conversely, the working of the instruments and correctness of analysis is judged by their fit with the theory. Theoretical assumptions may be “built into the apparatus itself”, as Peter Galison put it in his book How experiments end. This makes them a “closed system” (as Heisenberg referred to Newtonian mechanics) that is essentially irrefutable.

The theories of the laboratory sciences are not directly compared to “the world”; they persist because they are true to phenomena produced or even created by apparatus in the laboratory and are measured by instruments that we have engineered. This “true to” is not a matter of direct comparison between theory and phenomenon but relies on further theories, namely, theories about how the apparatus works and on a number of quite different kinds of techniques for processing the data that we generate. High-level theories are not “true” at all. This is not some deep insight into truth but a mundane fact familiar since the work of Norman Campbell (1920, 122-58), who noted that fundamental laws of nature do not directly “hook on to” the discernible world at all. What meshes (Kuhn’s word) is at most a network of theories, models, approximations, together with understandings of the workings of our instruments and apparatus. <…> Our preserved theories and the world fit together so snugly less because we have found out how the world is than because we have tailored each to the other⁵.

But, Hacking warns, “This is not to suggest that they are mental or social constructs.” He argues “[not for] idealism but rather for down-to-earth materialism.” Also, his thesis is about mature science rather than science at the cutting edge:

That can be as unstable as you please, even when it is what Kuhn called normal science. As a matter of fact such research is usually highly regimented. Results are more often expected than surprising. We well understand why: it is not that sort of short-term stability that is puzzling. I am concerned with the cumulative establishment of scientific knowledge. That has been proceeding apace since the scientific revolution.

Hacking notes that his thesis on the self-vindication of laboratory sciences is compatible with both realist and antirealist views, except the realist assertion that “the ultimate aim or ideal of science is ’the one true theory about the universe.’” This is because of incommensurability — in the literal sense of the lack of common measure — of theories enmeshed and co-evolved with different types of instruments and measurement practices, while each may remain “true to” their data domain.

It used to be said that Newtonian and relativistic theory were incommensurable because the statements of one could not be expressed in the other — meanings changed. Instead I suggest that one is true to one body of measurements given by one class of instruments, while the other is true to another.

Likewise, Heisenberg wrote in 1948:

some theories seem to be susceptible of no improvement… they signify a closed system of knowledge. I believe that Newtonian mechanics cannot be improved at all… with that degree of accuracy with which the phenomena can be described by the Newtonian concepts, the Newtonian laws are also valid.

To this Hacking adds, “It is rather certain measurements of the phenomena, generated by a certain class of what might be called Newtonian instruments, that mesh with Newtonian concepts.” The accuracy of theory and accuracy of instruments complement each other and depend on each other.

Hacking’s thesis is also compatible with both sides of the contingency vs inevitability debate, albeit at different levels. If mature science is self-stabilizing in the way he describes, its findings may become highly constrained by the interlocking elements of laboratory research, and thus in a sense inevitable.

But this self-stabilization is a contingent process in itself, since the particular theories, instruments, and practices of a mature science are not predetermined, but are a result of a historical process of mutual adjustment and co-evolution. While mature sciences may be stable, they could have stabilized around different research paths given different starting points.

Hacking’s thesis can be summarized as a coherence theory — except not of truth but of the entire scientific enterprise. Scientific stability and progress are brought about by the mutual adjustment of theory, experiment, and data into a coherent whole. Truth in this picture is a way of expressing our commitment to this evolving system rather than a matter of describing an independent external reality.

Ok! But how about vaccines and drugs and materials that came out of the lab and have been tested and clearly work in the real world? Hacking says they work because we’ve essentially remade much of the world in the image of the lab. Success of the scientific enterprise does not necessarily imply that our theories have captured preexisting truths about the world. Instead, we reshape our environment so that it behaves in accordance with our theories:

We remake little bits of our environment so that they reproduce phenomena first generated in a pure state in the laboratory. The reproduction is seldom perfect. We need more than the topical hypotheses and the modeling of the laboratory apparatus; we need more thinking of the same kind as those. But the application of laboratory science to a part of the world remade into a quasi-laboratory is not problematic, not miraculous, but rather a matter of hard work.

This is a radical statement, to be sure. I can’t accept it wholesale but it does prompt the question: How much of the world can we indeed remake in the image of the lab? What proportion of lab science does not translate into solving real life problems because of its disconnect with the world at large, while being internally consistent?

