Summary
Social media algorithms reward established news outlets for chasing clicks, and this shapes the unstable information system we live in. 'Surface news' dominates that is dramatic, personality-driven and optimised for engagement. We lose complexity, context and emerging reality, leaving room for deepfakes and propaganda to fill the gap. The news industry could shift from being 'Seekers of Truth' to 'Architects of Context and Verification Pathways' that use AI to help audiences test claims against evidence. That is not a smaller ambition than truth-telling; done properly, it could be a more honest one. But it only works if news organisations solve the trust problem it creates before they build the technology.
The crisis
Today's polarised 'information chaos' is a structural failure in news delivery, sharply accelerated by social media. Take the reporting of the Strait of Hormuz crisis: coverage has mostly lacked nuance, context and sophistication, and the result has been public anxiety and confusion rather than informed debate about the strategic and economic questions underneath it.
- The Surface (noise) - social media and legacy news amplify the theatre of conflict. Coverage of one side or another 'blockading' Hormuz, or rebroadcasting dramatic statements from leaders, is engagement-driven and often partial.
- The System (signal) - the emerging reality is more subtle: a new equilibrium involving oil-price effects, military trade-offs, workaround payment routes and bilateral deal-making. Much of this goes unreported or unexplained, because the tools most major outlets use (on sites like the BBC, CNN, WSJ and NYT) don't encourage that kind of complex exploration.
- The Gap - because coverage stays on the surface, or offers single-perspective commentary dressed as analysis, the public is kept in a state of reactive anxiety rather than equipped for strategic judgement. That is a direct erosion of public value: policy made under panic is rarely policy made well.
The same pattern shows up in coverage of health crises, trade disputes and migration. Surface news lurches from triumph to disaster within a single day; the steady default is negative headlines paired with angry opinion. The long-run effect is corrosive, bringing sustained anxiety, and a slow erosion of public trust in the free press that is meant to inform democratic life. This is not a partisan failure. Populist leaders across the political spectrum have learned to exploit surface-news dynamics; so, in a different way, have mainstream outlets and tech platforms that profit from the same attention economy.
A new construct: news as hypothesis
The 20th-century model treated news as a binary: something is either verified and reported as fact, or it isn't reported until it is. That model still has a place - some things genuinely are verifiable, and where they are, they should be stated plainly as fact, not hedged into false uncertainty. The problem is everything else: the much larger category of unfolding events where the full picture isn't yet knowable, but where audiences are currently offered false certainty anyway, dressed up as reporting.
For that larger category, AI makes a different model conceivable for the first time:
- Probabilistic reporting - instead of a single confident headline, coverage states what is known, what is likely, and what remains genuinely uncertain, and lets the audience explore the reasoning, not just the conclusion.
- A clear boundary between three things currently blurred together: verified fact, calibrated probability, and commentary or speculation, each visibly labelled as what it is.
- Context-seeking - most events can only be understood against historical, economic, technological and cultural context. There is often no single moment of clean verification, only accumulating layers of understanding.
- Signal-testing - claims are treated as hypotheses, continuously tested against live data (shipping and trade data, on-chain transactions, historical precedent) rather than filed once and left unrevised.
The goal is not to make news more vague. It is to stop dressing up incomplete information as settled fact, while staying direct about the things that genuinely are known. News outlets should design their algorithms and interfaces to encourage this kind of exploration, rather than simply rewarding time on screen.
Human–AI partnership: from correspondents to architects
In this model, audiences would still meet a headline, but with the option to ‘dive’ beneath it: AI-guided prompts that open context, competing perspectives and emerging evidence, in place of a single authoritative statement.
Take the Hormuz example. Alongside the headline, a ‘dive’ function might prompt a reader to:
- explore what a blockade would practically require, and what military assets are involved
- look at how past blockades and strait closures have played out historically
- examine what is actually being observed in tanker movements, insurance markets and oil pricing (and also what cannot be known, for instance ships that have switched off their transponders)
- compare how the claim is being framed in say Washington, Tehran, Riyadh or Beijing side by side
This reframes the journalist's role from correspondent (‘I think X’) to architect of context and verification. Three functions follow: surface curation (identifying which signals are worth investigating), pathway design (building the lines of enquiry the ‘dive’ function follows), and AI deliberation (using AI systems to sift data and stress-test claims against counter-evidence, within guardrails set by editors, not by the AI itself).
The interface for this need not be a feed. It could be closer to a deliberative workspace: a gateway (the surface event), and a set of exploration pathways that let a reader seek context, test assumptions against real-time data, or examine how the people involved are likely thinking, rather than being told what to conclude.
