News and commentary
Announcements, appointments, events, and writing for policy and public audiences from across the Centre and its affiliates.
-
Does worker retraining protect against automation?
As AI raises fresh concerns about job displacement, retraining is often proposed as the answer. In Did US Worker Retraining Reduce Participant Automation Exposure?, Jacobs and Canedy test that idea using over 23 million records from the Workforce Innovation and Opportunity Act, the main public funder of retraining in the United States. They find that participants rarely move into jobs less exposed to automation, and most return to their previous field. Where outcomes improve, it is mainly because wages recover, not because people change occupation. Employer-led apprenticeships do best.
Read the working paper → -
What can physical AI actually do on the factory floor?
After visiting a new deployment and IMTS in Chicago — the largest manufacturing technology show in North America — Professor Sentance asks what physical AI can really do in a factory. A century of process engineering has been about removing variation, he argues, so AI’s great strength — adapting to the unfamiliar — adds little to today’s linear production lines. The real opportunity lies in the adaptive, non-linear processes that manufacturers abandoned because they were too unpredictable to manage or too dangerous for people.
Read the post on LinkedIn → -
Finding causal relationships in incomplete data
In “PAIR-CI: Calibrated Conditional Independence Testing for Causal Discovery with Incomplete Data”, Robinson and Lall introduce a machine learning method for uncovering causal relationships in data with missing values — a pervasive problem in the surveys, firm records, and cross-national datasets that social scientists rely on. Where existing approaches can mistake gaps in the data for genuine causal links, the new method keeps such false discoveries rare. The paper will appear at NeurIPS 2026, the premier venue for research in artificial intelligence and machine learning.
Read the preprint on arXiv → -
Economic policy for AGI
In one of the inaugural essays of the newly launched DeepMind Institute — a platform started by researchers at Google and Google DeepMind to publish and debate ideas about a world with AGI — Jacobs and Imas assess eleven policies for managing the economic disruption AGI could bring. Rather than a single prescription, they propose “least-regret” responses tied to clear empirical triggers: expanded unemployment insurance, tax credits, and employer-led retraining for mild disruption; a negative income tax if displacement deepens; and universal basic capital as a backstop should labour and capital income decouple.
Read the essay → -
The chain of peace: do supply chain chokepoints deter war?
Advanced semiconductors cannot be made by any one country. Design software is American, lithography machines Dutch, optics German, photoresists Japanese, fabrication Taiwanese — and remove any node and the chain breaks. Pham and Scholten argue that this structure, not Taiwan's foundries alone, is what deters conflict in the Taiwan Strait, because a chip fab seized by force produces nothing without the network behind it. The uncomfortable implication is that export controls and reshoring, each sensible in isolation, are steadily dismantling the interdependence that keeps the peace.
Read at War on the Rocks →