Weiß die KI, dass sie nichts weiß?

written on November 11, 2026, read November 2026

I think there is a social bias: we hold people in particularly high regard who create, invent, found and develop things. And to a certain extent, that’s right. These people bring progress. They move humanity as a whole forward, often in many small steps, sometimes in one giant leap.

However, I consider another group of people just as important: the explainers.¹ The ability to create something unfortunately doesn’t always go hand in hand with the ability to explain it to others. Over the years, I’ve met quite a few brilliant people who couldn’t explain to anyone what they do – even if their lives depended on it. It’s the explainers who bring along the individual people that “humanity” is made of. So that they understand what is happening, can assess the consequences and can use progress for themselves.

We need these explainers to find our way in an ever more complicated world. Their audience doesn’t have to be large – not everyone is a Carl Sagan, a Neil deGrasse Tyson, a Mai Thi Nguyen-Kim or a Martin Moder, and I’m certainly not. It can simply be your family or your circle of friends.

In my private life, I notice this quite often when it comes to medicine. I sometimes have the impression that there are some terrible explainers among doctors: people who can only communicate within their own jargon and with whom, as a layperson, you’d be better served by silence. I don’t know how much time I’ve spent reading up on a subject in order to explain something that should really have been someone else’s job: someone who can judge whether the explanation is professionally sound and not oversimplified. But here we are.

It’s similar with AI. Whether you consider it the best thing since sliced bread or the fifth horseman of the apocalypse (on a horse faintly glowing blue) – this technology is in massive need of explanation. And since it will foreseeably affect everyone’s life, it has to be explained to everyone. It has even become a topic at family gatherings, over coffee and cake.

That’s why I’m constantly looking for new mental images that help explain AI. And I’ll admit it: I shamelessly help myself to the work of people who have already explained something well. It’s just sometimes so damn hard to find a good analogy.

For this reason, I recently bought a book by the computer scientist Katharina Zweig: “Weiß die KI, dass sie nichts weiß?” (roughly: “Does AI Know That It Knows Nothing?”).

What I particularly like about this book is a small device that I find incredibly important. She points out that we use human terms for much of what an AI does – terms that don’t really fit. Because AIs do something that looks similar from the outside but works entirely differently. “The AI concluded” or “the AI thinks” – but really, it’s neither concluding nor thinking. It’s something else, something that very often delivers very good results, but for which we lack a fitting word. So Katharina Zweig often puts a tilde in front of it: “~thinking”. I think this is a good way to keep different things apart, even when we lack a more precise word for them.²

I notice in myself that this is necessary. Even though I know fairly well how an LLM works and what it does, I often say “sorry”, “please” or “thank you” when dealing with one. That’s not as strange as it sounds: pilots sometimes climb out of the flight simulator pale and drenched in sweat after a particularly difficult scenario, even though they know it was all simulated – because the simulation is real enough in their heads. It’s the same with LLMs: sometimes the simulation of a conversation is so good that we assume there’s someone on the other side who thinks, rather than “merely” ~thinks.

The title already gives away the general direction: it takes for granted that AI knows nothing – the only open question is whether it knows that itself. Still, the book shouldn’t be mistaken for a reckoning with AI. It’s a well-founded assessment of how AI works, what it can and can’t do – and therefore what it’s good for and what it isn’t.

As more and more politicians use the word “AI” at every turn, as if the abbreviation itself were a problem-solving mantra rather than simply the name of a tool, I think it’s important to get an overview: from critical perspectives, positive outlooks and, above all, explanatory, realistic insights. For the latter, Katharina Zweig’s book is really good. Because she is an explainer.³