Anthropic's Red herring: Yes, AI will kill us all. But different than you think

At the end of the first film in the Matrix Quadrology, the AI ​​software “Agent Smith” teaches Morpheus, who is trapped in a virtual world, that people always spread, used up all the resources in an area, destroyed it and then moved on. In the film, the war against artificial intelligence has long been lost. The robots have banished people into a virtual world and use their bodies to generate energy. In the real world, this confrontation still awaits us.

Within the next decade, AI systems would have a 10 percent chance of wiping out humanity, according to an Anthropic employee whose company offers the Claude AI model. OpenAI boss Sam Altman made a similar statement to the UN Security Council - his company is behind the ChatGPT application.

The taz is an independent, left-wing and opinionated daily newspaper. We have been arguing in our comments, essays and debate texts since the taz was founded in 1979. We often cannot and do not want to agree on an opinion. That's why you'll find some completely opposing positions here - all part of the very broad left-wing spectrum of opinions.

The claims are so ridiculous that they could only have sprung from the narcissistic minds of the Tech Bros. And possibly their fear that their investors will soon ask themselves why their trillions of dollars have not yet produced a functioning economic model for their companies. In fact, their software is impressive, but not dangerous - and it's not the software, but the tech companies that develop it that are a danger.

This text first appeared in wochentaz, our weekly newspaper from the left!

Every week, wochentaz is about the world as it is – and as it could be. A left-wing weekly newspaper with a voice, attitude and the special taz view of the world. New every Saturday at the kiosk and of course by subscription.

Large language models – often equated with AI in everyday life – produce text. They don't do anything else. Programmers write ordinary software to turn the text into an instruction that is executed on a computer or in the physical world. This is even needed for the most trivial first step, namely displaying the text in a human-readable form on a screen.

So when the AI ​​company bosses claim that their AI has hacked a computer or could even cause the end of the world, then that primarily means: They have programmed software that has these capabilities. They've also written ordinary, switchable software that scans the output text of a large language model, interprets it as a computer command, and then executes it - whether to surf the Internet, power the wheels of a car, or launch a nuclear missile. The language model itself cannot do any of this.

Imagine if someone built a leopard enclosure without fences or a ditch and then ran around breathlessly declaring that the enclosure was dangerous because the leopards had escaped and regularly chewed the faces of passersby. Everyone else would naturally think that the enclosure was built incorrectly and that the builders were too lazy or incompetent to secure it properly. Nobody would buy the absurd reversal.

Software that runs on its own, does not deliver the desired results or even causes damage is a defective product, not a masterpiece. The big AI companies like Anthropic and OpenAI produce malware on a large scale. They try to shift responsibility for their poor production standards by claiming there is no other way. By warning of a fictitious future end of the world, they try to force others to act.

In fact, their software not only produces virtual damage, but also directly threatens the existence of humanity. Large language models are extremely inefficient and their now ubiquitous use reinforces the very real and present climate crisis. Its land use is so huge that its planned size is partly determined by how many “Manhattans” it covers. Their energy consumption is so enormous that dozens of new fossil power plants are planned around the world, and their water consumption is pushing some communities to their limits.

The software is based on a mass theft of copyrighted works. It contributes to the even more rapid spread of disinformation and to the financial overextension of the entire sector. People are now speculating on profits that are mathematically hardly possible. So we risk a global economic crisis. So there are many good reasons to join forces worldwide to stand up to the tech companies. But none of them are super-intelligent robots ready to take over the world.