Introduction
When an idea appears again and again, in different forms, then you can be confident that there is something meaningful behind it. This was the case at NAMS 2026 regarding ideas about the restructuring of journalism due to the inexorably improving capabilities of AI.
In his closing keynote, and in the essay included in this anthology, Rasmus Kleis Nielsen described a 4-part framework for understanding journalism’s reaction to AI, with the two primary responses termed ‘Romantic Journalism’ and ‘Modernist Journalism’. The former, articulated in a recent manifesto by Lea Korsgaard and Joshi Herrmann, is centered on the author’s “bond of affinity with the people who are reading or listening or watching” and emphatically decenters AI from journalism. The latter, articulated by Nielsen as a form of journalism in which “Everything that can be automated, will be automated”, anticipates an information ecosystem that is intensely mediated by AI.
At the same event, Shuwei Fang used different terms in her comprehensive talk on the economics of AI-mediated information. Fang (who I work with) described the ‘Barbell Effect’ – a generalized and well-studied phenomena in which the appearance of a scaling technology causes an industry to separate into two extremes: a ‘differentiated’ extreme, based on intangible value; and a ‘scale’ extreme, based on applying the new technology to its full extent. Businesses in the middle range of the industry decline, in the manner of department stores in an era of boutique brands and e-commerce.
This same essential division shows up repeatedly, in different places, and with different terms. It is, in some ways, built into the organizational structure of every news publisher using the terms ‘editorial’ and ‘technical’. This essay is about that division, about how AI is changing the relationship between these two interpretations of journalism, and about why the journalism industry and journalism’s research institutions are extremely ill-equipped to handle that change. It is also about the potential to both clarify and expand journalism by accepting a scientific alternative, and the potential of science to supercharge journalism in the same way it has supercharged so many other fields over centuries. I will try to make my arguments as fundamental as I can, and so I will use a different pair of terms that I believe are closer to the heart of what is going on. Those terms are ‘humanistic’ and ‘scientific’.
The two cultures
In his famous 1959 lecture The Two Cultures, C.P. Snow argued that modern intellectual life had become divided into two increasingly disconnected cultures: the humanistic culture, represented by writers, artists and literary intellectuals, and the scientific culture, represented by scientists, engineers and technologists. Snow argued that these cultures differ not merely in their subject matter, but in their fundamental ways of understanding the world. The humanistic culture emphasizes subjective experience, interpretation, judgment and the ‘qualia’ of human experience. The scientific culture emphasizes empirical observation, experimentation, quantification and the accumulation of objectively verifiable knowledge. Snow saw these two cultures as equally valid and necessary and bemoaned the division between them as a barrier to human progress.
In his 2004 book Hackers & Painters, Y-Combinator co-founder Paul Graham makes a similar argument about the intellectual differences between the humanities and the sciences, but was less neutral in his interpretation. Graham suggests that while many physicists could probably complete a PhD in French literature, relatively few professors of French literature could complete a PhD in physics. His argument is not that scientists are more intelligent, but that scientific disciplines are more directly constrained by objective tests of truth and falsity, whereas humanistic disciplines allow greater scope for social influence, intellectual fashion and established authority. Scientific disciplines, essentially, have more discipline.
This structural difference between humanistic and scientific cultures can be interpreted as more than just an asymmetry. In Homo Deus, his 2015 book anticipating the AI era, Yuval Noah Harari argues that human civilization has changed its fundamental assessment of the source of ultimate authority in the past, and is currently changing it again. Harari identifies three distinct eras – Theism, Humanism and Dataism. He posits that the printing press caused a change in civilization’s source of ultimate authority from God (Theism) to human opinion (Humanism), and that computation, the internet and AI are now causing a further change from human opinion to, essentially, data (Dataism). Before the printing press it was difficult to argue with God, and during the humanist era it was difficult to argue with personal opinions and collective human decisions as expressed in elections and markets. Harari's point is that this authority of human opinion, which humanism treats as foundational, may instead be historically contingent and in decline. If it is data that now settles arguments, then perhaps the era of the dominance of human opinion is behind us.
