Marina D'Amico
Affiliation: University of Milan
Category: Philosophy
Keywords: Causation, Causal reasoning, Norms
Date: Friday 5th of September
Time: 14:30
Location: GSSR Plenary Hall (268)
View the full session: Causal & Theoretical Reasoning
The notion of causality pervades human discourse, in both scientific and everyday talk. Yet much work remains to be done to fully understand the role (or roles) that this concept plays in our reasoning practices. Among the features of ordinary causal reasoning that need to be clarified is causal selection—the habit of identifying one (or a few) of several causal factors contributing to an event as "the" cause, while dismissing the others as background conditions. In recent years, empirical studies have shed new light on this issue by showing that causal judgments are systematically sensitive to violations of norms, in particular statistical and prescriptive norms. Two tendencies have been observed, depending on the causal structure. When two causal factors are jointly necessary for an outcome, an agent who violates the norm tends to be judged as more causal. Conversely, when causal factors are individually sufficient for the outcome, norm violations lead to judgments of lower causality (e.g., Knobe and Fraser, 2008; Icard, et. al., 2017; Morris, Phillips, Gerstenberg and Cushman, 2019). While various models have been developed to explain the evidence and capture the cognitive processes underlying our causal judgments (Icard et al., 2017; Morris et al., 2018; Quillien, 2020), less effort has been directed at explaining why events that are symmetrically related to the outcome lead to different causal attributions and why norms influence them. What, if any, is the purpose of causal selection that norms serve? I aim to provide a functionalist account of causal selection that answers this question. To clarify why it is worth considering an alternative account, I first review how and to what extent the two main approaches in the literature explain the evidence. The first proposes that causal judgments are aimed at assigning responsibility or blame (e.g., Alicke, Rose, and Bloom 2011; Sytsma, 2021), while the second proposes that causal judgments point to factors that are appropriate targets for intervention (Hitchcock and Knobe, 2008). The latter idea, that people's causal attributions assign greater relevance to those factors that allow us to intervene effectively in the world—and thus better support the purpose of manipulation and control that causal reasoning serves (Woodward, 2003; 2021)—also underlies my hypothesis. Specifically, I will suggest that the specific goal of causal selection is to identify the factor that, when manipulated, allows us to achieve optimal control over what we expect to happen. In other words, we attribute more causality to the factor on which intervening makes more difference to the expectation that an outcome will be achieved or avoided.
This idea is more precisely expressed in terms of a measure of control effectiveness using the language of structural equations (Pearl, 2009). Given a causal model M = <V, E>, let <C_1 = c_1, C_2 = c_2, ..., C_n = c_n> ∈ V \ E be the variables upon which to intervene at their actual values, and let E = e be the actual outcome. If Prob(∙) represents our expectations, then the effectiveness of a variable C_i in controlling Prob(E = e) can be expressed as:
CE(C_i, E, ci, e) = Prob(E(C_i = ci) = e) - Prob(E(C_i ≠ c_i) = e)
The variable to which we attribute the most causality is the one that maximizes this value:
arg max (C_i ∈ V \ E) CE(C_i, E, c_i, e)
Within this hypothesis, the influence of different types of norms is thus reduced to a single role, that of shaping expectations and providing the relevant contextual information that serves the function for which the concept of causality is designed: enabling the most effective interventions. Support for this proposal is provided by showing that it can explain those effects of norms in two-variable scenarios that are more difficult to explain by the alternative accounts. Furthermore, it can also account for some causal judgments that have recently been observed in scenarios involving more than two variables (Quillien and Lucas, 2023).