Since September, the world of mathematics has been in turmoil: AI agents from OpenAI are said to have solved one of the seven so-called Millennium Problems, among the hardest open questions in mathematics. What people did not manage in roughly a hundred years, AI, according to the company, achieved within 88 hours.

Exciting as this news was for the wider public, it has little immediate bearing on people's everyday lives. For mathematicians it is a very different matter: they see their work to date, their professional identity and their sense of purpose called into question.

That makes a closer look worthwhile. In mathematics we can currently watch an entire field lose authority over its own work, and watch the community affected struggle to win it back. It is grappling with its own purpose, its understanding of performance, its culture and its worth, and asking what changes follow from this for its organisation and its structures.

Similar questions are likely to arise soon in many organisations. How mathematics deals with this shift therefore offers us a chance to think early about how we want to shape our own future with AI.

A profession wrestles with its contribution

Mathematics has long been regarded as a pillar of intellectual thought and forms the basis for many sciences. In recent decades, however, that understanding has increasingly receded into the background. Mathematicians are often judged by their publications, and in particular by their ability to prove theorems and find solutions to mathematical problems.

It is hardly surprising, then, that OpenAI's release of more than 700 mathematical papers on 6 October is perceived as a serious threat. That holds even though it is not yet clear how many of the proofs it contains will withstand mathematical scrutiny. The release followed a series of announcements over the course of the year. These included the refutation of a conjecture put forward eighty years ago by Paul Erdős, as well as the purported solution of the Navier-Stokes problem in September, one of the seven Millennium Problems.

For researchers who have devoted years or decades to particular problems, this can be an existential experience. Beyond recognition for their work so far, their future research questions, career paths and professional prospects are also up for debate. If AI delivers the very achievements they have defined themselves by, what then is their own contribution in future?

The danger of mistaking performance for purpose

The mathematician and theoretical computer scientist Ankur Moitra of MIT, the engineer and fluid-dynamics researcher Steven Brunton of the University of Washington, and the number theorist Alex Kontorovich of Rutgers University take up this question in a more than three-hour episode of Lex Fridman's podcast. Their reflections carry over to other fields of work as well. We often judge the value of what people and organisations do by the visible, measurable results of their work. In doing so, we risk mistaking those results for the actual purpose of the activity. When AI can produce the same results faster or more efficiently, it is not only the previous way of delivering them that is called into question, but also the self-understanding bound up with it.

Ankur Moitra illustrates this with the so-called Doorman Fallacy. Anyone who sees a doorman's work solely in opening the door might replace him with an automatic one. That would overlook everything else he contributes: he welcomes guests, recognises regulars, conveys a sense of safety and shapes the experience of arriving.

Moitra applies exactly this thought to mathematics. The most visible achievement of a mathematician is to prove theorems. But that, he argues, is not the core of the work. Mathematicians are people who think deeply, develop new theories, make connections and open up new perspectives. He therefore defines his professional identity not through producing proofs, but through thinking itself. Alex Kontorovich adds to this with the question of whether mathematics has not mistaken a yardstick for scientific excellence for its actual purpose.

Many mathematicians argue in the same direction in their own contributions. They recognise that AI's involvement in mathematical research calls for a fundamental redefinition of their role and of the purpose of their work. And many see that purpose in enabling humanity to understand the world more deeply and in making mathematical insight usable.

What if humans cannot understand the AI?

Stephen Wolfram, a physicist and computer scientist whose research into complex systems also underpins our own theory of the Digital Society, approaches the question from a perspective characteristic of him: he argues that humans can grasp only a limited section of the entire space of mathematical possibility. Our knowledge, our experience and our capacity to process information determine which mathematical relationships we recognise and which concepts we are able to develop.

With AI, that section could expand considerably. Language models can draw on an enormous body of mathematical theories, proofs and concepts already developed, and combine them in ways that would be barely possible for any individual. At the same time, computational methods open up the possibility of systematically investigating formal rule systems and discovering mathematical structures that have so far lain outside our attention.

