Early knowledge
The origin story of (human) knowledge began hundreds of thousands of years ago with the transmission of ideas, art, and culture via oral communication, whether in the form of song, dance, prayer, or folklore (Kelly, 2024). Written knowledge started as people began to draw or carve ideographic symbols, whether onto cave walls, bones, or clay tablets (Von Petzinger, 2017). As humans evolved to further value the concept and practice of externalising information—and what it then enabled through communication—shared writing structures began to emerge (Suenaga, 2023). These took the form of phonemic symbols representing different sounds, or other logographic structures, such as cuneiform writing, or indeed Chinese characters in which each symbol represents both sound and meaning (Norman, 2023).
These collective developments qualitatively changed human knowledge practices. They first allowed for the recording of simple transactions, thus enabling trade. But further expansions, such as the insertion of vowels into the Phoenician script by the Ancient Greeks; (considered the first alphabet; Woodard, 1997), allowed for more nuanced written creations. This alphabet spread quickly, its success put down to its simplicity, for unlike other writing structures that included hundreds of symbols (thereby necessitating intensive—and exclusive—training procedures), the Phoenician alphabet contained just 24 characters (Hock & Joseph, 2009). It was, in fact, with the vowel-enriched Phoenician alphabet that Homer’s Odyssey was written. A Cumaean adaption of Ancient Greek eventually evolved to become the Latin alphabet known today, and from which, during the 16th century, the 26-symbol English alphabet then emerged.
Written knowledge to printed knowledge
In the century preceding the establishment of the English alphabet, written knowledge production shifted gears significantly with the invention of the printing press. Around 1440, in a small town in Germany, a local craftsman Johannes Gutenberg invented the first fast-moving durable type mechanical printing machine (Briggs & Burke, 2009; Füssel, 2020). The rapidity of its adoption was breath-taking (for the time). By the end of the 15th century the rhythmic thudding of printing presses could be heard in over 270 cities across Europe (Eisenstein, 2005). In fact, it has been estimated that from 1450 to 1500, the same quantity of books was printed as produced in the previous millennia by handwriting scribes (Dewar, 1998).
This machine allowed—for first time ever—the means to circulate knowledge at scale. Previous scribal culture had allowed only for a one-to-one distribution, meaning that knowledge transmission relied on a single person capturing information from another, and conveying it to someone else, either in the form of a manuscript or an oral communication (Dewar, 1998; Eisenstein, 2005). This resulted in a precarious preservation of knowledge, reliant on mnemonics, elitist structures, and the note-taking skills of ‘wandering scribes’ (Eisenstein, 2005). The printing press now provided the means for one voice to speak to many (Wallace & Green, 1996), and in so doing, this simple device went on to play an unarguably influential role in the significant societal changes that took place over the rest of the millennium.
By today’s technological standards, the printing press is a simple, mechanical device through which metal shapes are used to uniformly press ink onto paper. And yet, the industrial-level production of information afforded by this machine democratised society through its role in allowing knowledge to passed to unprecedentedly large numbers of people (Zaret, 2000). As a result, the printing press is now considered pivotal in the disruption of Catholic power during the Middle Ages, the cultural Renaissance that took place as Europe then transitioned towards modernity, and relatedly, the ensuing scientific revolution (Cochrane & Chartier, 2014; Eisenstein, 2005). The shift away from elitist information systems began almost immediately following its invention. In 1476 William Caxton set up a printing press in London to capitalise on the growing demand for books written in the English language, as opposed to the previously dominant Latin texts (Blake, 1991; Morgan Library, 2015). As popular works such as Chaucer’s Canterbury Tales spread, the rapid increase in literacy rates contributed to the restructuring of social hierarchies, and with it, the emergence of the ‘middle class’ (Tarkhnishvili & Tarkhnishvili, 2013). It has been argued that much as the printing press revolutionised society over hundreds of years by allowing for the sharing of human knowledge, the digital or ‘information revolution’—which began less than 100 years ago—has equal (if not greater) revolutionary potential (Builder, 1993; Dewar, 1998).
Printed knowledge to digital knowledge
Over the last century, the amplification of knowledge has accelerated exponentially with the arrival of computers, digitalisation, and networking. This next step change in information production, particularly the interconnectivity of computers and users, has evolved the distribution of information from an already revolutionary one-to-many, to a new paradigm-shifting reality of many-to-many. Described as the defining characteristic of the information age, networked computers have radically altered the process, experience, and scope of information sharing (Dewar, 1998), and in so doing, changed not only the nature of knowledge acquisition, but also the production of knowledge itself.
