Philosophy of Physics Graduate Lunch Seminar

Abstract: I develop a naturalised theory of causation, building a bridge between recent literature on the foundations of statistical mechanics—in particular the literature connected with Albert and Loewer’s Mentaculus (Albert, 2000; Loewer, 2020)—and the framework of causal Bayesian networks. To do so, I first recap how the Mentaculus implies that the probabilistic macro-trajectory of the universe is approximately (though not perfectly!) branching in structure. I then sketch a proof of the following:

Theorem. Insofar as the macro-trajectory of the universe instantiates the branching structure of the Mentaculus (Loewer, 2020), the relationships among sets of macro-variables admit a unique causal representation, oriented with respect to the arrow of time, with the implied causal effects given (roughly) by the Loewer semantics for counterfactuals (Loewer, New Probabilistic Account of Counterfactuals, 2023).

Following the framework of causal Bayesian networks, a “causal representation” is here understood to be a directed acyclic graph over the variables such that (i) causal effects can be computed using the do-calculus (Pearl, Glymour, & Jewell, 2016), and (ii) the joint probability distribution factorises in the right way determined by the graph – or equivalently, the causal Markov condition is satisfied (Hausman & Woodward, 1999).

I discuss two technical subtleties with the proof and two substantive philosophical objections to this whole line of argument. Time permitting, I compare my approach to David Papineau’s (2025) new regularity theory, concluding that, despite the apparent reversal in the order of analysis, the two approaches bear many striking similarities. 

_{Philosophy of Physics Graduate Lunch Seminar Convenors: Paolo Faglia, Gregor Gajic and Rachel Pedersen}