A way to anticipate emergence in complex human systems — for people who must decide before the evidence is in, so that weak signals become action.
Technological forecasting, cybersecurity economics, and social-interaction science — research and translational work, built for applied decision-making.
Six research perspectives, each sized by current intensity, all attacking one fundamental problem from a different angle — anticipating emergence in complex human systems. Every one of them is the same sentence, finished differently. Hover a perspective to complete it.
Research, talks, teaching, and the hackathons and programmes I run. Open one for the story behind it — recurring subjects share a single explanation.
The first hackathon of the UNIGE Faculty of Medicine turns digital phenotyping on mental health into 32 hours of building. Around fifty master's and doctoral students, clinicians, researchers and engineers work in interdisciplinary teams on six challenges submitted by researchers themselves, so the problems are the ones the field actually has. It is run with Campus Neuro, the Synapsy research centre, the Swiss Centre for Affective Sciences, the Wyss Center, the Geneva University Hospitals and EPFL, at Campus Biotech. The aim is deliberately concrete: teams leave not with slideware but with project sketches solid enough to be funded, bridging clinic, laboratory and engineering in one room.
A build course rather than a lecture: Vibe-Coding Social Robots, run with Eva Wiese's cognitive ergonomics group at TU Berlin and taught with Matt Huebert. It opens with a needs library seeded from the lab's own research problems — validating and sharpening those needs is also how students earn one of the credit-bearing seats. Then five teaching sessions on AI-assisted building and on AI stewardship, delegating, verifying, overriding and refusing; an on-site hackathon at the MAR campus; and three supervised weeks hardening the result. Every team ships a runnable artefact the group can actually pick up — a stimulus generator, a gaze-cueing harness, an evaluation probe for LLM participants.
A keynote in Kunming on what education owes students who will spend their working lives inside the SDG problem space. The argument: the SDGs are three problems at once — an innovation challenge, a coordination challenge, and a measurement challenge — and none of the three is taught by lecturing at people. Learning by doing is the honest response, which is what ten years of SDG Open Hack across Geneva, Tsinghua, Singapore and Yunnan have been building. AI sharpens the point rather than settling it: used alone it isolates and produces fast wrong answers; used collectively, with peers, review and shared evidence, it lets small teams analyse, prototype and localise faster than ever.
A talk inside a week-long build at ETH Zürich, organised by Dirk Helbing's group: two mornings of short papers — vTaiwan and participatory budgeting in Winterthur, adaptive preference elicitation, deliberation, voting as pluralistic optimisation, open government data, generative AI — then teams form and spend three supervised days making something, closing with presentations and prizes on the Friday. My slot falls on the Tuesday afternoon, immediately before the building starts, which is the right moment for the question: what does the evidence actually say about vibecoding, and what does AI-assisted building change about what a hackathon can be expected to produce in five days?
Competitive research funding across ~15 years. Bar position & length = project period; bar thickness = amount; colour = my role.
A revisited Scholar timeline — how citation intensity builds across years and shifts between fields. Darker = more intense; hover a cell for detail.