Responsibility and Blame

Up to this point in the book, causality has been treated as an all-or-nothing concept; either A is a cause of B or it is not. In this chapter, the definition is extended to take into account the degree of responsibility of A for B. For example, if someone wins an election 11--0, then each person who votes for him is less responsible for the victory than if he had won 6--5. A notion of degree of blame, which takes into account an agent's epistemic state, is then defined. Roughly speaking, the degree of blame of A for B is the expected degree of responsibility of A for B, taken over the epistemic state of an agent. The relationship between the notions of responsibility and blame and the notion of graded causation defined in Chapter 3 using the notion of normality is then examined.

Complexity and Axiomatization

Causal models are potentially quite large and difficult to describe, especially once they include information about normality and typicality. In this chapter it is shown that it is often possible to achieve a compact representation of causal models without losing any information. In addition, a complete axiomatization of a language for reasoning about causality is provided.

Graded Causation and Normality

Recent work in psychology and experimental philosophy has shown that judgments of actual causation are often influenced by consideration of defaults, typicality, and normality. This chapter shows the definition of causality introduced in Chapter 2 can be extended to defaults, typicality, and normality into account. The resulting framework takes actual causation to be both graded and comparative. Thus, it allows us to say that one cause is better than another. Examples showing the power of the approach are considered.

Introduction and Overview

The basic intuitions of causality defined in terms of counterfactuals are introduced, and a brief overview of the key ideas of the book – including responsibility, blame, normality, and explanation -- is provided.

Explanation

The structural-equations framework is used to provide a definition of causal explanation. Essentially, an explanation is a fact that is not known for certain but, if found to be true, would constitute an actual cause of the fact to be explained, regardless of the agent's initial uncertainty. This definition is shown to handle well a number of problematic examples from the literature. It can also be extended in a natural way to provide several notions of partial explanation and explanatory power.

Applying the Definitions

The book concludes with a summary of the key points made, and a discussion of three applications areas from computer science: causality in databases, causality in program verification, and the use of causality in accountability (e.g., determining the cause of a major data breach).

The Art of Causal Modeling

According to the definition of causality considered in the previous two chapters, whether A is a cause of B depends on the model used. A can be the cause of B in one model and not another. This chapter considers how a “good” model should be constructed. Factors considered include the choice of variables and their values, and the normality ordering chosen. Of particular interest is the extent to which the question of whether A is a cause of B remains stable as more variables are added to the model in a conservative way, that is, without changing the relationships between variables already in the model.

The HP Definition of Causality

Three variants of a definition of actual causality are introduced. These definition uses structural equations to model counterfactuals. The definition is shown to yield a plausible and elegant account of causation that handles well examples that have caused problems for other definitions. Although transitivity is not transitive according to this definition, conditions sufficient to guarantee transitivity of causality are provided. Although the definition given assumes that everything is known, it is shown to easily extend to a situation where there is uncertainty modeled using probability. A notion of sufficient causality is also considered, as well as causality in nonrecursive models, where there are circular dependencies