Opening Days 2026: “Good Science Needs Uncertainty”

Professor Osman, how much uncertainty can science tolerate?

Magda Osman: A great deal. Without uncertainty, no questions arise. However, the term is not as clear-cut as it initially seems. Uncertainty has two sides: it can stem from a lack of knowledge (epistemic uncertainty) or it reflects something about the inherent unpredictability of the world (aleatory uncertainty). Also, what often gets confused is confidence and uncertainty. Sometimes there two things are treated the same, and there is a lively discourse on this in scientific literature.

Does uncertainty automatically mean ignorance?

No. We can know something and still have a certain degree of uncertainty about it. Science seeks to reduce uncertainty by increasing what we know and correspondingly the confidence we have in our claims. We ask questions because we want to expand our knowledge. But we also ask questions when we want to test assumptions.

Isn’t it also a risk, conversely, if I’m too certain of my scientific results?

Then you should ask yourself again: Have I actually verified the results sufficiently? It’s worth questioning your own assumptions and reviewing your own understanding. Good science also means being willing to question your own convictions because sometimes they are held too strongly at the expense of what the evidence is telling you.

What about transdisciplinary research? Does uncertainty increase when scientific findings are applied to the real world?

Yes, often it does. I, for example, want to understand how people make decisions under extreme uncertainty. In the lab, I can create situations in which people are confronted with uncertainty. Then I can carry out highly controlled experiments and test theories. These results are of interest within the scientific community. But how do they benefit people outside of research? If I can identify these different groups, I might be able to develop a training program that helps people learn to cope better with uncertainty. But then I have to ask again: Does this training actually work? How should I present the information? Which method is effective? And all these questions bring additional complexity and additional uncertainty. In the lab, you can control many factors. In the real world, however, numerous factors interact with one another and this makes things harder, but more interesting as a scientific researcher.

Science works with data and models. How reliable are they, really?

Part of the problem always lies in the measurement. We must ask ourselves: Is our measurement reliable? Are our measurement tools valid? That means, does it actually measure what we have asserted it measures? That is a fundamental question in all sciences. We try to reduce uncertainty by using different methods and gathering additional evidence. But even then, we must take into account that results can be influenced by assumptions and measurement issues.

What can cause measurement issues?

We always face a problem because we’re trying to measure something in reality. As a scientific community, we must agree on which measurements are meaningful and which methods we can use to gather evidence. And we must be able to distinguish: Is what we’ve found actually a phenomenon, or did it just happen by chance? That’s why we have statistical methods. Confidence intervals, p-values, effect sizes, are all ultimately part of the infrastructure we use to try to determine how accurate we are in the findings we are generated based on the measures we are using. So, you can’t just say, “I’ll let the data speak for itself.” You can’t because the data needs interpreting, and this is why we really need theory to do this properly. 

You’re referring to the assumptions we bring to our research. There’s the concept of “motivated reasoning”—the practice of interpreting evidence in a way that fits one’s own expectations. How can a researcher avoid this cognitive trap?

If I conduct an experiment — for example, testing a decision strategy to help make an effective choice — and find that it wasn’t effective against a control, then I have to accept that. No result is also a result. Even if I had previously assumed that the decision strategy would work, or others have previously found that it had worked. I can’t say, “I know my assumption is correct, so the result must be wrong.” I have to accept what the data shows, while at the same time considering the reasons for why it didn’t work by going back to theory, as well as looking at the measurement tools we used, and the statical methods themselves.

Artificial intelligence is currently lulling us into a false sense of security. What advice can you give young people when it comes to dealing with generative AI?

Be skeptical! Understand how answers are generated. These systems are not currently designed to simply go through the scientific process and discover new knowledge. They’re designed to process existing knowledge and generate answers from it. But what they can do is prompt us to ask better more precise questions which improve our skills as well as the answers AI gives us back. This is the situation for now, but this will likelychange in the future. AI systems will get better at generating questions of their own and developing data sets through experiments that can be used to test scientific hypotheses. In science, we’ve developed methods for verifying knowledge: experiments, replications, statistical methods, peer review, and critical discussion within the scientific community. We need to accept the fact that AI can work with us, and this means having a re-think about the fundamental practices of the scientific community.

Would you like to ban AI from universities?

No, on the contrary — I actually have a problem with how academia often handles AI. We must ask ourselves: How do we prepare students to work responsibly with these technologies when they leave university and enter the real world? I think there’s still a great deal of resistance here because many academic institutions simply want to ignore AI, and are poorly prepared to handle the seismic changes that AI presents. Not least because there is work showing that AI is as good at, if not better, at teaching students than academics. Just as with research, with teaching we need to understand the capabilities of AI and the needs of the students to prepare them for the professional world they will be entering. I’d say that for teaching and research, above all, we need to understand how to educate and train people to use AI critically and effectively, and that includes the academics as well as the students.

Will AI significantly change academia?

Yes, it will. And that’s exactly why we need to talk about how to deal with it. It can do things that we can’t, and it can do things better than we can, so we need to find how it complements our capabilities, while making sure that AI is designed to align with our values and interests. 

If there was just one piece of advice you’d give to our new master’s students and doctoral candidates, what would it be?

Remain critical and decide on what you are prepared to challenge. This is the driving force behind the questions I ask, and what I do in my research. A large part of scientific work should consist of remaining critical. If someone makes a certain claim and suddenly everyone else starts treating that same claim as a given, well, when that happens, I become immediately skeptical. It takes a lot to establish ground truths, and to get to that point we have a responsibility to evaluate strong claims that are made with such conviction that they effectively stifle challenge. This is precisely where we need to look at the methods, the theory and the evidence to judge how certain we can be of the claims being made. 

Thank you very much for the interview!

Prof. Dr. Magda Osman is a psychologist and professor of decision research at the University of Leeds and a Research fellow at the University of Cambridge. Her research focuses on decision-making under uncertainty, risk, human agency, and behavioral change, and she explores the application of psychological insights to public policy making.