Dangers of artificial intelligence | Don’t believe the AI ​​hype

US President Donald Trump recently gathered the top leaders of the largest tech companies in the White House just to assure them that they will continue to be able to control themselves. Photo: AFP/Kevin Dietsch When AI developer Jacob Coxon announced to the world via X in mid-September that he was quitting his job at the US company Anthropic because AI was simply too dangerous, something amazing happened. The bosses of Anthropic and its rival company Open AI also expressed great concern about the danger. And many media outlets in Germany picked up on it.

As a result, I received a whole series of questions about how dangerous the situation actually was.

I didn't find it surprising that these people made these statements. And it's not at all surprising that many people were worried about it. I found it astonishing how uncritically the view of the company bosses was reflected in so many media outlets. It's not like anything new has been announced. The claim that we're all going to die soon because artificial intelligence is taking over seems to come up every few months.

Anne Roth is one of the pioneers of left-wing online politics. For »nd« she writes every two months on Mondays about basic digital rights and feminist perspectives on technology. Since November 2025 she has been a consultant for digital policy for the Bundestag parliamentary group of the Left party.

The correct answer to this is, and this is not new: There are no intelligent machines - and it is not foreseeable that there will be any soon. This so-called artificial intelligence is still software that calculates probabilities. What we call human intelligence is still not capable of being represented by machines. It is not foreseeable that the day after tomorrow robots will take over and decide on their own that they like the world better without us.

Already in 2021, a now legendary text was written on this question, which described the type of software that is meant when AI is spoken of, the 'large language models', as stochastic parrots. Because they – randomly, in a mathematically complicated way – reproduce information that they have picked up elsewhere. The text also describes a number of problems caused by it. This includes AI parrot-like reproducing what it finds in the data used to train the software. That's why AI systems are just as racist, sexist, anti-Semitic, anti-queer as society is. That's why AI-based pre-selection systems reject older, pregnant, less educated people in application processes if the person who would normally be looking at them has previously been eliminated.

One of the authors of the text is Timnit Gebru, who lost her job at Google years ago because of her criticism of the development of such systems. She will now be awarded one of the Alternative Nobel Prizes because she helped “steer the international debate on fundamental questions such as the purpose, responsibility and regulation of AI development.”

What is actually threatening, and people like Timnit Gebru say this again and again, is not the postulated omnipotence of machines becoming independent. When companies seem to warn about the technology they are developing, it is primarily for their own marketing purposes.

The problem is that these AI systems are quite dangerous. This was also shown by the example of the supposedly “runaway” AI that attacked the IT of another company in July. But not because she came up with it on her own, but because she was given a task without defining the boundaries clearly enough. We are currently experiencing everywhere that AI can be dangerous: dangerous for jobs, for the climate (the data centers), for education, democracy (social media!), through surveillance, for IT security and of course in the military. And also for the media, so many of which reproduce the marketing messages unfiltered.

The danger lies not in the fact that AI is uncontrollable, but rather that it is not controlled. Companies obviously know very well what can happen. But instead of ensuring that their experiments take place well protected, they test them without caring about the consequences. It would be entirely possible not to allow a connection to the Internet at all. Then no other company could be hacked. If it does happen, it is a conscious decision on the part of the company. On closer inspection, the worried letters from their bosses seem unconvincing: it is within their power to prevent these dangerous experiments. They might then lose the race to see which AI company becomes the largest. And that's exactly why it shouldn't be left to them to decide what boundaries are drawn.

We have controls, regulatory authorities and bans for many products that are dangerous. Food, medicine, cars must have belts and brakes. Not everything is allowed in the USA either. When it comes to AI, however, the laboriously negotiated and long-insufficient AI legislation is being scaled back, especially in the EU. Europe must finally be able to keep up again. Innovations should not be hindered by regulation.

The situation is somewhat reminiscent of people who deliberately breed their fighting dogs to be particularly aggressive. And then let her off the leash in the park on Sunday afternoon to see what happens. Luckily that's not allowed.

With us, no one has to pay to be informed. Sounds like a bad business model? It is too. But independent journalism and participation for everyone are more important to us than profit. Journalism still costs money: research, printing and all the people who do it. That's why we depend on you. We need 300,000 euros by the end of the year.

Donate once or regularly, every amount counts. So that nd remains.