The fundamental and emblematic manifestation of computer networking came in 1960s in the form of the internet. A global system of interconnected computers, the internet emerged thanks to developments in data communication and packet switching (simply speaking, the grouping of data into short, structured forms able to be transmitted across digital networksFootnote 2). The communication protocols enabling functional internetworking came out of R&D collaborations between the US Defence Agency (DARPA) and universities in the US, the UK and France. The transition into the World Wide Web—involving commercial networks and enterprises alongside accelerated public engagement—took place later in the 1990s.
On April 30th 1993, the internet was released into the public domain (Dinmore, 2023; Ring, 2023). All a user had to do was launch a new programme called a ‘browser’ and then type in an electronic address, known as a URL (Uniform Resource Locator). More importantly, anyone (leaving aside important issues of digital exclusion; Shaban, 2025) could publish information themselves online. Whether text, photos, music, or videos, this new form of information dispersal could also include links to other people’s information, literally accessible at the click of a button. By the end of 1995 more than 24 million people in the US and Canada alone spent on average 5 h per week perusing these new information sources. Today, 64% of the world’s population regularly uses the internet (Statista, 2025).
In its early days, the internet was an unregulated, organic, proliferating mass of mostly crowd-sourced information. Within this new and emerging communication culture, an abundance of discussion groups was created. Known as USENETS, these have been described as “an enormous billowing crowd of gossipy, news-hungry people wandering in and through the Internet on their way to various private backyard barbecues” (p. 2, Sterling, 1993). The potential for almost instantaneous mass-scale information sharing was, therefore, enormous. Users could (and would) navigate through thousands of newsgroups, reading hundreds of thousands of articles on a variety of topics, and sharing their own thoughts and reactions.
This new information network changed almost every aspect of society. From hobbyists to political activists, the ability to connect with others across the globe on the basis of shared information and interest was, quite literally, revolutionary (Tudoroiu, 2014). Knowledge was now being crowd-sourced, impacting anything from travel choices to political mobilisations. But equally, the internet began to reshape more formal knowledge processes. Researchers no longer required long visits to libraries but instead could access previously unimaginable quantities of information through browsing online. Scientific collaborations subsequently increased, alongside a substantial acceleration in the dissemination of findings (Gholizadeh et al., 2014; Teasley & Wolinsky, 2001).
This information expansion changed not only the creation of new knowledge, but the validation of existing knowledge. Prior to the printing press, errors written into manuscripts (whether due to a tired scribe or more conscious informational distortions) remained in permanence (Dewar, 1998). The printing of books changed this. Creating a culture in which authorship was associated with ownership (Eisenstein, 2005), the printing revolution meant that facts could be debated, contested, and—when necessary—updated (albeit slowly). Today, the information age allows for facts, ideas, and perspectives to be both created and overturned in real-time (Chan et al., 2018; Flemming et al., 2017). Reflecting Thomas Jefferson’s far-sighted description of the nature of ideas, the internet—at least in its early days—held the promise of infinite knowledge expansion: “That ideas should freely spread from one to another over the globe, for the moral and mutual instruction of man, and improvement of his condition, seems to have been peculiarly and benevolently designed by nature, when she made them, like fire, expansible over all space, without lessening their density at any point, and like the air in which we breathe, move, and have our physical being, incapable of confinement or exclusive appropriation” (Jefferson, 1813). Today, as we peruse our AI-generated online search results, the impact of this form of information mediation on the future of human knowledge is unclear.
Digital knowledge to AI-generated knowledge
Today, generative AI presents us with a powerful new way of accessing information and interacting with knowledge. The term ‘generative AI’ describes a particular use of Artificial Intelligence, largely based on machine learning and Natural Language Processing (Zewe, 2023). The product (simply speaking) is a dynamic prediction algorithm, able to respond to questions or prompts with an uncanny human-like demeanour, whether the output is text or voice, visual or musical (Bandi et al., 2023). Generative AI does this by drawing from the vast range of information upon which it is trained, allowing it to compute the statistical probability that informational components (such as words or pixels) cluster together (Salakhutdinov, 2015). For statistical accuracy, this technology must, therefore, train on vast amounts of data, hundreds of billions of words in the case of human language tools (Hughes, 2023).
