AI systems are becoming increasingly powerful. But this makes one question all the more important: can we trust their results? This is precisely what Alexander Jung is researching. As Associate Professor of Machine Learning at Aalto University and, since 2025, as a part-time lecturer at IMC Krems, he focuses on trustworthy artificial intelligence.
Story
How can we trust artificial intelligence?
When AI needs to be explainable
How can systems learn from data stored in different locations? When can we trust their predictions? And how can AI decisions be explained in a way that people can understand and act upon? These questions lie at the heart of Alexander Jung’s work. At Aalto University in Finland, he researches and teaches on trustworthy machine learning, distributed optimisation, and AI ethics and law. In doing so, he combines mathematical fundamentals with the practical challenges of deploying artificial intelligence.
He places particular emphasis on explainable AI. Alexander Jung explores how AI can be communicated in an understandable way and what organisations need to know about its responsible use. Preparing organisations for the requirements of the EU AI Regulation is also a key topic of his lectures and seminars.
Making complex technology understandable
For Alexander Jung, research and teaching go hand in hand. He is the author of the books *Machine Learning: The Basics* (Springer, 2022) and *Federated Learning: From Theory to Practice* (Springer, 2026), and develops teaching materials with the aim of making key concepts of machine learning understandable without reducing them to buzzwords.
His expertise is also in demand outside the realm of university research. Among other roles, he was Principal AI Scientist at Silo AI and a consultant for AI start-ups in the healthcare sector.
International expertise and cutting-edge know-how for our students
Since September 2025, the native of Lower Austria has been teaching on the Bachelor’s programme in Business Administration and Digital Management, as well as on the Master’s programme in Engineering Responsible AI Systems. In doing so, he brings his experience from international research and applied artificial intelligence directly into the classroom.
His combination of research, application and teaching is particularly exciting: the more AI becomes part of our professional and social lives, the more important it becomes to look beyond mere technological knowledge. It is also crucial to understand how AI arrives at its results and where its limitations lie.
We asked Alexander Jung
AI systems are delivering impressive results these days. But from a scientific point of view, when is an AI system actually trustworthy, and how can we tell that we can rely on its results?
Imagine you have a friend who claims she can predict the weather. Would you believe her? Probably not until you’d checked a few things. Firstly: has she been right often, and not just on days she’s familiar with, but also in unusual weather conditions? Secondly: Can she explain to you why she expects rain tomorrow – for example, because of the dark clouds in the west? And thirdly: Does she honestly say ‘I don’t know’ when she’s not sure?
These are precisely the three things we also demand of an AI system. It must have been tested on data it has never seen before and still function reliably. It must be able to justify its predictions in a way that a human can understand. And it must know when it is uncertain. A system that always answers with confidence but never admits that it is guessing is like a friend who never says ‘I’ve no idea’. You shouldn’t blindly trust such a system, especially when it comes to important matters.
Trust, therefore, does not arise from impressive answers, but from verifiability. This is no different in science than it is in everyday life. The more frequently AI prepares or influences decisions, the more important traceability becomes.
What are currently the biggest challenges in making AI truly explainable?
The biggest challenge is that ‘explainable’ always depends on the person receiving the explanation. A doctor, a judge, a bank customer and a software developer all need completely different explanations for the same decision. Technically, we can often show today which input data had the greatest influence on a result. But a list of numbers is not yet an explanation. A good explanation must be tailored to the recipient’s prior knowledge, and that cannot be retroactively applied to a finished system.
The second challenge lies with the models themselves. Modern language models have billions of parameters. No one can understand in detail what individual parameters mean. We make do with simplified explanations, and these simplifications can themselves be misleading. A plausible explanation is not automatically a correct one.
And thirdly, there is an organisational problem: explainability requires effort and, initially, brings no visible benefit during development. The EU now requires it for many applications, and in my view, that is the right approach. But organisations must first learn to factor explainability into their plans from the outset, rather than tacking it on at the end as a compulsory exercise.
What skills should students develop today so that they can not only apply AI, but also critically assess its results and handle it responsibly?
Firstly, a solid basic understanding of how machine learning works. You don’t need to be able to programme neural networks, but you should understand that AI does nothing more than identify patterns in data. This immediately leads to the most important question one should ask of any AI system: what data was it trained on, and does my situation resemble that data? A solid basic understanding also includes the ability to verify results. I advise my students to treat AI responses as suggestions, not as facts. Anyone who cannot assess a result for themselves should not pass it on without checking it. This applies just as much to a sales forecast as it does to a legal text. And finally, an awareness of responsibility. AI does not make decisions; people do. Anyone who implements an AI system within a company bears responsibility for the consequences, and the EU AI Regulation now makes this responsibility legally enforceable. Students who understand both the technical fundamentals and this legal and ethical framework will be in demand in every sector in the coming years. It is precisely this connection that I try to convey in my courses at IMC Krems.