It is as difficult to understate the significance of these changes as it is to ignore their reality in our direct experience. At some level these ideas are, to use a common idiom, “playing for all the marbles”, and for many journalists it will be easy to reflexively identify with the humanistic perspective and even to retreat into an ‘us and them’ defensiveness. But this need not be the only response. The changing dynamics between humanistic and scientific cultures in the AI era does not mean that humanistic journalism is ending – on the contrary it may become more valuable than ever, as the ‘differentiated’ side of the Barbell Effect or the argument for Romantic Journalism make clear. Instead, it can mean that it is time to expand journalism in ways that allow for the meaningful inclusion of scientific culture. If the objective of journalism in the AI era is truly to serve society with useful, accurate information that empowers citizens, then perhaps it is time to build a ‘scientific journalism’ around AI – a parallel journalism entirely within the scientific culture.
AI’s tectonic impact on humanistic culture
AI has deeply unsettled the balance between humanistic and scientific cultures. This is, obviously, because AI significantly encroaches into what was previously the sole preserve of humanism. AI can research, create new ideas and write, and therefore AI is a researcher, a creator and an author. AI has opinions and judgement. It can interpret and contextualize. It can make moral assessments. The usefulness of AI’s output can be tested and measured. That output can be compared with the output of humans and is found, in many cases and by many measures, to be superior to that of humans. All of this is true now, and the reach of AI into humanistic domains is likely to continue as AI models improve, as more agentic systems are built and as we learn how to use them.
For journalism, this disruption has been rapid and practical. AI authorship of varying kinds, whether overt or furtive, is now thoroughly embedded in most newsrooms. AI Overviews and similar experiences are providing more and more of the answers that consumers previously sought from journalists. Single individuals now operate news publications that would previously have required a newsroom. Consumers are poised to gain access to personal agentic news editors, and AI platforms are poised to bypass news articles as sources and gather relevant information directly.
More generally, journalism has experienced a ‘vibe shift’. The long, slow managed decline of journalism caused by the internet and social media has become a new, urgent realisation that we may be approaching some kind of endgame. The entire industry seems to now operate on a year-to-year basis and is pervaded by anticipation of sudden irrelevance. Journalism, as an industry, is waiting for an AI hammer to fall.
This situation is deeply uncomfortable and is made more so by the lack of any clearly defined path forward. For journalists this means anxiety, career uncertainty, questions about identity and even a kind of grief. For those who study journalism it means dilemmas about what is meaningful scholarship and what will be trivial or irrelevant work. For investors in journalism it means an almost incapacitating inability to allocate capital effectively beyond a few years.
There is no hiding from this situation. It is real, it is happening and the pre-AI status quo is gone forever – regardless of any plausible regulatory or legal development. Even if the observed impact is relatively light now, the ability to automate large portions of humanistic work is an enormous Sword of Damocles hanging over humanistic journalism’s head. The possibility of enormous impact by AI on large parts of humanistic culture must be accepted, but there is a coherent response despite the uncertainty: Journalism must develop a new conception of itself that firmly includes the scientific culture, operating alongside humanistic journalism. To move forward with confidence, despite the uncertainty, we need to clearly delineate between Romantic and Modernist journalism and between the differentiated and scale ends of the barbell.
Humanistic Journalism
Most journalism in our current information ecosystem is humanistic journalism. It is defined by the centrality of the opinions of specific humans called journalists, editors and publishers. It is an artisanal craft in an economy that has been industrializing for two hundred years – a handmade product that is still shaped almost entirely by the intuitive choices of individuals. Its practitioners define themselves through socialisation, identity and values, and not through the acquisition of an accumulated body of knowledge that defines the field. As knowledge work, humanistic journalism has previously been impervious to automation.