Therein lies the power of generative AI. Using the fastest processing power (not to mention large amounts of energy; Chien et al., 2023; Luccioni et al., 2024), this technology can—within the parameters of human-coded algorithms—train autonomously and, through recursive self-improvement, adaptively (Madaan et al., 2024). The fuel which drives its training is (at least at this point in timeFootnote 3) online sourced human-originated data. This includes books, websites, reports, poems, songs, and articles… ultimately anything digitised and available on the web, including text exuded from community platforms like Reddit, X, or Facebook. Despite the enormity of this data resource, however, generative AI models remain hungry, and current practice—although ethically debatable (Zhang & Yang, 2024)—is to use any and all information input by current human users as an ongoing source of ‘live’ training material (Fui-Hoon Nah et al., 2023; Lucchi, 2024; Porter, 2023).
This debatable practice encapsulates many of the ethical controversies surrounding generative AI. These include not only issues of Intellectual Property (IP) ownership but also concerns over the quality of the information upon which AI models are trained. Known levels of bias are present in the data, both selection bias (not all segments of society get to put their information online; Giorgi et al., 2022), but there is also substantial evidence of gendered, racial, and ideologically bias in generative AI’s outputs, presumably due to these forms of social distortion being present in the original training data (Wach et al., 2023). This uncomfortable reality has required generative AI systems to be further re-trained, adjusted, and controlled—by humans (not algorithms)—to provide an end product that is, loosely speaking, acceptable for human re-consumption (Kirova et al., 2023; Rani & Dhir, 2024).
Despite these controversies, the uptake of generative AI technologies across society has been unprecedented. Marketed as everyone’s new best friend, whether co-pilot, research assistant, or life coach, these new knowledge tools are currently free, available 24/7, answer pretty much any question you might care (or dare) to pose, and are (mostly) polite (Quan & Chen, 2024). Further, they are (largely) not domain-specific but can comprehensively answer questions on any subject, from eco-gastronomy to puppetry. Recent surveys are reporting that over one-quarter of American adults say they are using applications like ChatGPT (Motyl et al., 2024; Pew Research Center, 2024), with this figure increasing for younger users as well as people with higher levels of educational attainment (Park & Gelles-Watnick, 2023). 20% say they use it for entertainment, 19% for learning, and 16% to assist them with tasks such as summarising information. In the workplace, 75% of over 31,000 ‘knowledge workers’ (from 31 different countries), reported using generative AI to help them with their jobs (Microsoft & LinkedIn, 2024). In the health domain, a survey of 1,000 British general medical practitioners, reported that 20% were using tools such as ChatGPT for notetaking and diagnostics (Blease et al., 2024). Uptake is also prolific in student populations. A report published by the Digital Education Council showed that 86% of tertiary students use AI tools to assist them with their learning (Digital Education Council, 2024), and in Australia, a Youth Insight analysis showed that 70% of 14 to 17-year-olds were using generative AI for information sourcing, assignment writing, and general learning (Denejkina, 2023).
It is clear, therefore, that not only is this technology quite literally awe-inspiring, but so too is our uptake of it. Today, its revolutionary potential lies not just in the extraordinary reality of a machine communicating like a human (no small feat), nor the fact that this technology can successfully pass various tests of both intellectual and emotional intelligence (Kosinski, 2024; Sumbal et al., 2024), but more subtly yet potentially more impactfully, the fact that by blithely sidestepping established information-sharing practices, structures, and even laws, this tool has the potential to radically disruptFootnote 4 not just the information we hold dear, but the very landscape in which we, as a global community, currently create, debate, and produce our most valuable commodity, that being knowledge.
AI-generated knowledge to disruptive knowledge
Despite the unprecedented uptake of generative AI knowledge tools, as of 2024, there is little peer-reviewed, validated, or replicated research informing us as to the effect that outsourcing our thinking to AI knowledge tools and systems may have on our (human) capacity to reflect, understand, and learn (not to mention feel, experience, or empathise; Smith et al., 2024; Yan et al., 2024). The closest findings we have relate to what we know about how human memory processes changed as a result of internet use. Referred to as ‘the Google effect’ (or digital amnesia), research has demonstrated that when faced with difficult questions, today people’s minds jump first to a digital device for solutions (Sparrow et al., 2011), and that further, memory is reduced—both quantitively and qualitatively—the more time a person spends on their digital devices (Çinar et al., 2020; Robert et al., 2024). Alarming as this sounds, this is the nature of disruptive innovation.
AI is a disruptive technology, and generative AI in particular—a product promising to curate every human’s knowledge needs—takes disruption to potentially unprecedented levels. In the light of the socio-technical dilemmas presented by the processes of disruption, there are those who advocate for the application of a ‘precautionary principle’ (rather like the old adage, better safe than sorry). However, there are also those who argue that such a principle suppresses innovation (Cross, 1996; Hellström, 2003; Jonas, 2009). Regardless of one’s position on this debate, the question of what generative AI might be doing to our most valued commodity presents us with an interesting (and possibly long overdue) opportunity to reflect—at this timely moment in history—on the nature of human knowledge, and the need (or not) to apply caution.
