Philosophy and Causation

Catalog description

Philosophy, Causal Discovery and Causal Representation learning, Machine Learning, and Statistics all consider many of the same topics from different viewpoints: e.g., questions regarding counterfactuals, fairness, simplicity, etc. The goal of this course is (i) to investigate how these different viewpoints are related to each other and may leverage each other to inspire unified solutions, (ii) to investigate whether and how these different viewpoints suggest questions and directions of research in these areas and beyond, and (iii) to provide some history of the philosophical viewpoints on these issues, which may help us understand the future. Some examples of the type of questions to be considered include: What do counterfactuals mean, what are their truth conditions, how can they be reliably inferred, and how are they naturally related to modern machine learning problems such as domain adaptation? Why and how can causality, including hidden causal variables and causal relations, be learned from observational data? How can causality inform game theory? What makes a good model "simple" and what is the justification for preferring simple models? What makes a model, or a prediction "fair" and how can we tell when a model is fair? To what extent could a computer program be able to replicate what human scientists do? What abilities do they currently lack? How should AI be designed and regulated in order to promote human welfare?