S1-305 – How ideas travel
From individual insight to collective knowledge
In the winter of 1676, Isaac Newton wrote a letter to Robert Hooke that contained one of the most quoted sentences in the history of science: “If I have seen further, it is by standing on the shoulders of giants.” The sentence is usually read as a gesture of intellectual humility: the great scientist acknowledging his debts to his predecessors. It is also, read more carefully, a description of a specific mechanism: the mechanism by which knowledge accumulates across generations, by which the insight of one person in one time and place becomes available to another person in a different time and place, and by which the second person can begin not from nothing but from the point that the first reached. The sentence is about transmission. It is about the infrastructure that makes cumulative knowledge possible. And it is about something that is easy to take for granted precisely because it works: the fact that what one person has figured out does not die with them, but can be preserved, transmitted, and built upon by people who were not there when the figuring out occurred.
This infrastructure is not natural. It was constructed, over millennia, through a series of technologies (writing, the codex, the printing press, the scientific journal, the peer review system, the internet) each of which changed not only the speed and scale at which ideas could travel but the character of the ideas that traveled, the selection pressures that determined which ideas survived transmission and which did not, and the reliability of the knowledge that arrived at the other end. Understanding how ideas travel is understanding how human knowledge is actually produced, which means understanding both why the accumulated knowledge of civilization is as reliable as it is, and why it is not more reliable than it is.
Memory without writing, and its limits
Before writing, knowledge was transmitted entirely through the living memory of human beings: through speech, demonstration, and the gradual imitation that article S1-603 describes as the mechanism of cultural learning. This is not a primitive or inadequate mechanism. It is extraordinarily powerful within its domain. The oral traditions of cultures without writing developed sophisticated technologies of memory: rhythmic structure, formulaic repetition, narrative embedding, the ritual contexts that made certain kinds of knowledge memorable and recoverable across many generations. The Homeric poems, composed centuries before they were written down, were preserved with remarkable fidelity through the oral tradition. The accumulated practical knowledge of indigenous peoples about local ecologies, medicinal plants, seasonal patterns, and navigational routes was transmitted through oral tradition across generations with a reliability that modern researchers have found repeatedly impressive.
But oral transmission has structural limits that writing was invented, in part, to overcome. The first is capacity: the amount of information that can be reliably transmitted through living memory is bounded by what human beings can remember, which is small relative to the total body of knowledge that a complex civilization requires. The second is fidelity: as article S1-302 established in the context of the telephone game, each transmission introduces small errors that accumulate across successive transmissions, and oral traditions have no error-correction mechanism comparable to what writing provides. The third is range: oral knowledge can only travel as far and as fast as the people who carry it, which means it is local in a way that written knowledge is not. And the fourth is persistence: oral knowledge dies with its carriers. The Alexandrian library, when it burned, destroyed written knowledge. The equivalent destruction of oral knowledge (the death of the last person who knew a particular body of information) left no trace.¹
Writing changed all four of these limits simultaneously. It extended the capacity of human memory to whatever could be written down. It provided a fixed record against which the accuracy of transmission could in principle be checked. It allowed knowledge to travel further and faster than any human carrier. And it made knowledge persistent across the death of its original holders. These are not small advantages. They are the conditions that made the progressive accumulation of knowledge (the standing on the shoulders of giants that Newton described) possible at all.
The printing press and the democratization of error correction
Writing solved the persistence problem but created a new one: the reproduction problem. A manuscript had to be copied by hand, which was expensive, slow, and introduced its own errors. The number of copies of any given text was therefore small, their distribution was limited, and the errors introduced by each successive copying accumulated in the same way that oral transmission errors accumulated, without the oral tradition’s advantage of being continuously checked against living practice.
The printing press, developed by Gutenberg in the middle of the fifteenth century, changed the reproduction problem in a way that had consequences for the reliability of knowledge that went far beyond the obvious advantages of speed and cost. Before the press, the primary error-correction mechanism for written knowledge was the scholar who compared different manuscripts and tried to reconstruct the original from the variants. After the press, a much more powerful error-correction mechanism became available: the simultaneous distribution of identical copies to many readers, who could each independently check the text against their own knowledge and observation, and whose corrections could be incorporated into subsequent editions.²
This shift (from sequential to parallel error correction) is one of the most consequential developments in the history of knowledge production. It is the mechanism that made the scientific revolution possible, not because the printing press produced better ideas but because it produced a social environment in which ideas could be more quickly and more reliably tested. The natural philosopher who published a claim in print was exposing it to the scrutiny of every reader who had access to the printed text: a scrutiny that no manuscript tradition could match in scale, speed, or reliability. The ideas that survived this scrutiny were more likely to be correct than the ideas that had circulated in manuscript, not because the printing press selected for truth but because it selected against falsehood more efficiently.