But this could also produce results that are alien to us and hard to follow: mathematical proofs that are correct, yet whose reasoning we can barely understand. What use would such solutions then be to us? Mathematics does not consist only in finding correct answers. It also develops concepts and abstractions with which we recognise connections and describe the world.

For Wolfram, what must take centre stage in future is how AI can help us answer relevant questions and, in doing so, gain insights that are accessible to mathematicians and meaningful to humanity.

«Mathematics has changed. Let's act like it.»

The mathematical community must now reach an understanding of what the purpose of its work is and how it wants to work with AI in future. If what it regards as its contribution changes, it must also rethink how it realises that contribution. The various statements and demands of recent weeks show that this is already happening. The mathematics professor Frank Calegari puts it in a nutshell in the title of his article: Mathematics has changed. Let's act like it.

In this spirit, mathematicians from various countries are already putting forward concrete proposals for a realignment. Three areas illustrate how far their thinking reaches and what consequences they draw from a changed self-understanding for scientific practice:

Culture: Mathematics is marked by intellectual independence, specialisation and competition for new insight. As AI increasingly develops proofs, the question arises of how collaboration, exchange across disciplinary boundaries and shared thinking can be given greater weight. At the same time, it needs to be clarified how achievements are attributed and how an open research culture can be preserved.

Structures: If finding relevant questions and meaningful approaches gains importance over publishing proofs as the measure of achievement, that has consequences for universities, research institutions and publishers. How must education, career paths and incentive systems change? By what criteria is scientific performance assessed? And which structures enable the cross-disciplinary exchange being called for?

Communication: The way mathematical knowledge is shared is also up for discussion. Papers and seminars should do more to explain why an insight matters, which connections it makes visible and which new questions arise from it. This concerns exchange within the mathematical community as much as conveying its insights to other sciences and to society.

In our view, this approach is exemplary. It consistently takes the actual purpose of the activity as its starting point and derives from it how human and artificial intelligence can contribute to realising that purpose. Politics, business and other fields of science would do well to take up this way of thinking too. For AI expands the space of what we can recognise, understand and shape. Which of these possibilities we pursue, and what future we want to create with them, we must determine together as a society.

New leadership tasks for executive teams and boards

The example of mathematics shows how closely a profession's self-understanding can be tied to particular achievements. When AI can suddenly deliver them faster or better, that self-understanding is thrown off balance. The actual purpose, however, need not be lost in the process. Rather, it becomes visible where the purpose of an activity had been equated with the previous form of delivering it.

For organisations, this means thinking beyond automating today's processes and handling existing tasks more efficiently. They have to reach a shared understanding of the contribution they want to make for their customers, their employees and society.

This examination concerns an organisation's purpose and self-understanding as much as its strategy, its culture, its responsibility, its structures and its relationships with its environment. Executive teams and boards thus face the task of also questioning the organisational models themselves into which they want to integrate AI.

Four leadership tasks follow from this:

Examine your own understanding of performance. Which results count today as evidence of the organisation's value? And to what extent do they actually capture the contribution it wants to make? This distinction becomes decisive once AI can take over existing tasks.

Recognise the foundations of your own value creation. What particular knowledge, know-how and interplay of capabilities creates value for employees, customers, other stakeholders and society? This is what you can build on to deploy AI deliberately and to strengthen the resilience of the organisation and its members.

Create orientation and shape change together. When tasks, roles and established measures of performance are in question, people need an understanding of where their organisation wants to develop and what contribution they can make to it. Leadership and communication have to enable this shared understanding and to involve other stakeholders as well.

Rethink structures and business models. AI can fundamentally change the conditions under which an organisation operates, and do so in a short time. It is therefore worth examining which forms of organisation and collaboration will be needed in future to realise its purpose.

The central leadership task is to think from the organisation's actual purpose and to unlock the possibilities of AI in its service. This includes the potential to combine familiar concepts and approaches in new ways, and thereby to develop entirely new business models and value chains. How AI can make such connections, and what possibilities arise from them, is the focus of the second part of our Strategic Shifts series on what we can learn from the recent developments in mathematics. It will appear shortly.

Sources and further reading

Mathematical publications and context

Mathematics, AI and the future of the profession

Scientific autonomy, critique and research funding