Opinion is everywhere in humanistic journalism, from decisions about commissioning and newsworthiness, to editing and word choices. The best humanistic journalists have a ‘voice’ – an undefinable but unmissable point-of-view or attitude that in many cases approaches literary art and is sometimes of great value. The very best are adept at storytelling – shaping narratives in ways so powerful that they produce rich and vivid experiences in the minds of their consumers. More commonly and mundanely, humanistic journalism often follows ‘party line’ editorial policies set and enforced by Editors-in-Chief, Publishers and shareholders. As a product largely defined by human opinion, and valued partly for the opinionated judgement of its producers, much of humanistic journalism may not benefit from automation.
As a humanistic endeavour, humanistic journalism is intensely social. It is a person ‘speaking’ personally to other people, and it carries all of the social considerations and signals present in such a relationship – trust, reputation, status, norms and conventions. Within humanistic journalism success is often defined through relationships – opportunities and status are decided through the favour of an editor or the receipt of an award from peers. In its relationship with audiences, humanistic journalism is often adept at triggering ‘limbic brain’ responses, like emotion, empathy or outrage, or even, for particularly skilled journalists, curiosity. In its relationship with its own elites, humanistic journalism is often conceived of as a tool for the exercise of influence within human societies. If we define humanistic journalism as a social activity, then automating it may be pointless, or even meaningless.
Humanistic journalism is ‘romantic journalism’. It is differentiated by the myriad indefinable things that differentiate an individual or a community, and it is positioned in opposition to scale on the barbell. It is literally defined by its human origins, and it is possible that those human origins may become far more valuable in an information ecosystem mediated by AI – not because of vaguely defined ‘quality’, but precisely because it is human. We have seen this before in the popularity of chess following Deep Blue’s defeat of Garry Kasparov, and of Go following AlphaGo’s defeat of Lee Sedol. Human beings are deeply social and often assign value based on social criteria, regardless of ‘quality’. Humanistic journalism has value now, and it will probably have value a hundred years from now – without automation.
Scientific journalism
Scientific journalism exists today only on the margins of journalistic culture. Aspects of it are present in data journalism, in the more numerical parts of sports or financial reporting, in the details of elections results, in investigative projects centred on databases and document repositories, and in fringe innovation projects. It is not a prestigious activity in most news organizations, and some journalists barely consider it to be journalism at all.
Scientific journalism is defined by abstraction, and thus by its independence from human opinion. It is a form of journalism that values precision and objectivity. It seeks to be systematic rather than selective, and to build systems rather than to handcraft individual artifacts one by one. It seeks to be clinically neutral and rational about the facts that it is reporting. Its judgements are technical and methodological – explicitly discussed and written down rather than opaquely socialized. The units of information that it gathers and processes are all treated the same, and it is assessed by its completeness and its accuracy. Its measure of success is its fidelity to an objective and measured reality – and its value to its consumers is judged primarily by the volume and intensity of their engagement with it. The introduction of the opinion of the practitioner of scientific journalism into its output is considered an error, not an attribute. Scientific journalism is not seeking to be socially influential – it is seeking to be useful.
Scientific journalism is eminently automatable. Automation of this kind of journalism began well before the arrival of general-purpose LLMs, especially using template-based natural language generation (NLG) systems. Although early scientific journalism was possible only on highly structured or numerical forms of information, this was a purely technical constraint rather than a fundamental requirement. Like humanistic journalism, scientific journalism is essentially a philosophical perspective about what journalism is and can be. Just as humanistic journalism is an expression of the larger culture of humanism, so scientific journalism is an expression of the larger culture of science, technology and engineering. Scientific journalism is very different from humanistic journalism but it is also, in the eyes of its producers and consumers, real journalism.
