AI in schools
“Mistakes are now corrected by AI”

Last year, teacher Hendrik Haverkamp explained in DIE ZEIT how ChatGPT is transforming teaching. What has happened since then? A discussion about the time-saving benefits of artificial intelligence and how it may spell the end of homework.

By Martin Spiewak

DIE ZEIT: Mr Haverkamp, about one year ago, ChatGPT catapulted artificial intelligence into our everyday lives. Has AI now arrived in schools?

Hendrik Haverkamp: At our school, certainly. Most of my colleagues have realised how much easier AI can make their work. For example, it can produce a text at different levels: you can have a sophisticated version for native speakers, and a simpler version for a Ukrainian pupil who only speaks a little German. If I want to introduce a new topic, ChatGPT will suggest different lesson plans which I can then customise for my class. And just think of all the marking ...

…a really time-consuming task for teachers.

Haverkamp: Marking a German exam for a year-10 class used to take me up to six hours. Now, AI can make the first corrections, check spelling and grammar, and do an initial assessment of the structure and content. This reduces my workload significantly.

But it’s still you who decides what mark to give?

Of course. But if teachers can tell the programme what the requirements of the exam are, it will also make a suggestion. We’re still experimenting with this and looking at how grades differ.

How many teachers make use of this kind of help?

I’d say ChatGPT has impacted the work of 80 percent of my colleagues.

Your school has had a strong digital focus for 20 years. When ChatGPT appeared, many schools were initially shocked and wanted to prevent it from being used.

This resistance only lasted for a short while. I have experienced a lot of openness at the training courses that I give. Teachers soon realised that AI in schools is here to stay. Education authorities have reached the same conclusion. The Standing Scientific Commission now advocates the regular use of AI from year 8 onwards in all subjects where texts play a central role. The question has shifted from whether AI should be introduced in schools to how soon this will become a reality.

What are the problems you face?

So far, schools in only three federal states – Saxony-Anhalt, Mecklenburg-Western Pomerania and Rhineland-Palatinate – have official, or GDPR-compliant, access to ChatGPT. This is not acceptable.

The situation appears to be different among schoolchildren. Surveys have revealed that 70 percent use AI for schoolwork, mostly from a private account.

At our school, it’s 90 percent. Pupils use ChatGPT to do research, find explanations about subjects they’ve been taught in class or to get their texts corrected.

How do you verify that pupils have done their homework themselves?

We’ve said goodbye to traditional homework. Instead, we’ve introduced self-study periods at home, where pupils work independently, preparing for class tests or catching up on lessons they’ve missed. They’re required to keep a record of these learning periods.

“To a certain extent, we have to trust our pupils”

What happens in the senior years when all pupils are required to do subject-specific projects?

We’ve taken the bull by the horns, as it were. We encourage pupils to use AI, but they have to highlight the sections they’ve written with the help of AI.

And does that work?

Better than expected. As teachers, we help pupils with these projects. We are well aware of their level of knowledge. And often the topics are so specialised that AI doesn’t help much anyway. But it’s true that ultimately, the school relinquishes some control. To a certain extent, we have to trust our pupils.

And what about exams?

We should allow AI to be used selectively. I call it the calculator moment. For a long time now, pupils have been allowed to use calculators for parts of maths tests. We could adopt a similar approach with AI. Pupils could do one part of the exam on their own, as they have in the past, in the other part they could use ChatGPT. In its recommendations, the Standing Scientific Commission of the Standing Conference of the Ministers of Education and Cultural Affairs (KMK) speaks of “co-activity between pupils and AI”.
  How do you mark work like that?

I think that, in future, schools will concentrate on the process more than just the outcome. How did the pupils come up with their text? Did they scrutinise AI’s suggestions? How did they develop them? And there’s also the quality of the prompts ...

... the queries or input given to ChatGPT.

I can mark them too. For example, by looking at the quality of the chat or how the pupil interacts with the programme. Obviously, depending on the prior knowledge of the pupils, there can be significant differences.

So, it would be wrong to assume that AI makes knowledge less important?

Knowledge is actually becoming more important. For one thing, I have to ask ChatGPT the right questions. I can only do that if I have a certain level of knowledge. Secondly, I have to assess the extent to which AI’s responses are appropriate. And for that I need prior knowledge too.

“We need a broad training initiative in the field of AI”

Is the education gap becoming wider or narrower as a result?

We’ve found that AI improves the performance of very good pupils but offers fewer benefits to weaker pupils. They tend to rely on copy-and-paste.

But there are already signs that AI will bring about significant changes in schools.

I certainly hope so. Conventional exams, for instance, now seem rather outdated. Although schools promote communication and collaboration as important learning objectives, paradoxically they are forbidden in exam settings. I think one of the greatest opportunities AI offers is stronger personalisation in learning.

In what way?

Teachers are usually only able to give feedback to a few pupils in a class of, let’s say 30, when they’re working on an assignment. That’s all the time they have. AI doesn’t have the problem of time constraints. It can tell each pupil individually which questions they’ve answered correctly, what mistakes they’ve made and where they can get help, for example by recommending a short tutorial video.

The promise of personalised tuition through technology has frequently been made.

Now it could work. Here’s one example. Together with a programmer, I developed the tutorial programme “fiete.ai”. Pupils log in with a QR code and work on assignments the teacher has previously uploaded. One question in physics, for example, was: Which form of energy – solar power, wind energy or hydropower – is the most effective? The pupils receive feedback on their answer.

And what does the teacher do?

They not only monitor the pupils’ progress, they also identify areas where repetition is needed, especially when a lot of pupils struggle with the same task. Six thousand teachers have now signed up to fiete.ai. Saxony-Anhalt plans to make it available to its schools very soon.

How many programmes like this exist?

Unfortunately, not that many so far. There is a lack of AI learning programmes specifically geared to subjects like maths, Latin or chemistry. Programmes are being developed, but schools rarely understand the benefits of the new tools.

It’s not something individual schools can really check.

No, this has to be done by the education authorities, in the same way they’ve always carried out quality checks on school books. I think this control function is one of the three major tasks of policymakers.

What are the other two?

Teachers across all federal states need access to a programme like ChatGPT as soon as possible. And we need a broad training initiative in the field of AI. But all this takes time.