From Statistical to Causal Learning

Abstraction: Survey of AI learning paradigms from symbolic to causal interventional models

Key points:

• Paper by Bernhard Scholkopf and Julius von Kugelgen (arXiv:2204.00607, 2022) traces AI approaches from symbolic methods through statistical learning to interventional causal models.
• Hard open problems in ML/AI — including distribution shift and out-of-distribution generalization — are intrinsically linked to causality.
• Progress requires advances in how to model and infer causality from observational data.
• Statistical learning alone is insufficient for tasks requiring reasoning about interventions or counterfactuals.

Connections: Bernhard Scholkopf · Causal Inference · Statistical Learning · Causal Reasoning

Source: https://arxiv.org/abs/2204.00607