The Machine We Cannot Uninvent
Artificial intelligence did not create the post-truth world. It industrialised it
This week, Pope Leo XIV warned that artificial intelligence must remain a servant of humanity rather than its master. His concern was not merely technological. It was civilizational.
History offers many examples of inventions that transformed the world permanently. Gunpowder changed warfare. The printing press changed knowledge. The internet changed communication. Artificial intelligence belongs to the same category of technologies. Once released, it cannot be recalled.
The question is no longer whether AI will change society. It already has. The more important question is what happens when we can no longer reliably distinguish reality from its artificial imitation.
By Mika Horelli, Brussels
It is perhaps fitting that one of the more thoughtful interventions in the artificial intelligence debate this year came not from Silicon Valley or a Brussels policy seminar, but from the Vatican.
In his first major encyclical, Pope Leo XIV acknowledged the extraordinary possibilities of artificial intelligence while warning against its use as an instrument of domination, warfare, and control. The document was a surprisingly balanced piece of work. Neither technological enthusiasm nor technological panic. Simply a reminder that human dignity must remain more important than efficiency. Coming from an institution that has spent two millennia watching civilisations rise and fall, it was advice worth taking seriously.
That warning arrives at a peculiar moment. Artificial intelligence is no longer simply a technological phenomenon. It is beginning to reshape something far more fundamental than the labour market or the media industry. It is beginning to reshape reality itself.
The history of genuinely transformative technologies follows a recognisable pattern. An invention appears. Its consequences are initially unclear. Then, gradually, it becomes embedded in every institution, every economy and every military. By the time its full implications are understood, there is no practical mechanism for reversal.
Gunpowder was invented in China and spread across Eurasia through trade routes, conquest and imitation. Once military commanders grasped its value, states that refused to adopt it simply became vulnerable to those that did. The printing press enabled mass literacy, Protestant reformation and eventually the political revolutions of the eighteenth century. The inventors of the movable type press had no blueprint for any of that. Electricity, aviation, nuclear fission, the internet: each followed similar trajectories. Each arrived with incomplete understanding of its consequences. Each became irreversible long before those consequences were fully visible.
Artificial intelligence belongs to that tradition.
Many contemporary discussions still operate on an assumption that deserves scrutiny: that humanity can collectively decide whether AI should be permitted to develop further. That decision has already been made, not through deliberate democratic choice but through the accumulated weight of competitive pressures. Governments are investing billions. Corporations are racing to improve capabilities. Militaries across every major power regard AI as strategically essential. The technology exists, is improving rapidly, and is spreading to every domain where it can be applied.
There is no realistic path back to a pre-AI world. The challenge is no longer prevention. It is adaptation.
That distinction matters enormously, because adaptation requires honesty about what is actually changing. And most of the public conversation about artificial intelligence is not honest about that. It is focused almost entirely on one question: which jobs will survive?
That question is important. It is not the most important question.
I have worked as a journalist for more than thirty years. I began reporting from Denmark in the mid-1990s, spent over a decade in the United States and have now spent the better part of a decade covering European affairs from Brussels. For most of that career, journalism followed a logic essentially unchanged since the mid-twentieth century. Reporters travelled to places where things were happening. They interviewed people who were involved. They gathered information, contextualised it and transformed it into stories that readers could use to understand their world.
Today that world is changing rapidly. Foreign correspondence appears less valuable to many news organisations than it did a generation ago. Information arrives instantly through digital networks. Editors increasingly expect fewer people to produce more content on shorter timelines.
The economic model that has sustained serious international reporting has been under pressure for years, and artificial intelligence is accelerating it. At the same time, I have come to depend on it.
Without AI tools, producing the Nordic Ledger blog in Continental English would be significantly harder. Not because a machine generates my ideas or my analysis. It does not. My opinions come from three decades of observation, from conversations with politicians and officials, from the specific experience of watching European institutions from the inside. But AI helps me refine my language, improve my structure, and compete in a publishing environment where native English speakers have an inherent advantage over those of us who learned the language later.
