A specter is haunting the schools. In teachers' rooms, people whisper about its amazing power. Some want to chase it away, others downplay it. Students announce on social media: "Never again homework!"

> The ghost is called Chat-GPT

It is the new chatbot based on artificial intelligence from the company Open AI. But what is a chatbot? Chat-GPT answers questions, formulates texts, summarizes specialist knowledge, at the touch of a button. Not always, but very often correctly.

The program generates home essays on all conceivable topics – you can choose whether to use simple language or a more sophisticated tone. A philosophy teacher complains: "This program writes the presentation cards for my students." Chat-GPT sometimes formulates something generically, but the quality of the texts is sufficient for a passable bachelor thesis. And no plagiarism detector in the world finds errors, because: The texts are not plagiarism. Chat-GPT does not copy, but regenerates its texts with each request.

The reports from educational institutions that ban the program across the board are piling up. The Paris University of Sciences Po threatens to be expelled if students use the software for their work without making it transparent. New York City schools have blocked access to chat GPT on their devices and networks. Has the homework come to an end? Are there grades now only for oral exams or tests under supervision?

> The AI is overwhelmed with personal essays

The high school teacher and didactics lecturer Philippe Wampfler does not believe that Chat-GPT changes everything. "Tasks like 'Summarize the second chapter of 'Effi Briest' were already obsolete before." Students have always found methods to minimize their effort. A smarter question that hasn't been answered hundreds of times on the internet is: Which scene in the second chapter touched you personally and why? Giving an authentic answer to this is also overwhelmed by the AI. Wampfler generally advocates getting away from the means of pressure of grading. "Ideally, you convince the students that they get more out of doing their work independently." That doesn't rule out control. However, this should not only look at the end result, but also accompany the development process. “The pocket calculator can also divide. Because you think it's important that children learn it anyway, the solution on paper also counts."

In the case of writing tasks, you can discuss the task verbally with students, then let them collect arguments and examples and only write something in the third step. The AI can also come into play here: You can test which arguments Chat-GPT proposes, depending on how you ask. Think about where these come from and where the blind spots are. «Speech AI will become a relevant tool in the working world. So you should learn in school to deal with it critically,” says Wampfler. He's a long way ahead of that. Most of his colleagues are busy trying to stop the use of AI.

> Why does the AI only draw women in 'sexy'?

What it means to playfully learn how to deal critically with AI can be observed in the Sole learning center in Mollis. In this free school, children are supported in activities that they feel like doing, based on the educator Maria Montessori. In contrast to the pure Montessori teaching, digital devices are expressly desired here in addition to books, watercolors and building blocks. The headmaster and learning guide Nils Landolt showed them the new AI programs that spit out images and texts. Now try some kids.

A boy entered "sexy" into the image generation program. 10-year-old Nea asks: "Why are there only pictures of women?" Landolt says: "Because there are mainly these types of pictures on the Internet with the keyword." – «Why?» asks Nea. Landolt hesitates. He tries to explain that a one-sided male view of women is particularly present online. Another tutor intervened: “If we were to ask around here, everyone would define 'sexy' differently for themselves. But the AI just learns that from the Internet.” A learning situation like out of a picture book. The subject of sexist distortion in AI has arrived.

Landolt believes that playful handling of the new technology is the best preparation for children. His motto is: "Do we want to produce mediocre robots at school or great people?" Landolt launched a media-effective appeal as early as the summer: He invited high school students to have their Matura thesis written by an algorithm. At that time he could not convince any students. This school year one or the other should come up with it themselves. And the questions that Landolt wanted to draw attention to at the time with his School Change Foundation are suddenly on the minds of many: What kind of school will do justice to what is felt to be accelerating innovation in the world of work? 

Landolt is quick to warm up when it comes to the future: diplomas are losing their value faster and faster, CVs are becoming increasingly irregular, and lifelong learning is becoming increasingly essential. “Basically, it’s good that the AI is now doing a lot of the work for us. The only question that arises is what people should then live on.” In the longer 

term, he believes, an unconditional basic income could be part of the solution.

> Ordering something from an artificial intelligence needs to be learned

Meanwhile, ten-year-old Nea is trying out delegation to the AI. With moderate success. Nea wants the AI to help her write a children's book. It's supposed to be a street dog. Chat GPT is to describe him. Nea enters in the field: "Buddy is 'bäsch' and shaggy." Chat GPT: «I'm sorry, but I can't tell you who Buddy is or what he looks like without more information. It would be helpful if you could give me more information about Buddy so I can better help you." Nea: "Buddy is a dog 'b zotige' (shaggy)." Chat GPT: «Oh, I see. I'm sorry I couldn't help you better. If you give me more information about Buddy I will do my best to help you." 

“Buddy is a dog with short, shaggy hair that is brown,” Nea inputs with the help of the tutor. Chat-GPT replies immediately: «Thanks for the further information about Buddy. It sounds like he could be a nice dog with his short shaggy hair that is brown. Could you tell me more about Buddy, like what he's like or what he likes to do? I would like to know more about him."

Nea laughs and sighs: "I think I'll write it myself.“ Even the illustration skills of the AI have disappointed them. Giving the artificial intelligence instructions that bring exactly the desired result is an art in itself. It takes a lot of knowledge about the context from which the AI comes to find the right keywords.

For children it works easier with the Google image search. There Nea finds pictures of dogs. When she's looking for a picture showing the evil dog catcher's vehicle, even Google comes up against its limits. Nea asks her classmate Gabriel, who is sitting at the table next to her, engrossed in drawing, for help and asks if he can draw the dog catcher for her. A reassuring insight: Nea doesn't let any ideas be forced on her. She has her own story in mind. When the computer suggests other things to her, she finds ways to defend herself. 

Her school friend's drawing is not as perfect as the one on the computer: but it corresponds to what he and Nea want to express. The Sole learning house has only been around for a year and a half. A lot has changed in them too. iPads now have passwords so that children don't watch YouTube videos without permission. The digital devices invite you to let yourself be sprinkled. Children don't always come to "pro-summieren instead of consuming" as Landolt imagines.

> Without basic knowledge, we make ourselves a tool of the AI

As a professor of robotics and director of the Center for Learning Sciences at the École polytechnique fédérale de Lausanne, researcher Francesco Mondada has been observing for years how digitization is changing learning. 

With research and workshops, he works to ensure that AI is not only debated in the niche of an alternative school, but in all schools in Switzerland. "A basic understanding of how AI works, that it generally reflects what you put into it, should be part of compulsory education."

Mondada considers the current hype about chat GPT to be exaggerated. "But I'm happy when people think about the skills children need to be mature in the digitized world." He warns against only looking at this latest innovation. Instead of checklists for Chat-GPT, teachers should be given principles that will remain relevant for the next wave of innovation and the one after that. Because anyone who is in school today will be confronted with completely different tools in working life.

With a research group, Mondada has worked out principles that will also apply in the future. The aim is to enable students to think about the new tools. You should be able to assess: What is the program that I have in front of me - and how do I want to behave towards it? “You need a basic knowledge of these tools in order to use them in a self- determined manner. If we don't have that, we ourselves become a tool of technology." 

He himself is preparing for an exam in robotics at the time of the interview. Students can solve them at home. Mondada is confident that he can still evaluate the work fairly. He is not afraid of Chat-GPT, on the contrary: he used the program himself to refine his exam questions.