Presentation on theme: "1 Pluralistic Theories of Causation Francis Longworth University of Birmingham and Ohio University Saturday February 10th, 2007."— Presentation transcript:
1 Pluralistic Theories of Causation Francis Longworth University of Birmingham and Ohio University Saturday February 10th, 2007
2 Main Claims 1.Counterexamples: There are counterexamples to all extant univocal analyses of causation. 2.Disagreement: Individuals often disagree in their intuitive judgments about what causes what. 3.Vagueness: Causation is a vague concept. Borderline cases. 4.Degrees of Typicality: Some instances of causation more representative/typical than others. Our concept of causation has a prototype structure. 5.Polysemy: Causation is a polysemous concept: there is a plurality of related senses of causation. 6.Best explanation of 1-5: Causation is a cluster concept. 7.No univocal analysis of causation (that respects our intuitions) is possible. If one hopes to develop a precise, univocal analysis of causation in physics or biology, that analysis will have to be revisionary.
Causation as a Cluster Concept There are a number of factors/properties/criteria that we pay attention to when making our causal judgments, and whichcount towards whether or not C is a cause of E: E counterfactually depends on C Lawlike regularity between C and E E is manipulable via C C raises the probability of E Some (local) physical process links C and E, and… C is (morally) responsible for E -- empirically confirmed C explains E etc.
Causation as a Cluster Concept 1.The conjunction of all these criteria is sufficient (and prototypical). 2.No criterion is necessary. 3.Some subsets are sufficient (with some indeterminacy over which). - (3) follows from (1) and (2). - No individually necessary and jointly sufficient conditions (2). The cluster set is the union of minimal sufficient subsets: each property is a NUS property: a Non-redundant part of an Unnecessary but Sufficient condition. - Potentially highly pluralistic: The subsets correspond to different (but related) senses, and each is less prototypical than the full set Cluster Set.
Causation is a Moral Concept Automobile Accident: You are driving steadily down a deserted highway when suddenly, without warning, a truck ploughs into the side of your car. It is later revealed that the driver of the truck was heavily intoxicated and had run a red light. Was your driving steadily down the deserted highway a cause of the accident? Billiard ball analogue elicits opposite intuition. Objection: Moral facts supervene on causal facts…
Friendship as a Cluster Concept Thesis: The causal relation is analogous to the friendship relation. Do C and E stand in the causal relation? Do S1 and S2 stand in the friendship relation? The friendship relation is vague. Not in quantitative sorites sense Not clear how many properties need to be instantiated, to what degree, and in which combinations, for S1 and S2 to be friends. Expect disagreement, borderline cases, degrees of typicality, polysemy. Would not expect to be able to give a precise, univocal analysis of friendship.
7 In Favor of Causation as a Cluster Concept 1.Evades major classes of counterexamples to univocal theories. 2.Explains theses 1-5 (Counterexamples, Disagreement, Vagueness, Degrees of Typicality and Polysemy). 3.Theory of Error: a.Why univocal theories: i.Fail ii.Were attractive in the first place iii.Are so numerous b.Why our intuitions are sometimes mistaken. 4.Provides a very natural account of how we make our causal judgments. 5.The cluster theory could be tested empirically (Properties, Disagreement, Vagueness, Degrees of Typicality).
8 Evades Counterexamples to Univocal Theories Preemption Counterexamples to Difference-Making Theories Trainee and Supervisor: Trainee and Supervisor are on a mission to kill Victim. Trainee shoots first and Victim bleeds to death. If Trainee hadnt shot, Supervisor would have stepped in and done so, again resulting in Victims bleeding to death. Counterfactual dependence not necessary for causation. Merely one unnecessary criterion. Trainees shooting counts as a cause in virtue of the local physical process (and perhaps his moral responsibility). Very natural diagnosis. Counterexamples explained. Cluster theory explains why there are counterexamples to counterfactual theories, and also why counterfactual theories are attractive. Inflation of dependence into a necessary (or necessary and sufficient) condition, when in fact only criterial.
9 Evades Counterexamples (2) Omission Counterexamples to Physical Process Theories Gardener: My plants died when I was away on vacation. If my gardener had watered them, as he was supposed to have done, they would not have died. A local physical process linking C and E is not necessary for causation. Causation in virtue of dependence, manipulability, moral responsibility, etc. Explanation of why omission is a counterexample to process theories: the criterion local physical process is inflated into a necessary and sufficient condition, when it is merely one property of the cluster. The intuition that Gardeners omission is not a cause (as would be defended by Dowe/Beebee) is mistaken for the same reason.
