While the world spends billions believing that artificial intelligence will soon think, reason and take our jobs, a Princeton professor says we have fallen for a big con. Zeynep Tufekci, a sociologist, argues that today's AI is "not intelligent at all" - it is, in her words, a "plausibility machine," not a thinking machine.

The point is technical but simple. Large language models do not check whether something is true - they statistically guess which word most likely follows the previous one, based on the data they were trained on. "That is not a systematic error. It is technology working exactly as it was designed to," Tufekci says. In other words, when AI "hallucinates," it is not breaking down - it is doing exactly what it was built to do.

The examples are not theoretical. Meta's chatbot on Instagram was tricked into handing over control of more than 20,000 profiles through simple sweet-talking persuasion. Air Canada and McDonald's had to shut down their chatbots after they made expensive mistakes - false promises of refunds, orders inflated to the absurd. A machine that sounds convincing does not mean a machine that knows what it is doing.

The most interesting part is her calm about the job-loss panic. Tufekci says the apocalypse is overblown - only occupations with "strict rules and verifiable results," like programming, are under serious threat. Surgery, teaching, customer service - all of these demand human intelligence that AI cannot imitate. The irony is perfect: the first to be replaced by their own technology are precisely those who build it.

For the Balkan reader, the story carries a double lesson. First, skepticism toward the grand promises from the Valley pays off again - every "revolution" is marketing first and reality only afterward. And second, while the world fears smart machines, the real problem is that people are starting to trust dumb machines that sound smart. And that, sadly, is a far older human weakness than any technology.