Full Program »

Establishing Constitutive Relevance in Mechanisms

How do we identify the components of a mechanism? On Craver’s (2007) mutual manipulability account of constitutive relevance, X counts as an actual component of a mechanism S if X is a part of S, an ideal intervention on X changes S, and an ideal intervention on S changes X. The account faces a dilemma (see Romero 2016; Baumgartner and Gebharter 2015; Leuridan 2012). For any actual component X, if X and S meet the criteria of MM, then we must say X and S are thereby causally related, since an ideal intervention on X with respect to S (or vice versa) would entail, by Woodward’s (2003) interventionism, that X is causally relevant to S (or vice versa). This is undesirable, as mechanists commonly reject that mechanisms and their components causally interact, for their constitutive relationship makes interlevel causation problematic. But if, on the other hand, it is impossible for X and S to meet the criteria, then we cannot rule X a component of S. It may indeed be the case that ideal interventions between X and S are altogether impossible to perform, since any intervention on one is simultaneously a direct intervention on the other, violating one of the criteria of ideal interventions. Thus, no actual component could be counted as a component by mutual manipulability, making it an inadequate account of constitutive relevance. The extant attempts to solve these problems are unsatisfying, for they leave one of the horns of the dilemma unresolved (e.g., Romero; Harinen 2014), or they make for an account of constitutive relevance that threatens to be descriptively inadequate (e.g., Baumgartner and Gebharter). I offer a simple reconstruction of mutual manipulability that resolves both horns of the dilemma in one straightforward move: instead of interpreting interlevel experiments as involving interventions between the variables X and S, we ought, following Harinen, to interpret top-down experiments as involving an intervention between S’s input and X, and bottom-up experiments as involving an intervention between X and S’s output. The experiments themselves favor neither interpretation, but only this three-variable interpretation prevents the dilemma. Interventions between S’s input and X, and X and S’s output, can indeed be ideal, since these variables are not constitutively related (so an intervention on one is not a direct intervention on the other). Furthermore, the causal entailments are not problematic, for the variables are indeed causally related. Showing that X, a part of S, lies on the regular causal chain between S’s input and output would establish that X is a component of S, since S consists in this regular causal chain. If, therefore, we switch the criteria of mutual manipulability to demand ideal interventions between S’s input and X, and X and S’s output, instead of between S simpliciter and X and vice versa, we resolve the dilemma entirely. This leaves us with an account of constitutive relevance that is satisfying theoretically and fruitful empirically.Author Information:

Emily Prychitko

Washington University in St. Louis