Hacking’s paper suggests that these limits to application of scientific knowledge aren’t just a matter of our theories being incomplete or imperfect. Rather, they reflect the fact that the world is in fact not a lab and can never be made to conform perfectly to our lab-based understanding. There will always be some discrepancy between the phenomena we create in the lab and the complex reality of the world.

The success of science comes from our ability to make the world more lab-like; the limitations come from the inherent resistance of reality to this remaking.

Many thanks to Amanuel Sahilu and Joe Walker for their thoughtful feedback on a draft of this piece.

If you have comments on this post, feel free to get in touch on twitter/X or email me at aghayeva.u@gmail.com.

This was illustrated vividly by Ray Bradbury in his short story “A sound of thunder”. ↩︎
The latter is a claim of scientific realism which I will return to below. To a first approximation, scientific realism is the view that well-confirmed scientific theories are approximately true; the entities they postulate do exist; and we have good reason to believe their main tenets. ↩︎
Discussed in Science as it could have been, p. 19. ↩︎
where a laboratory is defined as “a space for interfering under controllable and isolable conditions with matter and energy.” ↩︎
Elsewhere he writes: “The most succinct statement of the idea, for purely intellectual operations, is Nelson Goodman’s summary (1983, 64) of how we “justify” both deduction and induction: ‘A rule is amended if it yields an inference we are unwilling to accept; an inference is rejected if it violates a rule that we are unwilling to amend.” ↩︎

Multiple discoveries - a rule rather than an exception?

aghayeva.u@gmail.com (Ulkar) — Wed, 25 Oct 2023 00:00:00 +0000

"…There is even a stronger feeling inspired of the presence of that Truth to which we all profess to minister, when we find our own discoveries, such as they are, coincide independently with the discoveries of other men. The voice which is heard by two at once appears to be more real and external - one is more sure that it is no personal and private fancy, no idiosyncratic peculiarity, no ringing in sick ears, no flashes seen by rubbing our own eyes."

William Rowan Hamilton, in a letter to John Herschel, Nov 23, 1846

Merton’s hypothesis states that all scientific discoveries are in principle multiples, that it is the singletons that require special explanation and that, on a closer look, they often end up being multiples.

As first, the hypothesis may seem a bit far-out and immune to testing and verification. One can declare even the historically known singletons to be multiples-in-principle or would-have-been multiples, since we don’t have a perfect knowledge of history, and so absence of evidence (that they are multiples) is not evidence of absence. But there’s no need for such sophistry. Merton presents abundant evidence supporting the hypothesis which I will examine below. Moreover, the way scientists work is already built on an assumtpion that multiple discoveries (MDs) happen all the time and are a predominant pattern in the course of scientific progress, rather than a peculiar exception.

The first class of evidence has to do with rediscoveries:

Rediscoveries of previously unpublished work, as in the cases of Cavendish and Gauss discussed in the previous post, is one mechanism by which discoveries thought to be singletons turn out to be multiples.

A special category of rediscovery deserves its own subsection:

Curious Cases of Cryptomnesia among Scientists

There are many records of scientists who found, to their own amusement and incredulity, that an idea that seemed to them new and original at the time had in fact been written down or even already published by them years ago, and then forgotten - cases of self-rediscovery.

From the correspondence of Joseph Priestley, the English chemist and discoverer of oxygen¹:

I have so completely forgotten what I have myself published, that in reading my own writings, what I find in them often appears perfectly new to me, and I have more than once made experiments, the results of which had been published by me.

Augustus de Morgan, a 19th century British mathematician, had his own cheerful version of the experience²:

I have read a Paper (but not on mathematics) before now, have said to myself, I perfectly agree with this man, he is a very sensible fellow, and have found out at last that it was an old Paper of my own I was reading, and very much flattered I was with my own unbiased testimony to my own merits.

Cryptomnesia refers to the fact that sometimes an idea that seems fresh and original to its author may in reality be the trace of a forgotten exposure to an idea of another (or even themselves) - something that they’ve read or heard before but have no memory of having encountered. So self-rediscovery is also a case of cryptomnesia.

Awareness of the possibility of cryptomnesia can undermine the certainty that one has indeed come up with a new idea oneself, when encountering another version of the same idea developed by someone else (this is somewhat akin to the anxiety of influence in literature and music).

After William Rowan Hamilton, of quaternions fame, learned at 19 that his discovery in optics was “only” a rediscovery, he developed a lifelong morbid fear of being plagiarized and unwittingly plagiarizing others. In a letter to his friend de Morgan (same de Morgan who had a case of self-rediscovery), he wrote:

As to myself, I am sure that I must have often reproduced things which I had read long before, without being able to identify them as belonging to other persons.

and

Where is the priority business to end? I am sick of it as you can be; but still, in anything important as regards science, I should take it as a favour to be warned, if I were inadvertently exposing myself to the charge of plagiarising.