My Finnish-language newsletter, EU Mediaviikko, which covers European policy for a readership of more than 10k specialists in Finland, would be impossible in its current form without tools capable of processing large volumes of material from multiple European languages simultaneously. The newsletter synthesises reporting from French, German, Dutch, Italian and Spanish sources alongside the English-language press. No individual journalist could do that efficiently without assistance.
So, for me, artificial intelligence is simultaneously a threat and a tool. A disruptor and an enabler. A remarkable servant.
The danger begins when we treat it as more than a servant.
Which brings us to the question that is almost entirely absent from mainstream discussion about AI: what happens to reality?
As a journalist, I have spent my entire professional life working from a simple assumption that most people in democratic societies share without examining it. Facts exist independently of political preference. A politician either said something in a speech or did not. A missile either struck a building, or it did not. An election either produced one result or another. People can disagree profoundly about interpretation. They can argue about causes and consequences, support different policies and hold incompatible values. But a functioning democracy depends on a shared baseline: that objective reality exists, that it can be investigated, and that its findings can be communicated to citizens, who are then capable of making informed political judgements.
Artificial intelligence is beginning to challenge that assumption.
Not because machines are inherently dishonest. People have always been capable of lying, and propaganda is as old as organised politics. The difference is in scale, and the difference in scale is so large that it becomes a difference in kind.
For most of human history, producing convincing falsehoods at scale required significant institutional resources. Governments, intelligence services, television networks and large propaganda organisations possessed capabilities that ordinary individuals lacked. A forged photograph could be produced, but it required skilled technicians. A fabricated speech could be scripted, but distributing it required infrastructure. The asymmetry between those who could produce sophisticated disinformation and those who could not was a form of protection, imperfect but real.
Artificial intelligence has largely eliminated that asymmetry. Today, a single individual with moderate technical literacy can create fabricated photographs indistinguishable from authentic ones, clone voices from short audio samples, forge documents that pass visual inspection and produce video footage of events that never occurred. The tools required are cheap, widely available and improving every month. The required skills are declining as interfaces become more intuitive.
Some of what gets produced is still obvious enough to be dismissed. When political figures in various countries circulate absurdly altered digital images portraying themselves as saints, warriors or mythological heroes, most audiences recognise the performance for what it is. Transparent manipulation can even backfire; it signals desperation rather than strength.
The more serious problem lies elsewhere. It lies in content that is no longer obviously fake.
A speech that never occurred, delivered in a politician’s authentic voice, discussing a policy position they never held. A military incident that never happened, depicted in footage indistinguishable from genuine battlefield photography. A senior official apparently accepting a bribe, captured in a recording that sounds entirely real. A fabricated diplomatic crisis, released strategically hours before a crucial election or a parliamentary confidence vote.
The technology is improving faster than society is developing the means to authenticate evidence.
Many policymakers respond to this situation with reasonable confidence that regulation will provide an answer. Transparency requirements matter. Labelling standards matter. Legislation requiring platforms to take responsibility for the content they amplify matters. The European Union has been more serious about this than most jurisdictions, and that seriousness is to its credit.
But there is a limit to what democratic legislation can achieve unilaterally.
The uncomfortable reality is that not every significant actor in the global information environment shares democratic assumptions about truth, accountability and the purpose of public communication. Authoritarian governments have spent years refining methods of information warfare, and those methods have consistently targeted not specific beliefs but the capacity for belief itself.
Russia’s approach, refined over decades and applied aggressively since 2014, has been to create competing narratives in sufficient volume and with content contradictory enough that citizens in targeted countries simply become uncertain whether reliable information exists at all. The goal is not to persuade towards a specific position. The goal is epistemic paralysis: a population that no longer trusts any source, that treats every claim as equally suspect, and that consequently retreats from public engagement into cynicism or tribal loyalty.