The application of a scientific perspective to journalism also applies to journalism’s understanding of its audience and their needs and wants. For much of its recent history journalism did not need to care what its customers thought of its product, because near-monopolistic control of pre-digital distribution channels offered few alternatives. Even now, journalism’s understanding of its audience is remarkably thin – at worst paternalistic folk theories, and at best simplistic analytics and coarse surveys based on declared preferences. Similarly the academic study of journalism, as a soft social science, is overwhelmingly oriented towards its producers rather than its consumers, and towards its existing form rather than its fundamental function. In 100 years of existence, journalism studies has failed to produce a consensus theory of journalism that could be used to orient the field. Science offers a far richer understanding of why and how human beings value information – cognitive research techniques, functional MRI studies, controlled multivariate testing and theory-building based on neuroscience and evolutionary biology rather than on identity and social conditions. Beyond audiences, science offers many avenues of productive research into the flows of information through societies, including the application of network theory, Agent-Based Modeling, information topology and similar techniques.
As the arrival of powerful AI deeply unsettles the balance between humanistic and scientific cultures, it also dramatically expands the potential scope of scientific journalism, including into what was previously the sole preserve of humanistic journalism. The technical constraints of scientific journalism have been removed, and so a new opportunity to apply a scientific approach to much more of journalism has appeared. This empowerment of a scientific approach to journalism is a positive thing, even for humanistic journalists. Journalism as it stands is clearly stuck, clearly out of ideas, and clearly faced with what appear to be insurmountable challenges ahead. By explicitly and operationally separating journalism into the romantic and the modernist, the differentiated and the scaled, the humanist and the scientific, we can perhaps bring clarity, precision and conviction about the place of each in our AI future.
Society urgently needs scientific journalism
Science works. It works so well that our entire modern civilization exists because of it. The scientific revolution of the 17th century unlocked something immensely powerful in human affairs. The basis of field after field transitioned from socialised folk theories into rigorous evidence, and the results were near miraculous – from medicine to metallurgy. Science built the intellectual platform which made the industrial and agricultural revolutions possible, bringing a way of life that now supports more than 8 billion humans at a level of comfort and security that would have been unimaginable a few centuries ago. We need to seriously consider what that power might do for the empowerment of individuals and collective interests if it were brought to bear on the flows of information that course through our societies.
A technical interpretation of societal information flows is inevitable. The primary reason is that humanistic interpretations cannot plausibly adjust to the scale, scope and velocity of AI-mediated information. The seeds of this inevitability have been sprouting all around us, in start-ups, platforms and academic fields that have essentially zero overlap with journalism as a field. Probably the most significant example so far is X and X.ai, and Elon Musk’s explicit ambition to replace journalism – all of it – with an entirely new model of societal information flows, and to spend billions of dollars doing it. This has not worked well (so far) and has not meaningfully changed our information ecosystem, but it is perhaps a harbinger of a new kind of fundamental innovation rooted entirely in the scientific culture. The problem with this inevitable technical alternative to humanistic journalism is not its existence, or its value, or its effectiveness or even its ‘quality’ – all of that will be decided by consumers. The problem is its values. Whatever the reality of our legacy journalistic institutions, their declared values of truthfulness and independence, and their watchdog and ‘checks and balances’ incentives have served society very well. The potential dangers of misused scientific power applied to societal information flows is very real, and even urgent. We will need, therefore, a scientific journalism that does not merely operate in a scientific way, but which also pursues objectives and values that are similar to the objectives and values of science itself. As journalism becomes more technical it would probably be best if the values of humanistic journalism – essentially enlightenment values – became the values of an emerging scientific journalism too.
Science is a moral force, as well as a pragmatic one. It is, in fact, the most credible and legitimate source of authority in our societies today. Many people still trust the words of religious leaders, intellectuals, politicians and even of journalists, but not as many and not as much as those who trust science. The best of technology, too, is trusted daily, by billions of people in trillions of practical acts seeking to fulfil their most important needs. Science and technology are trusted because they work – they deliver value – but science is also trusted for the same reasons as journalism wants to be trusted. Like journalism, science seeks to provide truthfulness, objectivity and consistency, but unlike journalism, science is not perceived to be filtered through the gate of human opinion. This neutrality comes from science’s reliance on specificity, evidence, data, logic, peer-review, citations and specific, directed criticism. It comes from systematic rather than selective treatment of its subject matter. Of course scientists have opinions too, but science is a system, a perspective and a culture designed to diminish rather than to accentuate the influence of those opinions.