With the rise of collaborative research, the risk of cryptomnesia has likely fallen, since even if one member of the team forgets about an earlier exposure to an idea, others probably won’t. But it is remarkable how often it had happened in the past when research was largely conducted by individual scientists.

Let’s now turn to another class of evidence for Merton’s hypothesis, forestalled multiples - “discoveries that are historically identified as singletons only because the public report of the discovery forestalled others from making it independently.”³

There are reports in print stating that a scientist has discontinued research on a particular problem, well along the way, because a new publication has anticipated both their hypothesis and the experimental design to test the hypothesis. In other cases, despite having been anticipated, scientists still go ahead with the publication of their original work and often lament in the footnotes of their papers along the lines of “Since completing this experiment, I found that so-and-so had arrived at this conclusion last year and that so-and-so did it 50 years ago.”

This is from Lord Kelvin’s biography:

The experience of Lord Kelvin as an undergraduate of 18, when he was still the untitled William Thomson, who sent his first paper on mathematics to the Cambridge Journal only to find that he “had been anticipated by M. Chasles, the eminent French geometrician in two points . . . [and] when the paper appeared some months later, prefixed a reference to M. Chasles’ memoirs, and to another similar memoir by M. Sturm. Still later, Thomson discovered that the same theorems had been also stated and proved by Gauss; and, after all this, he found that these theorems had been discovered and fully published more than ten years previously by Green, whose scarce work he never saw till 1845.”

(You may start noticing that prominent scientists are especially prone to being involved in numerous multiples throughout their research careers - I will return to this point later.)

Many scientists cannot bring themselves to report in print that they were forestalled. These cases are usually known only to a close circle of friends and collaborators who are familiar with their work. As is evident from diaries, letters and memoirs of scientists, these people have dutifully or reluctantly notified them that their recent discovery that they were so elated about is a multiple. With the effect that those ideas never find their way into print.

W. R. Hamilton again, on a result in optics that he thought he was first to arrive at:

A fortnight ago I believed that no writer had ever treated of Optics on a similar plan. But within that period, my tutor, the Reverend Mr. Boyton, has shown me in the College Library a beautiful memoir of Malus on the subject. . . With respect to those results which are common to both, it is proper to state that I have arrived at them in my own researches before I was aware of his.

Third class of evidence is the behavior of scientists themselves. They act on an assumtion that singleton discoveries are imminent or potential multiples, that if one scientist does not soon make the discovery, another will.

The social institution of science puts a premium not only on novelty but also on chronological priority in discovery.

As Merton poignantly put it:

…the values and reward system of science, with their sometimes pathogenic emphasis upon originality, help account for certain deviant behaviors of scientists: secretiveness during the early stages of inquiry, lest they be forestalled; violent conflicts over priority; an unending flow of premature publications designed to establish grounds for later claims to having been first.

Along with lab fires and explosions, multiple discoveries are an occupational hazard of science. Scientists are affectively involved with their discoveries: nuggets of knowledge are revealed to them as a result of long, hard work, so it is no wonder that finding out that theirs is “just” another instance of a discovery already made by another is a matter of acute stress.

It may seem that the quote by W. R. Hamilton⁴ at the beginning of this post may testify to the contrary. But the effects of that charitable insight of his did not last long. Later in the very same letter he announced that he had anticipated the work on ellipsoids by the mathematician Ferdinand Joachimsthal in “a long extinct periodical of whose existence he [Joachimsthal] probably never heard, with a date which happened to be a precise decennium earlier.”

Autobiographies, letters and diaries of scientists like Wiener, Huygens, Newton, the Bernoullis and many others contain numerous examples of expressions of a similar sentiment.

Friends and family of scientists can be outspoken about matters of priority when the scientists themselves aren’t.

Elder Bolyai warned his son János about the necessity of publishing his [János’s] results in non-Euclidean geometry sooner than later:

first, because ideas pass easily from one to another, who can anticipate its publication, and secondly, there is some truth in this, that many things have an epoch, in which they are found at the same time in several places, just as the violets appear on every side in spring… Thus we ought to conquer when we are able, for the advantage is always to the first comer.

Friedrich Bessel, a faithful friend of Gauss, for years on end encouraged him to publish his discoveries to avoid being forestalled. Ironically, once Gauss published a manuscript on dioptrics and sent a copy to Bessel, it turned out that this work fully anticipated Bessel’s own unpublished results.

