When judging what caused an event, people do not treat all factors equally – for instance, they will say that a forest fire was caused by a lit match, and not mention the oxygen in the air which helped fuel the fire. We develop a computational model formalizing the idea that causal judgment is designed to identify “portable” causes - causes that are likely to generalize across a variety of background circumstances. Under minimal assumptions, the model is surprisingly simple: a factor is regarded as a cause of an outcome to the extent that it is, across counterfactual worlds, correlated with that outcome. The model explains why causal judgment is influenced by the normality of candidate causes, and outperforms other known computational models when tested against an existing fine-grained dataset of human graded causal judgments (Morris, A., Phillips, J., Gerstenberg, T., & Cushman, F. (2019). Quantitative causal selection patterns in token causation. PloS one, 14(8).).
- Causal selection
- Computational modeling