Rafal Urbaniak

About

I work on causal explanation and probabilistic reasoning for decisions that matter — in courtrooms, real estate, public policy, and ecology. The thread tying it together is that the things we want to know about — responsibility, fairness, how a system would respond if we changed it — are causal and counterfactual, while the data we have are statistical and partial, and I’m interested in bridging that gap explicitly rather than smuggling it in. Much of the work is spatio-temporal: it lives on trajectories and dynamics rather than on i.i.d. observations.

My main current project is PCI (Probabilistic Causal Impact). The toolset for causal explanation is split: actual-causality frameworks in the Halpern–Pearl tradition give principled verdicts but don’t scale, while methods that do scale — SHAP, LIME, gradient attribution — aren’t causally aware. PCI re-conceptualizes explanation as an estimation problem on an expanded probabilistic structural causal model, with explicit user-facing knobs (a variable-selection distribution, an alternative-value distribution, and a causal-impact function), and yields Monte Carlo-friendly causal attributions that line up formally with Halpern’s actual-causality verdicts. We test it on canonical scaled-up actual-causation problems and on a production spatio-temporal Bayesian valuation model for real estate with deep machine-learned components.

Alongside PCI I work on legal probabilism — forensic evidence evaluation and case narrations, and lately on imprecise probability, awareness growth and unanticipated possibilities, and gaps in evidence — and on spatio-temporal models of group foraging in animals, Bayesian approaches to bias estimation in word embeddings, and the design of counter-speech interventions against online aggression.

I did my PhD in Logic and Philosophy of Mathematics at the University of Calgary, and have since held positions at the Research Foundation Flanders, Trinity College Dublin, the University of Bristol, and the University of Gdansk.