The scientific journal and the formalization of peer review
The next transformation in the infrastructure of knowledge diffusion was the formalization of peer review through the scientific journal: a development that began in the seventeenth century with the founding of the Philosophical Transactions of the Royal Society (1665) and the Journal des Sçavans (1665), and that produced, over the following three and a half centuries, the system through which virtually all professional scientific knowledge is currently produced and validated.
The peer review system embeds, in the structure of knowledge publication, a specific set of epistemic norms: that claims should be supported by evidence, that methods should be described in sufficient detail to allow replication, that the work should be assessed by qualified experts before publication, and that the published record should be permanent and citable. These norms are not merely bureaucratic requirements. They are the operational expression of the falsifiability criterion that Popper identified as the mark of genuine scientific claims: a paper that describes its methods fully enough to allow replication is a paper that is making a claim specific enough to be wrong, and is providing the information needed to establish whether it is wrong.³
The peer review system is also, however, imperfect in ways that have become increasingly visible as the volume of scientific publication has expanded and as the social incentives of academic science have shifted. The replication crisis (the discovery, beginning in the 2000s, that a substantial proportion of published findings in psychology, medicine, economics, and other fields could not be reproduced by independent research teams) revealed that peer review, as currently practiced, does not reliably filter out false positives produced by underpowered studies, flexible analysis methods, and the publication bias toward positive results. Peer review is a necessary but not sufficient condition for reliable scientific knowledge. It is, in the terms of this series, a model with limits: a system that works well enough to have produced an extraordinary body of reliable knowledge, and that fails in specific and predictable ways that the system itself has been slow to diagnose and correct.⁴
The internet: acceleration without selection
The internet has done for the diffusion of ideas what the printing press did for the reproduction of manuscripts: it has eliminated the practical barriers to distribution that previously served, unintentionally but effectively, as quality filters. Before the internet, the cost and difficulty of reaching a large audience meant that only a small fraction of the ideas generated by the world’s population ever reached that audience. The selection pressure of limited distribution favored, imperfectly but genuinely, ideas that were interesting enough to attract the investment required to produce and distribute them.
The internet has largely eliminated this selection pressure. The cost of distributing an idea to a global audience has dropped to effectively zero, which means that the selection pressure that previously operated at the distribution stage has been removed. Ideas are no longer filtered by their ability to attract the investment required for distribution. They are filtered, if at all, by something else: by the engagement they generate, by the emotional responses they trigger, by their compatibility with the existing beliefs of the communities through which they spread. And as article S1-210 established in the context of the availability heuristic, the emotional and engagement-based selection of ideas is not a reliable guide to their truth. The ideas that spread most easily on the internet are not the most accurate ideas. They are the most arousing, the most affirming, the most outrage-generating, and the most compatible with prior belief. These are properties that are largely independent of truth and sometimes inversely correlated with it.
The specific consequence for collective knowledge is the fragmentation of the epistemic commons: the shared body of facts, sources, and standards of evidence that democratic deliberation requires in order to function. When the printing press democratized knowledge, it also created, over time, shared epistemic institutions: newspapers, universities, professional bodies, scientific journals. These institutions were imperfect: they had their own biases, their own blind spots, their own capture by powerful interests. But they provided a shared reference point: a set of facts and sources that people with different views could argue about from the same evidentiary starting point. The internet has not merely supplemented these institutions. It has, in significant parts of the population, replaced them with ecosystems of information that are organized not around shared evidentiary standards but around shared identities and shared emotional responses.⁵ The consequence is not merely that people have access to more information than before (they do) but that the information they have access to is increasingly selected for its compatibility with what they already believe, and decreasingly checked against shared standards of evidence that could in principle correct it.
