You can’t spell IIASA without AI

Quick experiment.

Before reading further, answer the following question:

What should AI be used for?

[] Solving climate change
[] Accelerating science
[] Improving public services
[] Increasing productivity
[] Making cat videos

If you didn't tick the last box, why not?

Last week, at the IIASA Interaction Festival, where staff members, collaborators, and Council members gathered to think about the future of international scientific cooperation, the question landed as a joke. I’m not convinced it should have.

The festival was established for a meaningful purpose. At a time when geopolitical tensions are rising and many of the institutions that once underpinned international cooperation are under strain, IIASA remains an unusual place: a space where people from different countries, disciplines, and political systems still sit down together and try to solve problems. Science diplomacy, in practice.

Or, as one Council member put it, perhaps the most important thing is that people keep talking to one another: listening carefully, disagreeing respectfully, and staying in the conversation precisely when agreement is hardest to find.

That sentiment lingered in the room during the panel on AI and digitalization. On paper, the discussion was about artificial intelligence as a strategic enabler across IIASA's research domains. In reality, it was about something larger: how the Institute should respond when the production of knowledge itself is being transformed.

Moderated by Kai Kornhuber, the panel brought together Verena Kain, Marcial Sandoval-Gastelum, Tamas Galosi, and Council members Olli Varis (Finland), KC Moon (Korea), and Lien Le (Vietnam). The conversation moved between practical applications and uncomfortable questions. AI can help scientists identify patterns in complex datasets and accelerate parts of the research process. Where does that leave institutions such as IIASA? And what happens when the tools shaping science are increasingly owned by someone else?

One speaker described using AI to identify promising interactions between genes and proteins before validating them experimentally in the laboratory. Another highlighted the use of large language models to make complex systems models more understandable for decision-makers. Across disciplines, the message was similar: AI is becoming a research partner.

Yet there was a noticeable difference in emphasis between generations. Younger participants often spoke about AI as something already embedded in their daily workflows. More senior participants tended to return to questions of judgement, interpretation, and responsibility. The contrast was not a disagreement so much as a difference in perspective. One group discussed what AI can do; the other focused on what humans should continue doing. That distinction matters.

Only weeks before the festival, Nature published an article on The AI Scientist, a system capable of generating research ideas, reviewing literature, writing code, and drafting scientific papers. One AI-generated paper was accepted to a machine-learning workshop after peer review. Not groundbreaking science, according to its creators, but good enough.

Good enough changes the conversation.

For decades, discussions about AI focused on automation of manual labor. Increasingly, the target is cognitive labor. Writing, reviewing, analyzing, synthesizing, even generating hypotheses – activities that universities traditionally regarded as deeply human – are becoming partially automated.

This is where the discussion returned to ethics. The real question is not whether AI should be used to solve climate change rather than generate cat videos. The real question is who gets to decide.

Today, the most powerful AI models are largely developed by a small number of companies, concentrated primarily in the United States and China. The infrastructure, training data, computing power, and increasingly the interfaces through which knowledge is produced sit within ecosystems that are neither neutral nor public.

Several panelists highlighted concerns around data sovereignty and governance. As AI tools become more deeply embedded in research workflows, questions of transparency, control, and accountability are becoming harder to ignore.

For Europe, and perhaps especially for an institution like IIASA, this creates both a challenge and an opportunity.

The Institute's strength is bringing together disciplines, countries, and perspectives that rarely meet in the same room. If AI is becoming a new layer of global infrastructure, then questions of governance, trust, ethics, and international coordination are not side issues. They are systems questions.

And systems questions are IIASA's territory.

So let's return to the cat videos.

Perhaps the problem is not that AI can generate them. Perhaps the problem is assuming there is an obvious distinction between worthwhile and trivial uses of intelligence. During the discussion, cat videos came to represent the supposedly less worthwhile side of AI. But the joke hides a serious question: who gets to decide what is worthwhile? Scientific breakthroughs, after all, often begin with curiosity rather than utility. The internet itself was once dismissed as a playground for distractions. Today it underpins research, commerce, politics, and everyday life. 

The more pressing question is whether, in an age of increasingly automated science, we retain the ability to collectively decide what knowledge is for, who benefits from it, and which futures we are building towards. 

That is not a technical question. 

It is a political one, an ethical one, and increasingly, a scientific one too.

Note: This article gives the view of the authors, and not the position of the IIASA Insights blog, nor of the International Institute for Applied Systems Analysis.