wiki

Actual Causality

3 min read

Actual Causality

Definition

Actual causality asks which concrete event, action, condition, or sequence in a particular history caused a particular outcome. It is distinct from type causality, which asks whether one kind of variable or event generally causes another.

Why It Matters

Actual-cause judgments support explanation, responsibility assignment, debugging, legal reasoning, fault analysis, and counterfactual diagnosis. In AI and formal methods, they help ask why an agent, system, or computation produced a specific result.

Formalism / Key Objects

  • In structural causal models, actual-cause definitions usually rely on counterfactual interventions such as do(X=x') and ask whether changing candidate cause variables changes the effect under suitable contingencies.
  • In Situation Calculus, candidate causes can be action sequences in a narrative. A minimal counterfactual cause for goal phi satisfies:

with z minimal under a chosen action-sequence ordering.

  • For narrative achievement causes, Liu2026 - A Counterfactual Cause in Situation Calculus removes a prefix z' from the action history, filters the remaining sequence to executable actions, and checks whether the goal would fail.
  • In Temporal Causal Models, causes can be indexed by time: X at step t may be a cause of Y at step t'.
  • In Causal Games, counterfactuals can require pairing an actual rational outcome pi with a counterfactual rational outcome pi' after an intervention changes the game.

Connections

  • Structural Equation Models provide intervention semantics for counterfactual dependence.
  • Causal Game Intervention Semantics records how counterfactuals change when agents may rationally adapt to pre-policy interventions.
  • Situation Calculus provides an action-history language for actual causes involving objects, actions, and fluents.
  • Temporal Causal Models make actual-cause questions explicitly time-indexed.
  • Situation Calculus Causal Models is the focused annex for SCCM-style counterfactuals and narrative causes.
  • Dyadic Morality is adjacent because moral appraisal can ask whether perceived intentionality is necessary for outrage, but its evidential back-filling of agency is a psychological judgment process rather than an actual-cause definition.

Common Confusions

  • Actual causality is not the same as statistical association or average causal effect.
  • A but-for test is often too weak for preemption and overdetermination cases unless supplemented by contingencies or richer structure.
  • In strategic settings, “what would have happened” can depend on whether the alternative world keeps agents’ decision rules fixed or lets them recompute rational responses.
  • In action formalisms, an action not occurring and a fluent being false are not the same kind of object.

Key Sources

Graph View

Backlinks