10 Disagreement and Vagueness Explained Squirrel Kick: A golf ball is rolling towards the hole and is about to drop in when a squirrel kicks the ball away. After a series of improbable collisions with trees, the ball nevertheless ends up in the hole. Did the squirrels kicking cause the ball to fall into the hole? There is significant disagreement about this example. And many individuals simply do not know what to say (indeterminacy/borderline case). Local physical process, but no dependence, no manipulability, no probability raising, squirrel deserves no praise. Not clear whether this is enough for causation.
11 Degrees of Typicality: Prototype Effects (Rosch) 1.Direct Rating: Asked to rate (on a scale from 1-7) how good an example of the category (e.g. bird) various members are (e.g. robin, penguin, ostrich, etc.) 2.Reaction Time: Tested for the speed of their reactions to question such as, A penguin is a bird. True or False? 3.Production: Asked to produce examples of the category (by listing or drawing). Some examples are rated higher than others, have shorter reaction times, and are more frequently produced. Appears that there are degrees of typicality.
12 Degrees of Typicality Explained Prototypical A kills B, rock breaks window. All cluster properties instantiated. Less Prototypical 1. Causation by omission: Fathers Inattention: Father fails to pay attention to his child during a visit to the local park. The child wanders off, onto a busy road and is crushed by a passing truck. Was the fathers inattention a cause of the childs death? No local physical process. The father's inattention is a less prototypical cause than the truck, which instantiates all of the cluster properties. Seems highly probable that empirical evidence would confirm the existence of prototype effects for causation.
13 Polysemy Not the same as ambiguity. Cricket bats vs. Bats in the belfry. The two meanings are unrelated; it is an accident that the same word is used. Homonymy. The book is on the table. The picture is on the wall. The shadow is on the ceiling. In these cases the meanings are clearly related. It is not an accident that the same word is used. One may not be aware of the polysemy at first; not obvious that there are different senses of on. Suggest that causation is like this: subtle polysemy. Each minimally sufficient subset corresponds to a different sense. For example, the father's inattention is a cause of the child death in different sense from the truck. These meanings are related in virtue of sharing properties of the cluster (e.g. dependence, manipulability, responsibility), and the related meanings therefore show a family resemblance.
14 Two Further Points In Favor of the Cluster Theory 1.Explains why there are so many theories of causation: a univocal theory for each property of the cluster. 2.Plausible account of how we actually make our causal judgments (cf. highly technical - epicyclic - theories).
Which Subsets are Sufficient? So far, have only said that some subsets are sufficient. But which? Hall (Two Concepts of Causation 2004) has suggested that dependence is sufficient for causation, and also that production is sufficient for causation. C is a cause of E iff (E depends on C) v (C produces E). Counterexample to Sufficiency of Dependence Queen Elizabeth: My plants died when I was away on vacation. If Queen Elizabeth had watered them, they would not have died. Reply: Need to add some normative condition: moral responsibility, responsibility or expectedness. Counterexample to the Sufficiency of Production Flip: Suzy flips a switch that diverts a trolley away from the main track, and down subtrack A. The trolley later regains the main track and crushes Victim. If Suzy hadnt flipped, the trolley would have stayed on the main track, but Victim would have met the same fate.
Which Subsets are Sufficient? Counterexample to the necessity of (Production v Dependence): Trainee and Supervisor AAAD: Trainee and Supervisor are on a mission to kill Victim. Trainee shoots first with his action-at-a- distance vaporizing gun and Victim is vaporized. If Trainee hadnt shot, Supervisor would have stepped in and done so, again resulting in Victims vaporization. Reply: Add dependence holding fixed some G to the cluster. I propose three candidate sufficient subsets: 1.Dependence & responsibility 2.Production & responsibility 3.Dependence holding fixed some G & responsibility Some differences from Halls Two Concepts: 1.One concept of causation (polysemous). 2.More sufficient conditions. 3.Vagueness.
Reasons for Wanting a Precisified Univocal Analysis of Causation in the Sciences 1.To force scientists to be clearer about what they mean when they say C is a cause of E. 2.To fit in with broader metaphysical commitments: a.Network model - in which negative events cannot be causes. b.Causation as an objective, non-moral relation. 3.To line up with experimental practice for discovering causes. 4.To be consistent with axioms that are useful for causal inference from non-experimental statistical data (e.g Markov condition, Faithfulness condition). Such analyses will have to deviate substantially from ordinary usage.
Conclusions 1.Causation is a polysemous and vague cluster concept, with a prototypical structure. 2.No descriptive univocal analysis can succeed. 3.While there are no necessary and sufficient conditions, there are a variety of sufficient conditions. 4.There are good reasons for wanting a precisified univocal analysis of causation in physics and biology. Such analyses will be highly revisionary. But this is justified if the scientific, metaphysical or epistemological payoff is significant.