Causality and Counterfactuals in the Situation Calculus

Summary

This partial ingest is based on the extracted full PDF text. Hopkins and Pearl argue that standard structural causal models are too inexpressive for some actual-causality problems involving actions, objects, time, and event/non-event distinctions. They propose situation calculus causal models (SCCMs), embedding structural-causal intuitions into Situation Calculus by pairing a situation-calculus specification with a potential situation, then evaluating counterfactuals through a natural executable subsituation.

Key Claims

• Structural causal models handle many counterfactual queries well, but their variable-assignment ontology can blur distinctions between transitional events, enduring conditions, actions, inactions, and object-indexed causal pathways.
• Situation Calculus supplies an action/history language expressive enough to represent dynamic causal scenarios with objects and temporal order.
• SCCMs map structural-model ideas into situation calculus: exogenous variables correspond to initial database axioms, endogenous enduring conditions to fluents, transitional events to actions, and mechanisms to successor-state and action-precondition axioms.
• Counterfactuals can be represented by removing actions from, or forcing actions into, a potential situation before computing the natural executable subsituation.
• The paper does not give a final definition of actual cause; it provides a framework intended to support richer definitions than propositional structural models can express.

Methods / Formalism

• A situation S=[a_1,...,a_n] is treated as an action sequence. For a set of ground actions A, S-A is the sequence obtained by omitting actions in A.
• A potential situation is a pair (S,F), where S is not necessarily executable and F:{a | a in S}->{0,1} flags actions whose preconditions should be ignored.
• A situation calculus causal model is

$$M=(D,P),$$

where D is a situation-calculus specification and P is a potential situation.

• Counterfactual model updates have the form [not a_1,...,not a_m,b_1,...,b_n]M, removing some actions and forcing others.
• The natural executable subsituation NES(M) is built by scanning the potential situation in order and appending action a_i when either Poss(a_i,NES(M)) holds or F(a_i)=1.
• Situation Calculus Causal Models records this SCCM construction and the counterfactual language.

Evidence / Experiments

• The paper is formal and example-driven rather than experimental.
• Running examples include firing squad, candle extinction, and bystander/assassin scenarios.
• The examples are used to show why distinguishing actions from fluents and occurrences from non-occurrences matters for actual-causality judgments.
• The framework also supports probabilistic SCCMs by putting a probability distribution over complete initial database specifications.

Connections

• Seeds Actual Causality in the wiki as a bridge between causal explanation and action histories.
• Connects Structural Equation Models to Situation Calculus by translating structural mechanisms into precondition and successor-state axioms.
• Complements Liu2026 - A Counterfactual Cause in Situation Calculus, which later gives a counterfactual achievement-cause definition in a modal situation-calculus variant.
• Extends the existing Situation Calculus cluster beyond executability toward counterfactual causal explanation.

Open Questions

• The paper leaves the actual-cause definition itself for future work.
• How should SCCMs choose the potential situation when observations are partial, concurrent, or probabilistic?
• Which parts of SCCM causal judgment should be handled by model structure, by the potential situation, and by the actual-cause definition?

Citation

Hopkins, M., and Pearl, J. (2005). Causality and Counterfactuals in the Situation Calculus. In Proceedings of the 7th International Symposium on Logical Formalizations of Commonsense Reasoning, 115-122.