The problem is not that AI learns from AI. The problem is that people stop learning from reality..

Lately, we have been hearing more and more claims that artificial intelligence could become a problem once it starts learning from its own content. At first glance, this sounds logical: a system that closes in on itself, loses touch with reality and starts cycling through the same patterns. But haven't we, in fact, already seen all of this in humans too?

But this actually misses the point. Not because that problem does not exist, but because it is not the central issue. The central issue is not how AI learns, but how people learn, or rather, how they are learning less and less.

People do not learn from information in the way we like to think. They learn from situations in which reality “corrects” them. Only when an assumption fails, when a decision produces a consequence, does a correction take place. Without that, there is no real learning, only an interpretation that sounds convincing until we put it to the test.

AI works differently. It does not know what an error is in terms of a real-world outcome. It does not know what it means to make a bad decision and bear the consequences. It recognizes patterns and calculates probabilities. In other words, it optimizes for what appears most frequently, not for what is necessarily correct.

When such a system starts learning from its own outputs or solutions, the result is not a breakdown, but something far more dangerous, because distinctions are lost and deviations disappear. Everything slowly “levels out” toward the average. The result is content that makes sense, that is orderly and logical, but that rarely offers anything new or genuinely valuable.

But even that is not yet the real problem. The problem arises when people start doing the same thing.

When they stop checking what they receive. When they stop questioning conclusions that sound sensible or logical. When they stop testing their ideas in a real-world context or environment.

That is when we lose what has kept things in check until now: contact with the real world. With reality itself.

At that point, a closed loop emerges. AI generates content, people use it without genuinely checking it, and that content feeds back into the system, further reinforcing existing patterns. There are no major errors to expose it all. No dramatic developments. Everything works, but at a level that makes no difference.

Organizations in such a system do not fail suddenly. They slowly become average.

And being average is the most dangerous state we can find ourselves in. Not because it is bad, but because it looks good enough that no one questions it. Decisions are reasonable, processes make sense, everything seems stable, but there is no progress, no differentiation and no real competitive advantage.

In a world where everyone has access to the same tools and the same information, technology is no longer what makes the difference. The difference comes from how people use what they have. More precisely, from whether they can recognize when something that sounds convincing is actually not good enough.

So the question is not whether AI learns from AI. That is a technical detail. The question is whether humans remain a corrective force or start behaving like a system that merely reproduces what it is given.

The moment that happens, learning in the true sense of the word stops. All that remains is repetition. And where there is only repetition, no value is created.