This will not appeal to everyone. Most news consumption is fast, low-effort and habitual, and a workspace that asks readers to explore will lose the audience who just want to know immediately ‘what happened’, or simply scroll until their attention is grabbed. That is a real limitation, not a footnote - this model is best suited to readers already inclined to look past the headline, which may be a minority. Its case rests on quality of engagement and subscriber trust, not on replacing the mass-market headline feed.
One key question remains: how much cognitive effort will users put in to arrive at something closer to a more realistic and rounded picture of the world? There may be some indications, from more interactive gaming and research platforms, that ‘effort’ is something that many users are prepared to supply, as long as the stimulation ‘rewards’ are high enough from such interactions.
AI is both the problem - and the possible answer
Most major outlets - the BBC, CNN, the FT, the WSJ among them - currently resist a move towards AI-assisted exploration, perhaps because they see their role as delivering ‘the truth’ rather than probabilistic search, or because they remain wary of AI-generated content generally. That resistance is understandable, but the case for change doesn't depend on AI being flawless. Early generative AI was overwhelmed by hallucination. Better-trained models and more complete underlying data now produce something closer to reliable synthesis – imperfect still, but now often a better starting point than either human-curated headlines or an unfiltered social feed.
There is, however, a harder problem this piece needs to name directly rather than gesture at. The obvious objection is this: if the credibility of legacy news is already in question, why would audiences trust the same organisations to design AI exploration pathways, rather than simply moving the manipulation one layer deeper and making it harder to see? A ‘dive’ pathway is still curated. Someone still chooses which prompts to offer and which counter-arguments to surface, and that choice is at least as consequential as choosing a headline.
The honest answer is that this model doesn't remove the trust problem, it relocates it from ‘do I believe this headline’ to ‘do I trust the process that built this pathway’. That is a real difficulty, and it only works if it comes with real safeguards: published, auditable rules for how pathways are built; visible sourcing for every dive prompt; and independent scrutiny of the editorial guardrails governing the AI, not just of the AI's output. Without that, this model is a more sophisticated version of the problem it claims to solve. With it, it has a chance of being something different, because the guardrails, not just the brand, become the trusted thing.
There is a second vulnerability worth naming. Probabilistic language - ‘there is a probability that X is happening’ - is also the native register of disinformation and stochastic propaganda, precisely because hedged claims are hard to hold anyone accountable for. A system built around calibrated uncertainty needs to be at least as rigorous about tracking and correcting its probability estimates over time as the old model was about running corrections. Otherwise ‘probabilistic’ becomes a shield rather than a discipline.
AI vs human judgement
This argument does depend on conceding that some forms of AI reasoning outperform human judgement in specific respects. Humans excel at synthesising new ideas and forming opinions about them, but are reliably drawn to the sensationalist, the personality-driven, the frightening, and tend to view events through whether we like or dislike the people involved. AI has no equivalent stake in the outcome. Used deliberately, it can act as a partial corrective to genuine human biases: short-termism, emotional decision-making, impulsive engagement.
That correction isn't automatic. AI systems can just as easily be trained to amplify polarisation as to reduce it, as recent years of AI-generated content and AI-optimised social feeds have shown. So, the central question isn't really about the technology. It is about who is trusted to shape it, under what scrutiny, and answerable to whom.
Conclusion: the value of signal over noise
Many in the industry will resist this shift, and some of that resistance is legitimate: it asks news organisations to give up the comfort of sounding certain and being the ‘tellers of truth’. But the public value case is real. When audiences can explore the emerging signal beneath the noise of political and media theatre, they are less susceptible to manufactured panic and manipulation. Policy made in that climate tends to be more considered, not less.
Whether this model can sustain itself commercially is genuinely open. It may be that advertising continues to reward the sensational over the substantive, in which case public-service broadcasters are best placed to build it first, since they answer to a mission rather than just the click-through rate. It may be that audiences, over time, tire of being treated as an engagement metric and start to pay for depth instead. But that is a bet, not a certainty.
What is clear to me is the desired direction of travel: away from news that tells the public what to think, and towards news that gives them credible, accountable pathways to work out what might be happening. That shift only earns public trust if the guardrails behind it are as visible and contestable as the headlines themselves. Get that right, and news becomes a stabilising force again. Get it wrong, and AI-assisted ‘exploration’ becomes just another, harder-to-see version of the instability it was meant to fix.
