Panel discussion "Research >> AI"

Before academic events start, while participants and audience find their seats and the technology is set up, classical music often plays. So it is here, a piece for piano and two strings, quite pleasant and vaguely familiar. Maybe it's Chopin. "What you heard," Simone Abels opened the podium, "was AI music." She pointed out that we most often encounter AI (Artificial Intelligence) as a creating entity - generating texts, images and indeed music.

Artificial intelligence can no longer be imagined without universities, the panellists agreed. Whether translations with deepL, data analysis, finding catchy headlines or general brainstorming: AI delivers good, reliable results on average. It is not suitable for research because, as Jill Walker Rettberg noted, it does not provide conclusive explanations. It only shows answers that are most likely according to the amount of data it has been trained with. But that's not a big deal, you have to look at the answers critically and check them as you would with any other source. What is problematic, however, are the biases, the unconscious presuppositions and narrow-mindedness that AI brings with it. One such bias (in the sense of passing off something particular as something typical) is, for example, that AI-generated images of female professors are more often than average shown cropped.

Sebastian Horndasch demonstrated that AI-generated photos are not very good in other respects either. He had an AI generate an image live for the keyword "professors": The picture showed white old men in long robes who looked like they were teaching at Hogwarts. As another example of misleading AI results, Rettberg cited the answer the Amazon chat box "Alexa" spits out when asked how many deaths there were in World War II: "There were 1000 deaths in World War II." It's easy to figure out how she comes up with that. In the Wikipedia entry "World War II casualties" there is a photo of the Battle of Tarawa. Below it is the caption "Over 1,000 American troops died in the fighting." Captions are considered particularly relevant by web crawlers, and so the Alexa algorithm concluded that this info was likely to match the question. "So to be able to use such technologies," Abels summarised, "you need a lot of knowledge and a lot of media literacy." But maybe that doesn't matter, Jan Müggenburg interjected, because after all, new technologies are designed to be as smart as possible - even if they make mistakes along the way: "We don't want trivial machines. We want creative machines." The constant further development, Horndasch added, cannot be prevented anyway; at best, AI-created texts and images could be labelled, as the EU is already planning for 2026.

"What role do universities have in this development?" asked Meikel Soliman. A central one, found Horndasch: "We live in changing times and it needs someone who makes it understandable." On a practical level, Rettberg suggested that teachers be quite open about it and even start their seminars with students generating AI texts - to then examine these texts and identify why this falls short and where exactly the AI is making mistakes. AI in teaching certainly offers advantages, Horndasch said. The social situation of students in Germany is worse than it has been for decades. AIs can have an egalitarian effect here: They are expensive, but not unattainably expensive, and all AI-s (from the cheapest to the most expensive) are essentially equally good. However, Müggenburg warned that one should not only be optimistic, because essential social problems - such as the shortage of teachers - cannot be solved with AI alone. Moreover, the mass use of AI is at least questionable in terms of climate friendliness.

When the podium was opened for questions from the floor, one student pointed out that AI is also used as a weapon of war. In the Ukraine war, for example, both sides used AI to spot missile targets. "Many entertainment, consumer and even research technologies have their origins in war technology," Müggenburg located this historically, "for example batteries or computer games." What distinguishes the Ukraine war in this respect, however, is how technologies are used as weapons there that are not advanced but can also be bought in the supermarket, such as commercially available (recreational) drones.
In response to a student's question about whether art still has a chance at all in times of artificial intelligence, Rettberg replied that this discussion is often reduced to 'good human art versus bad AI art'. Horndasch noted that much human art is also mediocre - works are not better simply because they were created by humans.
Finally, Abels invited the audience to include international perspectives on the topic and to question their own view of national and regional presuppositions. Earlier, a student pointed out: "The vast majority of people who develop AI are Indian. The vast majority of scientific papers on AI come from China. Why does AI reproduce Western hegemony so persistently?

"The panel discussion and our academic discourse as part of the Opening Days," explains Head of the Graduate School Anja Soltau, "expresses on a small scale what we want to achieve on a large scale with the Master's programme and the doctorate at the Graduate School: Scientific penetration of a current topic that can only be grasped in an interdisciplinary way, with impulses from international, renowned scientists, and in community with our entire graduate student community.