Situation Calculus Causal Models

Context

The situation-calculus causality papers use action histories rather than flat variable assignments to reason about actual causes and counterfactuals. Hopkins2005 - Causality and Counterfactuals in the Situation Calculus introduces SCCMs; Liu2026 - A Counterfactual Cause in Situation Calculus gives counterfactual achievement causes in a modal situation-calculus setting.

Formal Statement

Hopkins-Pearl define a situation calculus causal model as

where D is a situation-calculus specification and P=(S,F) is a potential situation. S=[a_1,...,a_n] is a not-necessarily-executable action sequence and F:{a|a in S}->{0,1} flags actions that should be forced despite failed preconditions.

The natural executable subsituation is computed by:

whenever Poss(a_i,NES(M)) holds or F(a_i)=1; otherwise a_i is skipped.

Liu-Belle define a minimal counterfactual cause for a static goal phi by requiring

with z minimal under a chosen ordering.

For a narrative C=<Sigma,z,phi>, a prefix z' is an achievement cause when all extensions from z' through z keep phi true, the filtered counterfactual remainder Filter(z\z') makes not phi true, and no smaller subsequence satisfies those conditions.

Derivation / Construction

  • SCCM counterfactuals remove actions from the potential situation or force actions into it before recomputing NES.
  • Initial database axioms play the role of exogenous variables; fluents are enduring endogenous conditions; actions are transitional events.
  • Successor-state axioms describe fluent mechanisms, while precondition axioms describe action mechanisms.
  • Filtering in Liu-Belle removes impossible remnants of a narrative after the candidate causal prefix is deleted.
  • The two approaches share the same modeling instinct: counterfactual causal queries should respect action executability and the distinction between actions and fluents.

Implications

  • Situation calculus can express object-indexed and temporally ordered causal stories that are awkward in propositional structural models.
  • Counterfactual causality over actions requires care: deleting an action can make later actions impossible.
  • The formalism is useful for explanation and diagnosis in agent histories, but minimality choices affect the returned cause.