Nov 23. 2020
The Limits of Causal Graphs: Density
<p>Speaker: Dean McHugh (University of Amsterdam), Time: Monday November 23, 16:30, Prague time. Place: Zoom (contact Radek Honzik for details).</p>
Causal reasoning is often modelled using causal graphs, such as Bayesian networks and structural causal models (e.g. Pearl 2000, Causality). Given the success of graphical models of causation, one may ask whether they can represent every instance of causal reasoning, or whether a more expressive framework is required. In this talk we show that graphical causal models are limited in their expressive power: some intuitive causal structures are impossible to represent using graphical causal models. Specifically, graphical causal models cannot represent dense causal chains; that is, chains of events where between any two events C and E on the chain there is a third event D on the chain such that C causally influences D and D causally influences E. Dense causal chains appear, arguably, in both our intuitive representation of the world and models in physics that assume spacetime is dense. Since any framework representing causal reasoning should be compatible with these models, a more expressive framework is required—one that can represent dense causal chains.