Viral ideas vs. true ideas: the selection problem
There is a distinction that the preceding analysis makes necessary to state explicitly: the distinction between an idea that spreads widely and an idea that is true. These two properties have never been perfectly correlated (false ideas have always spread) but the history of knowledge transmission has been, in significant part, a history of building institutions that increased the correlation: institutions that made true ideas somewhat more likely to spread than false ones, or that at least filtered out the most egregiously false ones before they reached large audiences.
The scientific journal increased this correlation by requiring that ideas survive peer review before achieving the prestige of publication. The newspaper increased it (imperfectly) by requiring that claims be attributed to named sources who could be held accountable for their accuracy. The textbook increased it by requiring that ideas survive long enough to be considered reliable before being included in the educational canon. Each of these institutions was imperfect and subject to capture by interests that distorted its error-correction function. But each of them introduced some selection pressure in the direction of truth.
The internet, in its current form, has introduced powerful new selection pressures in a different direction. The algorithm that determines which content reaches which users is optimized for engagement: for the clicks, shares, and reactions that generate revenue for the platform. Engagement is not randomly distributed across the space of available ideas. It is systematically higher for ideas that trigger strong emotional responses: outrage, fear, wonder, moral indignation. These emotional triggers are themselves not randomly distributed across true and false ideas. The research literature consistently shows that false news spreads faster, further, and more broadly than true news: a finding documented across multiple platforms, languages, and political contexts.⁶ The mechanism is not mysterious: false news is disproportionately novel and emotionally arousing, because true news tends to be constrained by the boring facts of what actually happened, while false news is free to be maximally interesting.
The result is a knowledge ecosystem in which the correlation between spread and truth has decreased, relative to the pre-internet knowledge institutions, in the specific domain of public affairs: the domain where democratic deliberation most depends on accurate shared knowledge. This is not an argument for restricting the internet or for returning to a past that was not, in fact, epistemically ideal. It is an argument for understanding the specific ways in which the internet’s selection pressures differ from those of previous knowledge transmission systems, and for developing, at both individual and institutional levels, the epistemic practices that the internet’s architecture makes more necessary rather than less.
The Conscious Look, applied to knowledge transmission
The practical implication of this article’s argument is a specific form of the series’ central practice: the habit of asking, about any piece of information, not merely whether it is true but how it traveled to us and what selection pressures shaped its journey.
Information that has survived peer review, replication attempts, and extended scrutiny by qualified critics is more likely to be reliable than information that has not: not because peer review is infallible, but because each of these filters applies selection pressure in the direction of truth, and information that has survived all of them has passed more tests than information that has not. Information that reached us because it went viral on a social media platform has survived a very different selection process (one that selects for engagement rather than accuracy) and should be held with correspondingly lower confidence, regardless of how plausible it sounds or how many people have shared it.
The provenance question (where did this information come from, and what did it have to survive to reach us?) is one of the most useful single questions available for calibrating confidence in specific claims. It does not always yield a clear answer, and the answer it yields is not infallible. Peer-reviewed findings have turned out to be wrong. Viral claims have occasionally turned out to be true. But the selection pressures of different transmission systems are real, they operate systematically, and they are knowable. Asking the provenance question is the application of the series’ principle of knowing the limits of a model to the specific model constituted by a piece of information: understanding not just what it claims but what it has been through, and what that history says about how much confidence it warrants.
Newton’s giants were real because the shoulders they provided were built from knowledge that had survived long enough and widely enough to earn the trust of subsequent investigators. Not all knowledge that circulates in the current information environment has survived comparable tests. The practice of distinguishing between knowledge that has survived scrutiny and information that has merely spread is, in the current environment, one of the most practically important applications of The Conscious Look available.
Further reading
Elizabeth Eisenstein’s The Printing Press as an Agent of Change (1980), available in a condensed single-volume version, is the definitive historical treatment of what the printing press did to the production and distribution of knowledge in early modern Europe: including its role in the scientific revolution and the Protestant Reformation, and the ways in which the simultaneous distribution of identical copies to many readers changed the error-correction dynamics of knowledge production.
Vannevar Bush’s “As We May Think,” published in The Atlantic in July 1945 and freely available online, is the essay that anticipated many of the features of the internet 50 years before it existed: specifically the idea of a machine that would allow researchers to follow associative links between stored documents. It is historically important and reads as an extraordinary demonstration of intellectual foresight.
