Versicherungskammer hosted the award ceremony for Germany’s national mathematics competition. The pairing turned out to make even more sense than I had anticipated.
Earlier this month, Versicherungskammer hosted the award ceremony for the winners of Germany’s national mathematics competition, the Bundeswettbewerb Mathematik 2025. Around 100 guests filled our atrium in Munich: representatives from the Federal Ministry of Education and Family Affairs, the Bavarian State Ministry of Education, the President of the German Mathematical Society, the Max Planck Institute for Mathematics, private sponsors, math educators – and, at the centre of it all, sixteen of the seventeen young national winners.
Bringing the ceremony to our headquarters was an idea I had proposed internally, and it found immediate and full support. The pairing turned out to make even more sense than I had anticipated.
What it takes to win
For readers outside Germany, a brief note on what the Bundeswettbewerb Mathematik actually is. It is not a single test on a single afternoon. It runs over nearly two years, in three rounds. The first two demand written, fully proven solutions to problems that go well beyond the school curriculum – the kind of problems where the answer matters less than the elegance of the argument. The final round is not an exam at all but a colloquium: a one-on-one mathematical conversation with a panel of professors and senior mathematicians, in which candidates defend their thinking, respond to probing questions, and demonstrate genuine mathematical maturity.
Those who reach the top are automatically admitted to Studienstiftung des deutschen Volkes, Germany’s most selective academic foundation. Some of the winners in our atrium were claiming the title for the second, third, fourth – in one case, the fifth consecutive time.
These are not children who memorised tricks. They are young people who have learned to think.
A surprising operation
The festive lecture was given by Prof. Jürgen Richter-Gebert (TU Munich, Geometry and Visualization). He started with a definition that any maths teacher would mark wrong in red ink:
From this seemingly naive operation, he unfolded the world of Farey sequences – with applications stretching from the visibility lines in orchards to the theory of musical intervals. Circles and spirals swirled across the screen, made not only visible but audible.
What we witnessed was mathematics in its most beautiful form: from something simple grows something deep.
The mathematical lineage of insurance
Insurance is one of the oldest applied probability professions in the world. We price uncertainty. We model tail risk. We turn the abstract – a fire that may or may not happen, a life that will end at an unknown date – into something contractually tractable. Long before “data-driven” became a slogan, insurance was already that.
Mathematical excellence is not a guest in our house. It is the lineage.
The cognitive foundation of AI governance
The more contemporary reason is the one I keep returning to in my own work.
Much of the conversation about AI governance treats it as a regulatory and organisational discipline. The EU AI Act, FiDA, internal policies, risk classifications, registries. All of this is necessary, and I have spent considerable time building exactly these structures.
But underneath every governance framework sits something more fundamental: the ability to think precisely about uncertainty.
Can a team understand what a model is actually estimating? Can they tell the difference between correlation and causation in a high-dimensional setting? Do they know what a confidence interval means, and what it doesn’t? Can they recognise when a system is extrapolating beyond its training distribution?
These are not legal questions. They are mathematical questions. And no compliance checklist can substitute for the people who can answer them.
The young winners in our atrium will, in ten years, be among the people who build, audit, regulate, or govern AI systems. Some will be at insurance companies. Some at regulators. Some at the labs themselves. What they share is the one thing AI governance cannot manufacture from scratch: the trained habit of asking precise questions about precisely defined objects.
The colloquium format of the final round is, in this sense, exactly the right preparation. It does not ask whether you can produce an answer. It asks whether you can defend it.
The right questions
In a time when AI delivers answers ever faster, we need – more than ever – people who can ask the right questions. That is what the Bundeswettbewerb Mathematik has been quietly cultivating for decades.
An insurance company, a mathematics prize, and a conversation about AI governance belong in the same room more naturally than most people assume. We just rarely arrange the chairs that way.