Adrien-Marie Legendre warned Karl Jacobi about the possibility of Niels Abel’s overtaking him in the race for coveted results in the theory of elliptic functions unless “you take possession of that which belongs to you by letting your book appear at the earliest possible date.”

Many such cases.

Lastly, there are institutional devices designed to protect scientists’ priority of discovery. It’s been an established practice, since the 17th century, for scientific academies and societies to have sealed and dated manuscripts deposited with them to protect the priority and the ideas themselves.

From Royal Society’s archives⁵:

When any fellow should have a philosophical notion or invention, not yet made out, and desire that the same sealed up in a box might be deposited with one of the secretaries, till it could be perfected, and so brought to light, this might be allowed for the better securing inventions to their authors.

The French Academy of Sciences also practiced this arrangement - among the many documents deposited under their seal was Antoine Lavoisier’s on combustion.

And of course, the patent system is a formalized way of securing priority of discoveries and inventions with a commercialization potential.

Merton’s multiple pieces of evidence all seem to testify that scientists, in practice, if not in theory, already assume that his hypothesis is true: all discoveries are potential multiples, and singletons are most often forestalled multiples.

I found these arguments pretty compelling, especially the last bit on the practical behavior of scientists - this I have witnessed first- (and second-)hand, having worked in several biology labs over the years.

In the next post, I will look into what MDs mean for the history of ideas and if and how the pervasive fact of multiples can be harnessed in economics of science. Thanks for reading!

If you have comments on this post, feel free to get in touch on twitter/X or email me at aghayeva.u@gmail.com.

Life of Priestley, Centenary Edition, p. 74 ↩︎
From his letter to William Rowan Hamilton, in Graves, Life of Hamilton, 3: 494 ↩︎
Robert K. Merton, Singletons and Multiples in Science (1961) ↩︎
By this point he has inadvertently become something of a main character of this post. I blame that on Merton who has quoted Hamilton extensively in his essays (though for good reasons). ↩︎
Thomas Birch, The History of the Royal Society of London (London: A. Millar, 1756), 2:30 ↩︎

Multiple Discoveries, Discovered Multiply

aghayeva.u@gmail.com (Ulkar) — Mon, 16 Oct 2023 00:00:00 +0000

If there be nothing new, but that which is
Hath been before, how are our brains beguil’d,
Which, labouring for invention, bear amiss
The second burthen of a former child!

William Shakespeare, Sonnet 59

This blog is setting sail as a stan account of Robert K. Merton. I first came across the name in a book titled “Sleeping Beauties: The Mystery of Dormant Innovations in Nature and Culture”. Robert K. Merton¹ is the founder of the fields of sociology of science and sociology of crime and deviance. Many of the sociological concepts now firmly embedded in everyday language owe their origin to Merton: ‘unintended consequences’, ‘role model’, ‘reference group’, ‘self-fulfilling prophecy’, ‘self-defeating prophecy’, ‘Matthew effect’ (with Harriet Zuckerman).

Merton authored a compendium of essays “The Sociology of Science: Theoretical and Empirical Investigations” ranging from establishing sociology of science as a distinct discipline to examining the reward system of science. The essays² that caught my eye at first were on the subject of multiple discoveries in science, since this was also the context in which I initially encountered Merton’s name.

Multiple discoveries, now also referred to as Merton’s multiples, happen when the same discovery or technological invention³ is made independently by multiple people - within a span of decades to years or even weeks. There can be debates as to what is meant by ’the same’ in this context but let’s put that aside for now.

I will divide the subject into 3 parts: this post will look into historical examples of multiple discoveries, in the next one I will go through the evidence of their ubiquity (so it’s not just isolated examples), and the third post in the series will consider what this says about “the shape of human ignorance” and what sociological and economic implications of multiple discoveries (MD from now on) are.

The most amusing thing about MDs is they are an example of themselves. “Almost,” says Merton, “it is a Shakespearean play within a play.” He notes that before him, at least 18 different historians of science and scientists themselves put together lists of varying lengths tallying MDs. The most famous study among them is by Ogburn & Thomas (1922) titled “Are inventions inevitable? A Note on Social Evolution”, documenting some 150 of MDs.

Without going too far back, in 1828 Lord Macaulay, a British historian and Whig politician, wrote about how the independent discovery/invention of calculus by Newton and Leibniz is an example of a larger class of instances of independent discoveries. He also notes that, for example

the doctrine of rent, now universally received by political economists, was propounded, at almost the same moment, by two writers unconnected with each other. Preceding speculators had long been blundering round about it; and it could not possibly have been missed much longer by the most heedless inquirer.