Walter Ong’s Orality and Literacy: The Technologizing of the Word (1982) provides the most thorough available treatment of the cognitive and social consequences of the transition from oral to written culture: what writing does to thought, memory, and social organization that oral culture cannot. It is the essential background for understanding why writing was not merely a faster version of speech but a genuinely different cognitive technology.
Soroush Vosoughi, Deb Roy, and Sinan Aral’s “The Spread of True and False News Online,” published in Science in 2018, is the most rigorous available study of the differential spread of true and false information on the internet: documenting the specific mechanisms by which false news spreads faster and further than true news, and the specific emotional properties of false news that make it more engaging. It is freely available and directly relevant.
Naomi Oreskes and Erik Conway’s Merchants of Doubt (2010), discussed in the context of article S1-207, is equally relevant here: it documents the specific ways in which the internet’s predecessor media (cable television, press releases, and the op-ed pages of major newspapers) were used to insert false or misleading information into the knowledge transmission system by mimicking the surface features of legitimate knowledge production without the substance.
Notes
¹ The loss of oral knowledge through the death of its carriers is not a historical abstraction. It is an ongoing process. The linguist David Harrison has estimated that the world loses, on average, one language every two weeks, and with each language, a body of knowledge encoded in that language that has no written form and no living speaker to transmit it. The specific bodies of knowledge most at risk are precisely those that oral traditions have been best at preserving: detailed knowledge of local ecologies, medicinal plant uses, navigational techniques, and the complex social and cosmological frameworks that organized life in particular places over long periods. Harrison, D. K. (2007). When Languages Die: The Extinction of the World’s Languages and the Erosion of Human Knowledge. Oxford University Press.
² The historian Adrian Johns has argued, in The Nature of the Book: Print and Knowledge in the Making (1998), that the standardization of knowledge achieved by the printing press was less immediate and less automatic than is often assumed: that the early centuries of print were characterized by considerable uncertainty about the reliability and authority of printed texts, and that the development of conventions that made printed knowledge trustworthy was itself a long historical process rather than an automatic consequence of the technology. The point does not undermine the argument made here, but it qualifies it: the printing press created the conditions for more reliable knowledge transmission; the realization of those conditions required centuries of institutional development.
³ The Philosophical Transactions of the Royal Society, founded in 1665 by Henry Oldenburg, is generally credited as the first scientific journal in the modern sense: the first publication that required contributors to describe their methods and that provided a permanent public record against which claims could be checked and cited. The peer review system, in its current form, developed considerably later: systematic pre-publication review by qualified referees became standard practice only in the twentieth century. The development of peer review is traced in Csiszar, A. (2016). The Scientific Journal: Authorship and the Politics of Knowledge in the Nineteenth Century. University of Chicago Press.
⁴ The replication crisis has been documented most thoroughly in psychology, where the Open Science Collaboration’s large-scale replication project (2015) found that only 36 to 39 percent of published findings could be successfully replicated. Open Science Collaboration. (2015). Estimating the reproducibility of psychological science. Science, 349(6251), aac4716. Similar findings have emerged in medicine, economics, and nutrition science. The structural causes of the crisis (publication bias toward positive results, underpowered studies, flexible analysis methods, and the incentive structure of academic publishing) are reviewed in Ioannidis, J. P. A. (2005). Why most published research findings are false. PLOS Medicine, 2(8), e124.
⁵ The concept of the epistemic commons (the shared body of facts and evidentiary standards that democratic deliberation requires) is developed most fully in Jonathan Rauch’s The Constitution of Knowledge (2021), which argues that the norms governing knowledge production are themselves a kind of institution, and that the erosion of those norms through the fragmentation of the information environment has specific and measurable consequences for the capacity of democratic societies to make collective decisions on the basis of shared reality.
⁶ Vosoughi, S., Roy, D., and Aral, S. (2018). The spread of true and false news online. Science, 359(6380), 1146-1151. The study analyzed approximately 126,000 news stories shared on Twitter between 2006 and 2017 and found that false stories were 70 percent more likely to be retweeted than true stories, that false stories reached their first 1,500 users approximately six times faster than true stories, and that the differential spread was driven primarily by human behavior rather than by automated bots. The emotional novelty of false stories (their tendency to trigger surprise, fear, and disgust more strongly than true stories) was identified as the primary mechanism driving the differential engagement.