Interview: How ChatGPT is changing teaching

Robert Lepenies, President of Karlshochschule International University in Karlsruhe, wants to integrate artificial intelligence into seminars. In an interview – which c’t conducted with him by e-mail – he advocates a new relationship between man and machine.

c’t: Mr. Lepenies, you say that certain types of exams are now unthinkable because text AIs in the social sciences can produce seminar papers or other papers that are indistinguishable from those of students.

Lepenies: There are studies that even experts cannot distinguish between human and artificial expertise. For example, in an experiment, philosophers Eric Schwitzgebel, Anna Strasser, and Matthew Crosby asked people if they could tell which answers to deep philosophical questions came from philosopher Daniel Dennett and which from GPT-3. Even Dennett experts had trouble distinguishing GPT-3 texts from Dennett’s work (The Computerized Philosopher: Can You Distinguish Daniel Dennett from a Computer?).

The fact that AIs can produce scientific texts will not be limited to the social sciences and humanities. Other fields of science and all forms of education will soon be affected. In concrete terms, this means that it will be difficult to assign and evaluate term papers. AI shakes us awake here and makes us ask: Is our view of what “education” and “achievement” mean still up to date?

c’t: Do you also know areas where the AI ​​isn’t that advanced yet?

Lepenies: She still just makes up scientific literature, for example: She “hallucinates” some – but not all – scientific sources.

c’t: How can universities deal with the fact that written essays can no longer be used for evaluation?

Lepenies: I spoke to our Vice President for Research, Wendelin Küpers, who confirmed that the independently prepared written elaborations and their evaluation remain important, because in a university of the future, students have an intrinsic interest in writing their own essays – after all, students want to learn , and make questions your own and take an evaluative position.

At Karlshochschule, we rely on creative and plural forms of examination anyway – which means that we may have less of a problem here than the big universities. More important than grades is qualitative feedback and joint reflection on learning.

c’t: Where can chatbots help in teaching?

Lepenies: GPT-3 is a potentially very intelligent and stimulating feedback partner. The smarter you ask the question and give feedback to the model, the smarter the answers will be. In most cases, the text AI can provide the first draft for any document.

For example, we could create new module descriptions within a short time or supplement and update existing ones with new learning objectives or literature. An example: we wanted to include a few case studies from the Global South in a fairly theoretical seminar on ethics and globalization – we then simply had the AI ​​work out the first version – which then gave us examples, for example a session on feminist issues labor movements in Latin America. We would certainly have figured that out ourselves – but not after five seconds. At the same time, we let the AI ​​suggest an evaluation grid for group presentations – we then of course have to check and validate this again.

c’t: You express the fear that students could delegate their own (re)thinking to ChatGPT & Co. Isn’t that the usual skepticism that accompanies every new technology, every new medium?

Lepenies: Like every technological development, this one is ambivalent. Wikipedia or auto-correction in word processing don’t make us dumber, but both (rightly) question what an educated person needs to know and be able to do today. But text AIs can do much more. Here we have to learn to correctly classify and evaluate the results of the machine and to develop interpretations and our own judgments. The key question here is: Where and how do we learn these basic skills? Of course I would say at our university.

The university is also about learning experiences together with other people, through questioning, through discussions in groups, through aesthetic and ethical arguments, through learning from other people, through the interpersonal – AI cannot do all of this alone.

c’t: You suggest using ChatGPT as a participant in creative group discussions. How does this work?

Lepenies: Some of our teachers have already tried this: they pretended that GPT was a participant in a discussion group – it was co-creative and stimulating. It is best to use GPT together in a group – and thus dissolve this bilateral man-machine relationship: discuss the AI, but also joke about it.

c’t: Beyond the teaching, where do you think systems like ChatGPT will make the biggest changes?

Lepenies: In the state parliament of Baden-Württemberg, Alex Salomon (the Greens, editor’s note) has just given the first speech written by AI. Of course, I’m now thinking about that for future speeches…

In general, I suspect that AI will revolutionize all activities that have to do with text or images, also in a linked form: “Write me a children’s book for my master’s thesis in the following languages, film it and compose the film music for it” – something like that will happen to be possible.

(yeah)