AI and the Truth Crisis: We Feared the Wrong Thing

Remember when deepfakes were the great AI bogeyman? We were promised an apocalypse in which we could not tell a video of the president from the output of a neural network. We prepared for a war over truth, armed with detectors and fact-checking. But we seem to have missed the most important thing: AI is not breaking the picture on the screen, but the way we perceive reality itself. It turns out that even when we catch a neural network lying, the damage is already done.

The concept of "truth decay" has been discussed for years, but it has now entered a phase few foresaw. Traditionally, we assumed the solution lay in technology: build watermarks, train classifiers, and everything will be fine. But psychological research shows something alarming. Even when a person knows content was created by AI, the false information still settles in their memory and shapes later judgments. It is a kind of mental virus against which we have no immunity.

What has really changed is the cost of manufacturing doubt. Undermining trust in an event once took the resources of entire media holdings or intelligence agencies. Now a couple of prompts to a language model are enough. But the irony is that the main victim is not the lie, but truth itself. In a world where anything can be fake, the easiest way out for the psyche is to refuse to believe anything at all. This state is called the "liar's dividend": any politician or corporation can now claim that a real, compromising recording is merely "AI-generated."

We spent too long focused on the question "how do we spot a fake?" and ignored the question "how do we preserve trust?" While engineers try to embed invisible marks in pixels, society is losing its common ground. We used to argue over the interpretation of facts; now we argue over whether those facts exist at all. And here lies the main trap: the more suspicion of AI we train into ourselves, the more vulnerable we become to manipulation in the real world.

The situation is made worse by social media algorithms, which still reward engagement rather than accuracy. AI content that provokes rage or delight spreads faster than any debunking. The result is an environment where truth is simply unprofitable in terms of reach. We are building digital infrastructure on a foundation of quicksand, and there is not a single serious architectural solution in sight that could fix it.

Instead of hunting for a technical "silver bullet", we will have to admit that the truth crisis is not a technical bug but a social pathology. We got used to trusting our eyes and ears, but that era is over. Trust must now rest on the reputation of institutions and the transparency of processes, not on image quality. But are we ready for such a radical rewiring of how we think?

The bottom line: the problem is not that AI lies too well, but that we too easily stop believing the truth. Can we rebuild trust before it finally becomes an artifact of the past?