A case study: Miscalibrated algorithms
Think back to one of 2024’s most interesting generative AI misadventures. The diversity algorithms in a prominent image generator over-zealously re-calibrated the ethnicity ratio of its historical depictions. When asked to visualise a British medieval king, users were presented with images of a fine-looking African man dressed in fur robes and sporting a crown (Field, 2024). Further historical faux pas came in the form of female popes, black Vikings, and ethnically diverse American ‘founding fathers’.
What generative AI inadvertently did was to expose the highly ‘constructed’ nature of knowledge, both its origin story (who, where and why) but also the subsequent representation of that knowledge as it is composited onto our digital screens. This particular form of AI, along with its limitations, biases, and hallucinations (Sun et al., 2024), serves us with a timely reminder of the fallible nature of knowledge (Emsley, 2023). For social scientists, this may not be news. But for the world at large—the voting electorate, the summoned juror, or even the doomscroller—this is big news. And when it comes to the need to understand and tackle the reality of mass-scale mis- or disinformation (which significantly predates the arrival of generative AI), this is much-needed news (Posetti, 2018).
Despite the concerning implications of a privately funded, opaque, and certainly not peer-reviewed technology quite literally re-writing history, this socio-technical ‘incident’ presents us with an unprecedented and ‘live’ insight into the nature of knowledge fabrication. More specifically, knowledge artefacts are—thanks to generative AI—being revealed as potentially subjective, idiosyncratic, and financially tainted. This is not something for us to shy away from. For anyone remotely interested in knowledge practices, this technological misadventure provides the raw material for one of the most exciting, engaging, and illuminating exposés in recent times, and seems like a fitting finale for the already dramatic social trajectory of human knowledge production.
In fact, in an ironic twist on this futuristic technology, the arrival of generative AI in our homes, classrooms, and offices transports us right back to some of the most fundamental of debates waged 3000 years ago in ancient Greece. The Sophists were ancient Greek ‘knowledge-makers’ whose job it was to deliver expertise, for a fee (Säfström, 2023). Little evidence of their work exists today, and there remains debate as to their ultimate purpose, with some considering them experimental thinkers using speculative reason (Crick, 2010), and others portraying them as intellectually dishonest charlatansFootnote 5 (Duke, 2025). Like generative AI today, the Sophists claimed to democratise knowledge, but it was their commodification of information that ultimately became their downfall, revealing as it did, the malleable, serviceable, and fallacious potential of knowledge.
Disruptive knowledge to responsible knowledge
Today, with the dramatic entrance of generative AI into our midst, we find ourselves back at the intersection of knowledge, society, and economics. In order to ensure the beneficent impact of this novel and disruptive technology, we would do well to use it to spotlight important epistemological debates—where knowledge comes from, how it impacts thoughts, attitudes, and behaviours, and how we can trust it. Given the advanced sophistry involved in each and every individually AI-enhanced digitised production of our daily dose of news, information, or educative material, these questions are more important than ever (Datta et al., 2021; Rodilosso, 2024). Answering them will prove fundamental to our ability to use generative AI responsibly, equitably, and for the betterment of future generations.
So, how to do that? Strangely, turning back once more to Ancient Greece can provide us with some ideas. Socrates was the only Sophist who refused to charge for his services (Cooper, 1998; Taylor, 2006). The original advocate of open-sourced information, this Ancient Greek philosopher developed the ‘Socratic method,’ a technique designed to scrutinise commonly held beliefs and assumptions through rigorous questioning to determine their validity, veracity, and applicability (Benson, 2011). Applying Socratic scrutiny to AI-generated knowledge would most likely conclude with an unsettling awareness of what is either distorted, politely overlooked, or categorically silenced (Worrell & Johns, 2024). At the very least, it should point out that the power of yesterday’s printing press to expand one voice to many, or indeed the power of digitisation to provide a ‘many to many’ information-sharing platform, is today being replaced by the power of generative AI to reduce ‘many voices to one’. Given the urgent need for society to become its own discerning audience (think pandemics, wars, climate crises, and ever-widening global inequities) this in itself would represent a profoundly useful piece of contemporary learning. Socrates is most famous for saying “The only true wisdom is in knowing you know nothing” (Cooper, 1998). If, as it seems, we may have ‘unknowingly’ moved into a period in which we are fast becoming information rich but thinking poor, acknowledging that we both know everything and nothing may be the one piece of knowledge we need right now.
