Alongside my empirical research, I work on how research gets done and how it gets used. Two threads run through it: the practice of doing rigorous empirical work now that AI is in the workflow, and the science-policy interface, where evidence from the social sciences has to be brokered into decisions within existing systems. This page lists some of the talks, essays and public writing that come out of that work.
TALKS
AI Is Not an Algorithm: What That Means for Research Practice, Machine Collaborators seminar series, 9 July 2026. [Slides (PDF)]
Generative AI is not deterministic, so supervising it is not code review. I set out a consistency-creativity dial for choosing how much latitude to give AI on a given task, work through failures at each end drawn from my own projects, and argue that the interesting territory is the middle: how to get AI's leverage without compiling away what makes the work yours. The organizing question is not whether to adopt or resist, but what you keep human, and why.
WRITING
The Compression Risk: What AI Synthesis Does to Contested Evidence, frontiers Policy Labs (with INGSA-Europe, Science Advice in Challenging Settings series), 4 September 2026. DOI: https://doi.org/10.25453/plabs.33437095Â
AI-assisted evidence synthesis creates risks of compression and correlation. Compression drops the context that explains what a finding means, and the cutting happens before an adviser ever sees the material. Correlation follows from advisers consulting the same few models, which narrows the range of their output and costs a team much of its capacity to catch errors without costing it members.
On AI and research work styles (in progress)
A longer essay on why most AI productivity advice is written for one kind of worker, and what changes when it is not.