I’ll name a couple more curious examples from Merton’s list of people who made lists of MDs:

in 1862-1865 the London Times, in light of the controversies over the patent system, published an entire roster of MDs, noting that “that the progress of mechanical discovery is constantly marked by the simultaneous revelation to many minds of the same method of overcoming some practical difficulty” (13th September 1865).
in 1869 Francis Galton in his Hereditary Genius, considered “it notorious that the same discovery is frequently made simultaneously and quite independently, by different persons” (in part referring to the Darwin-Wallace doublet in evolutionary theory), and returned to this subject in 1874, in his English Men of Science.
in 1906-1913 Pierre Duhem, the physicist-chemist and one of the fathers of the modern history of science, documented MDs and their implications in all of his major works of that period.

Ogburn and Thomas (1922) mention, among their 150 examples of MDs, that the pendulum clock was invented at least three different times, the thermometer seven times and the telegraph four times. The most famous inventors of the telephone are Alexander Bell and Elisha Gray (1876) but even before them, there were Antonio Meucci (1854) and Philipp Reis (1861) who came up with their own versions of the device.

Sometimes the fact of a multiple comes to light when unpublished results of a scientist become known, often posthumously. Henry Cavendish, for example, independently arrived at what we now know as Ohm’s law relating electrical current and voltage, and also discovered that combusting hydrogen produces water. But this wasn’t revealed until after his death in 1810. Augustus Harcourt published some of his work in chemistry in 1839, James Clerk Maxwell - his work in electricity in 1879, and Sir Edward Thorpe - his complete works in chemistry as late as in 1921. In the meantime, many of Cavendish’s unpublished results were independently discovered by his contemporaries and later researchers.⁴

Likewise, many important results in mathematics in the 19th century turned out to be multiples when Carl Friedrich Gauss’s private notebooks were unearthed and thoroughly studied. These include non-Euclidean geometry, quaternions, and Gauss–Seidel method for solving systems of linear equations (described by Gauss in 1823 and independently published by Seidel in 1874), among many others.⁵

Some published works are as good as unpublished because of their impenetrable prose or publication in obscure outlets. It has been said of Gibbs, of Gibbs free energy fame, that “it is easier to rediscover Gibbs than to read him.”⁶ (I will go more into this phenomenon in another post.)

Is it surprising that MDs happen and in such abundance? It was to me, even being aware of some of the most colorful textbook examples. So it was also - or at least noteworthy if not entirely surprising - to many people who took pains to document them, though it seemed a matter of course to others. Merton writes, in his characteristically eloquent prose:

the theory [of multiple discoveries] was most unevenly diffused among scholars and scientists. By the middle of the nineteenth century, it had become, for some, a commonplace and often deplored truth; for others, it represented an entirely new conception of how science advances through the uneven accumulation of knowledge and through immanently or socially induced foci of attention to particular problems by many scientists at about the same time.

Further, Merton suggests that

the pattern of independent multiple discoveries in science is in principle the dominant pattern rather than a subsidiary one. It is the singletons - discoveries made only once in the history of science - that are the residual cases, requiring special explanation. Put even more sharply, the hypothesis states that all scientific discoveries are in principle multiples, including those that on the surface appear to be singletons.

I will go into the evidence supporting this (seemingly outlandish!) hypothesis in the next post. Thanks for reading!

Many thanks to Ashley Zhang and Santi Ruiz for their feedback on a draft of this post!

If you have comments on this post, feel free to get in touch on twitter/X or email me at aghayeva.u@gmail.com.

Merton is an adopted name - he was born Meyer Robert Schkolnick and took a stage name ‘Merlin’ for his magic performances as a youth. This then morphed into Robert K. Merton, “to further Americanize his immigrant family name”. K. stands for King (no false modesty here). Also his son, Robert C. Merton, is a Nobel Laureate in Economics (1997). ↩︎
Singletons and Multiples in Science (1961), Multiple Discoveries as Strategic Research Site (1963) and The Ambivalence of Scientists (1963) ↩︎
Here, discoveries and inventions are considered jointly, without making the distinction. ↩︎
Among them: Black, Priestley, John Robison, Charles, Dalton, Gay-Lussac, Faraday, Boscovich, Larmor, Pickering… ↩︎
And these were independently discovered by Abel, Jacobi, Laplace, Galois, Dedekind, Franz Neumann, Grassmann, Hamilton… ↩︎
This is a paraphrase by Muriel Rukeyser in Willard Gibbs of a remark by Wilhelm Oswald’s in a preface to Studies in Thermodynamics. ↩︎