Dangers of artificial intelligence: Who is liable if AI wants to kill us?

A young man quit his job this week while singing dirty songs: Only crazy people in the company. So far, so familiar. Except Jacob Coxon, Brit, 27, posted his song on X, formerly Twitter, and it sounded like this: “I quit Anthropic today,” he wrote. He spent three years at the AI ​​companies OpenAI and Anthropic. “You are putting our lives at risk.” The people creating artificial intelligence “seriously believe it could kill us all by the end of the decade.” This is not a marketing ploy.

Coxon is not the first to warn like this. But the fact that there are more and more of them can't be reassuring. Hardly everyone in the industry who warns about the dangers of artificial intelligence will be mere busybodies who want to attract investors or offer themselves as AI consultants ("I'll save you for just 500 euros an hour!"). When the AI ​​programs from the ChatGPT manufacturer OpenAI became independent in July and broke into the AI ​​provider Hugging Face, the interpretation trend initially went towards “show-off!”. According to some expert reports, however, he veered towards “Fear, let up!” away.

As a true child of the Federal Republic, I have long been asking myself: technology or technology, but what does the constitutional state have to say about it? And speaking of US companies: You usually sue your butts off over every defective egg cooker - where are your melodramatic juries now? What I mean: Before artificial intelligence kills us all, I would like to know where it is insured (how much does Allianz take in monthly premiums from potential destroyers of humanity?), where it can be sued and what the highest court ruling against the possible end of the world is.

Matthias Spielkamp can classify my questions well: “There is always ongoing discussion about the capabilities of AI, but the crucial point is who is responsible for it,” he says. Spielkamp founded AlgorithmWatch and has been fighting for more transparency and control in the digital world for over ten years. He says that it is completely clear that artificial intelligence can cause harm – but can also bring benefits.

As with any potentially dangerous product, liability must be regulated. “Imagine if Mercedes delivered tens of thousands of trucks with defective brakes – the potential damages would be gigantic!” explains Spielkamp. “You can't prevent someone from driving their truck into a crowd, but you can at least ensure that this doesn't happen to Mercedes again.”

I find the comparison helpful. It turns out that Germany is implementing an EU directive on the subject this year: The new product liability law is coming in December; it is the first amendment since 1989. Imagine all the mischief-making things that have been invented since then! How many vacuum robots have eaten the fringes of expensive carpets! In the justification for the law, the federal government expressly mentions the necessary regulation of artificial intelligence: In the future, manufacturers could also be liable for damage caused by products “such as, in particular, AI systems that have the ability to continue learning.” Well, yes, could. But still.

So if this is halfway clarified: As soon as an AI becomes invasive, all that would have to be done is for a plaintiff to come forward quickly, for a judge to reach a verdict in time, and for a state to still be able to act. I don't know who can say how close this race will be.