Humanistic journalists sometimes consider scientific approaches to journalism to be somewhat naïve or ‘unsavvy’, claiming that reality and ‘truth’ are just too complex and nuanced to be understood, represented and communicated by technology – even technology as capable as advanced AI. This objection has not aged well, but it does have a point. Humanistic journalism excels at reporting what epistemologists call ‘inter-subjective’ truth or ‘social truth’ – the truth that is ‘true’ because it is agreed upon by a community of humans. There is clearly a significant demand for inter-subjective truth by information consumers, and this may well be a form of coverage in which humanistic journalists can specialise. But it is objective truth – scientific truth – that dominates human affairs. Scientific journalism, with its more systematic coverage and more neutral reporting, may perhaps offer society a way to coalesce around a shared perception of an objective reality in a way that journalism centered on human opinion cannot.
AI now gives us a new possibility for a genuine scientific journalism – for journalism engineering and for technical journalism – and thus for a new deal for journalism as a whole. Humanistic journalism has delivered immense value for societies over the centuries, and it will most likely deliver value for centuries to come. It can’t, however, become an end in itself – a sacred cow that must be privileged over the greater goal of open, democratic societies of empowered citizens with access to truthful, trustworthy information. Journalism can’t be limited by the desire to preserve familiar jobs and institutions or even cultures. It must expand or it will inevitably shrink.
Accepting and facilitating scientific journalism
Journalism must make space for the scientific culture within its purpose, but it is probably unreasonable to expect some kind of cultural merger with humanists, especially on the terms that legacy journalism institutions have required from technologists during the digital transition. The differences between humanistic and scientific cultures are, in my opinion, too great to be bridged – especially as we begin the exploration of journalistic possibilities from our new AI toolkit and as we automate many of the functions previously reserved for humans.
Instead I think that we need to explicitly recognize and separate humanistic journalism from scientific journalism – the differentiated side of the barbell from the scale side. We need to be clear about which is which – in leadership, in skills, in activities, in products and even in organizations – just as we also need to be clear about the need to transfer the values and mission of humanistic journalism into scientific journalism. This clarity and this separation can relieve both humanistic and scientific journalism of much of their uncertainty and unease, by unambiguously refocusing humanistic journalism on what it does best, while unblocking scientific journalism to stride ahead into the AI future. AI-driven and highly technical forms of journalism are coming, and it would probably be best for society if they developed with the support and encouragement of humanistic journalism rather than with its antagonism.
We also need a research and development ecosystem for scientific journalism. We need first principles applied research on information representation, automated verification, network coordination, incentive structures, socio-technical dynamics and more. We need audience research that applies the full gamut of cognitive science and neuroscience research tools to human information consumption and sensemaking. We need AI and data science oriented unapologetically towards radically expanded societal understanding, instead of as support functions for humanistic storytelling. We need an information ecosystem in which research institutes for scientific journalism are developing concepts, methodologies, frameworks and communities around the application of AI to societal information flows.
Humanistic culture and humanistic journalism will remain as long as humans remain, just as religion has remained long after the scientific revolution. Romantic journalism has enormous value, and the differentiated side of the barbell is real – claims that are supported by a market for romantic and differentiated physical goods that exceeds half a trillion dollars. This essay is itself a kind of humanistic journalism. We really do, however, stand on the cusp of a fundamentally new era, dominated by machines that can fulfil editorial functions. A more technical journalism is arriving – and again, ideally with the values and mission of humanistic journalism – and it can deliver radical new benefits to all of us, as individuals and as societies. Humanistic journalism and its institutions should help facilitate that arrival.