AI poses a subtle danger beyond job loss or weapons: epistemic drift—the gradual shift in what societies accept as “truth.”
Unlike past media shifts (printing press, radio, TV), AI does not just report reality—it enabled deepfakes and synthetic data.
History shows fragile trust—like the 2013 hacked Associated Press tweet that briefly wiped billions from markets.
Artificial intelligence is usually discussed in loud, dramatic terms. Some fear mass unemployment. Others picture autonomous weapons or machines breaking out of human control. Those risks matter. But the most danger is quieter, more gradual, and harder to see. Call it epistemic drift—the slow shift in what people, communities and even entire nations accept as reality.
Truth has never been set in stone. What counts as “real” has always depended on the tools we use to perceive it. The printing press spread literacy and created new political publics. Radio sped up markets and wars. Television turned politics into performance. Each technology changed not only what people knew but how they came to know it.
Artificial intelligence marks a sharper break. It does not just report reality—it can fabricate it. A photograph once suggested something had happened. A deepfake can conjure an event from nothing. A chatbot can generate experts, citations or whole studies that never existed. The danger is not simple distortion but substitution: one reality quietly swapped for another.
A sudden collapse—say, a deepfake swinging an election or an algorithm sparking a market crash—would be visible and dramatic. Drift is more dangerous because it is slow and largely invisible. Think of geology. The Colorado River did not carve the Grand Canyon in a single flood. It wore away rock grain by grain, until centuries later the canyon yawned wide. Epistemic drift works in the same way. Each small shift in what people trust, share or doubt seems trivial. Yet over time the ground beneath societies is altered. By the time the change is obvious, it is too late to reverse.
Signs of drift: The drift is already under way.
Misinformation inflation. With AI, almost any claim can be backed up with convincing “proof”—fabricated images, fake quotes, synthetic witnesses. The line between truth and invention blurs.
Trust outsourcing. Algorithms decide what is worth seeing. The old habit of checking credibility for oneself begins to fade.
Synthetic memory. Platforms already curate collective memory by choosing what to highlight and what to bury. AI will edit the archive even more aggressively.
Artificial intimacy. People are already forming relationships with AI therapists, companions and influencers. These voices shape emotions and norms, even though they are not part of human society.
Each of these shifts looks manageable in isolation. Taken together, they amount to a reconstruction of reality itself. The effects stretch far beyond culture. Markets rely on reliable anchors—earnings reports, supply chains, consumer demand. If AI can produce not just fake reports but entire synthetic datasets, those anchors may drift.
Businesses already use AI to track risks, competitors and customer moods. But what if the data itself is AI-generated? The loop becomes self-referential: machines feeding on their own inventions. The threat is not one spectacular fraud but the gradual erosion of confidence in information. Markets without anchors begin to float.
History shows how fragile confidence can be. In 2013, a hacked Associated Press tweet falsely reported explosions at the White House. In minutes, the Dow Jones fell by 150 points, wiping billions in value before recovering. That was just one false tweet. A future of AI-generated, persistent misinformation could corrode confidence more deeply.
The risks to democracy are starker still. Democracies depend on a shared field of facts. Citizens may disagree on policies, but they must agree on what has actually happened.
Imagine one group of voters seeing a video of a leader confessing to corruption, while another sees a different video of the same leader in tears at a rally. If both are convincing deepfakes, no common ground exists. Debate dissolves into parallel realities.
Authoritarian states face the opposite incentive. Instead of censoring, they can overwhelm. By flooding citizens with synthetic material, they make it impossible to distinguish truth from falsehood. That is not simply information control. It is reality control.
The geopolitical risks are large. The battles of the 20th century were ideological—capitalism versus communism, democracy versus fascism. The battles of the 21st may be epistemological: fights over what counts as reality itself.
Epistemic drift will not unfold evenly. Some societies will embrace AI-generated knowledge wholesale. Others will resist or regulate it. Still others will blend it with older traditions.
Picture one legal system that accepts AI-generated evidence in court, and another that bans it. When such societies trade or negotiate, the clash will not only be political but ontological—over what qualifies as a fact. Within countries, divides will sharpen. Wealthy elites may afford tools to authenticate content, while most citizens swim in a feed where truth and fiction blur. Reality itself risks becoming a privilege.
Why humans fail to notice? Human brains are tuned to sudden shocks, not gradual erosion. We spot predators in the bushes, not the slow creep of climate change. Epistemic drift exploits that weakness. By the time people realise that their categories of truth and falsehood have shifted, the gap is already unbridgeable. Conspiracy theories show how it works. They spread slowly, one meme or video at a time, until entire communities inhabit alternative worlds. Generative AI will accelerate this process by delivering personalised falsehoods tailored to every bias. The danger is not one big lie but the normalisation of countless small ones.
How to respond! Stopping drift will not be easy. Traditional fact-checking is necessary but insufficient. It assumes there is a stable ground to compare claims against. Drift erodes that very ground. What else might help?
(a) Epistemic literacy. Citizens need to learn not just to doubt claims, but to understand how realities can be built, manipulated and reshaped.
(b) Clearer boundaries. AI-generated content should carry labels or cryptographic watermarks to flag its origin.
(c) Trusted archives. Societies should invest in tamper-resistant records of events, secured so that history cannot be easily rewritten.
(d) Slower cycles. Markets and political institutions could build in pauses for verification before acting on AI-driven acceleration.
These measures are scaffolds, not solutions. The deeper task is cultural: learning to live in a world where reality is more fluid, without losing the ability to anchor it.
Epistemic drift is frightening because it strips away the illusion that reality was ever pure. History shows that knowledge has always been shaped by tools, institutions and power. AI does not invent unreality. It makes its constructed nature impossible to ignore. Handled wisely, this could even be an opportunity. Societies might learn to treat truth not as a given but as a commons—fragile, shared, and in need of defence. Handled badly, the result could be fragmentation into epistemic tribes that no longer recognise one another’s worlds.
The most serious danger of AI is not conquest but calibration. Machines are already nudging our sense of reality, one quiet degree at a time. Future generations may wonder how early 21st-century societies let truth dissolve. The answer may be depressingly simple: because drift was profitable, convenient and comfortable.
Epistemic drift does not shout. It whispers. And if we fail to hear it, the silence of truth may arrive long before the noise of catastrophe.
(Nishant Sahdev is a physicist at the University of North Carolina at Chapel Hill, United States and a columnist for Mid-Day, The Tribune, and Mint.)