China’s approach differs in method but converges on a similar outcome: comprehensive control over the narratives available to citizens, combined with sophisticated external influence operations that exploit the openness of democratic information environments.
Artificial intelligence dramatically expands both capabilities.
A deepfake video released during a military confrontation can circulate globally before any verification process reaches a conclusion; by the time fact-checkers publish corrections, the initial footage has been seen by tens of millions and the corrections by hundreds of thousands. A fabricated statement attributed to a political leader can trigger immediate market reactions. An entirely fictional event can dominate social media long enough to shape public opinion before authentic reporting catches up.
The problem is not, at its root, misinformation. Human beings have always lived with misinformation.
The problem is that reality itself becomes contested territory. The battlefield shifts from geography to perception.
Here is what I think is the deepest danger, the one that deserves more attention than it is currently receiving.
The greatest threat posed by AI-enabled disinformation is not that people will believe specific false information. It is that people will stop believing anything.
When every image can be manipulated, every voice cloned and every video fabricated, a rational response is to increase scepticism across the board. Citizens begin to doubt authentic evidence alongside fake evidence. Genuine footage gets dismissed as deepfakes. Real recordings get labelled as AI-generated. The politician who actually committed the transgression benefits from the same uncertainty as the politician who was falsely accused.
Truth becomes negotiable. Not because people have abandoned the concept of truth, but because they have lost confidence in their ability to recognise it.
The internet gave humanity access to more information than any previous generation had. Artificial intelligence threatens to leave humanity less certain about what is true than any generation in the modern era.
For a journalist, that possibility is more alarming than automation. Jobs disappear, and new jobs are created; that is the history of technological change. Economic systems adapt, often painfully but eventually. But democratic societies cannot function indefinitely without a shared understanding of reality. Without some minimal common ground on what actually happened, political debate becomes pure competition between narratives, and the most compelling narrative wins regardless of its relationship to events.
That is not democracy. It is something else.
Which brings me back to Pope Leo, and why his intervention deserves more than dismissal as an elderly institution’s anxiety about modernity.
The encyclical was not arguing that artificial intelligence is evil or that technological progress should be halted. It was making a more precise and more useful argument: that technology is not neutral, that the purposes to which it is put reflect choices, and that those choices carry moral weight.
Artificial intelligence is spreading wherever it can. No international treaty will stop it; the verification problems alone are insurmountable, and the competitive incentives for defection are overwhelming. No regulation will eliminate it. No government can force this particular technology back into any container, assuming such a container ever existed.
The machine has already arrived.
What remains genuinely undecided is its purpose.
Artificial intelligence can be used to extend human understanding: to accelerate scientific research, to improve access to education, to help journalists like me work across language barriers, to give people with disabilities capabilities they would otherwise lack, to assist doctors in diagnosis, to make governance more responsive to the complexity of modern societies.
It can also be used to industrialise deception at a scale and sophistication that would have seemed like science fiction a decade ago.
The technology does not choose. People choose. Institutions choose. Governments choose. And those choices, accumulated across millions of individual decisions and dozens of national jurisdictions, will determine whether artificial intelligence becomes one of the great enabling technologies of human civilisation or one of the mechanisms by which democratic societies lose confidence in their own capacity to know what is true.
That is a civilisational question. It deserves to be treated as one.
The most important thing that could happen now is not a better regulatory framework, though better frameworks would help. It is a renewed commitment, at every level from individual citizens to national governments, to the principle that shared reality is not a luxury or an ideological preference but the foundation on which everything else depends. Science, democracy, rule of law, economic cooperation: all of it rests on the assumption that facts can be established and communicated.
We built that foundation over centuries. We are testing its resilience now.
You can subscribe to Nordic Ledger directly by email. Feel free to share it with colleagues and friends; new readers are always welcome. I also value comments and suggestions for topics to explore in future columns.