November 2026
Abstract: AI systems can now read and write mathematical arguments, but long proofs remain hard to check, for humans and machines alike. After a short introduction to how mathematical texts are built from definitions, statements and proofs, I will present a setup built from freely available software that works together with a sufficiently capable large language model (LLM), currently in practice a commercial frontier model, to handle larger mathematical texts written in natural language. An AI agent breaks a mathematical document into a network of linked notes, one per definition, statement or proof. Other AI reviewers then check each proof using only what it cites. When a statement changes, the setup shows exactly which proofs need checking again. This is a personal engineering experiment in building a tool for my own work. It draws on many similar ideas from others and is not meant as an original research contribution. I would welcome feedback and discussion.
Slides: available here.
Abstract: Within a few years, generative AI has made the production of text, images, sound and video seamless and immediate. This lecture argues that this shift constitutes a new media condition, hyperreproduction, in which cultural artefacts are generated as endlessly variable instantiations from learned representational spaces: copies without originals, drawn from a substrate of absorbed material.
Driven by platform capitalism, it produces content at scales and velocities that outpace the perceptual and evaluative horizons of any individual, institution, or archive. Its most visible effect is the latent flood: hyperreal synthetic material saturating feeds and catalogues, with platforms like Suno already generating more music every two weeks than Spotify's entire catalogue. Read through critical theory and media theory from Adorno to Kittler, latent space appears as the medium of this regime, and the platforms that own it as its new culture industry.
The lecture identifies and theorises a constellation of interlocking conditions that constitute hyperreproduction: latent floods, semiotic deflation, the liquefaction of the cultural object, infrastructural capture, recursive loops, latent enclosures, the autosimulacral turn, contextual voids, Dead Internet dynamics, and the post-authentic condition, arguing that hyperreproduction operates simultaneously across individual, platform, cultural, and ecosystemic levels as both a condition and a structurally accelerating tendency of contemporary media ecologies. The lecture maps the current and near-future media landscape and seeks to make the transformations now under way more visible and open to collective reflection.
Extended abstract available here.
Slides: available here.
Abstract: AI systems now do mathematics: quickly, forcefully, sometimes brilliantly, and sometimes still very stupidly. By the standards we have long set ourselves, with prizes for solving the great problems, this must count as progress of a sort. Yet the same tools may change mathematics beyond recognition. Is this the end of the beginning, or the beginning of the end? And can we, or should we, stop?
I will speak as a working mathematician who is torn. I am wary of revolutions and of a discipline shaped by large private companies, and worried about ecological cost, unequal access, attribution, and results arriving faster than human understanding. But I cannot simply dismiss a tool of real scientific value.
I will argue that the answer depends less on the machines than on what we decide mathematics is for, and what it is about. The ethical questions depend heavily on the task, and AI use should be disclosed. Above all, the heart of mathematics, thinking and explaining together, is ours to keep unless we give it up.
Slides: available here.
October 2026
Peter Scholze shared his critical perspective on the rapid development of AI and its implications for mathematics and the mathematical community in an informal conversation. Here are the questions he was asked by Sylvie Paycha.
Abstract: As a woman mathematician witnessing the rapid development of AI and its growing impact on society, I wish to draw attention to some important questions and challenges it raises. I speak not as an AI expert, but as an engaged observer seeking to encourage discussion.
AI’s potential to reshape society makes deciding its future a shared responsibility, not a privilege reserved for a few. Women and other underrepresented groups must have a meaningful voice—not only to ensure fair representation, but also to bring diverse perspectives, challenge assumptions, identify overlooked risks, and help safeguard the public interest.
As European Women in Mathematics (EWM), we have contributed to this discussion through our declaration on AI and the mathematics profession. My aim is not to offer definitive answers, but to raise questions and open a discussion about who shapes AI’s future and whose voices are heard.
Slides: available here.
Abstract: Frank Quinn described an invisible revolution in twentieth-century mathematics: a profound change in our understanding of proof and rigour, which gave rise to modern logic, math and, eventually, theoretical computer science. Today we are living through another transformation. Logic helped create computation, but the current revolution in artificial intelligence has come largely from statistics, neural networks, data and massive computational power. Now logic is needed again: AI systems can produce remarkably convincing mathematical arguments, but convincing is not the same as correct.
This makes formal mathematics newly important. The better machines become at generating mathematics, the more we need reliable ways of checking what they produce. But I do not want the consequence to be that mathematicians must learn to communicate like proof assistants. I will discuss attempts to combine the precision of formal mathematics with the flexibility and economy of ordinary mathematical language, using AI itself to help bridge the two.
Finally, the mathematics we are building is not only technical infrastructure. It depends on open publications, libraries and communities, and on decisions about whose work becomes visible and receives credit. Drawing on my experience with Women in Logic and Women in Formal Mathematics, I will argue that the challenge posed by AI is therefore not simply how machines will do mathematics, but what kind of mathematics—and what kind of mathematical community—we want to build.
Slides: available here.
Recording: available here.