Project on Causal Understanding
Project Workshop 2005:
Causal Understanding and Intuitive Physics
Timetable
Friday, 1st April
| 14:00 | Coffee |
| 14:15 | Welcome & Introduction |
| 15:00 | John Campbell: Grasp of Concepts as Grasp of Causal Role |
| 16:30 | Coffee |
| 17:00 | Brian Scholl: The Nature of Causal Perception |
Saturday, 2nd April
| 9:30 | Josep Call: Action, perception, and causality in apes |
| 11:00 | Coffee |
| 11:30 | Daniel Povinelli: Chimpanzees' conception of weight: A test case of the Reinterpretation Hypothesis |
| 13:00 | Lunch |
| 14:00 | Laura Schulz: Naïve Physics, Savvy Science? Experimental design, causal determinism and preschoolers. |
| 15:30 | James Woodward: Interventionist theories of causation in psychological perspective |
| 17:00 | Coffee |
| 17:30 | Su-Hua Wang: Intuitive physics in infancy: Effects of priming and teaching |
Sunday, 3rd April
| 9:30 | Alison Gopnik: Causal maps, spatial maps and Bayes nets: Bridging covariation, intervention and mechanism. |
| 11:00 | Coffee |
| 11:30 | Concluding discussion |
Abstracts
Josep Call (Psychology, Max Planck Institute for Evolutionary Anthropology, Leipzig): Action, perception, and causality in apes
Studies on causal knowledge in nonhuman animals have traditionally been based on individuals using tools to get out-of reach rewards. Although some species have succeeded in some of these problems, there are also numerous studies that have shown robust and somehow unexpected limitations. These negative results on causal knowledge are often difficult to interpret because the predisposition to use certain actions may seriously interfere with causal knowledge. Another approach that has received much less attention in comparative research is based on presenting natural causality problems without requiring subjects to use tools to get the reward, but instead use simple actions to select between alternatives. I will present several studies in which we investigated the ability of chimpanzees, gorillas, orangutans, and bonobos to make causal inferences regarding the location of hidden food. In these studies subjects have to choose between two containers (one of which is baited) after they witness the effect that various manipulations have on the containers. For instance, subjects are allowed to choose one of the containers after they are shaken by the experimenter. I will argue that these results show that subjects understand the causal connection between the food and certain features in the environment, and that they can infer the food location in the absence of direct visual or auditory information about the food. I will contrast these seemingly more positive results based on perceiving events with those based on producing events and discuss several aspects of causality in relation to these results.
John Campbell (Philosophy, Berkeley): Grasp of Concepts as Grasp of Causal Role
I argue that in some basic cases, grasp of a concept is a matter of knowing the causal role of a property. I locate this idea in a interventionist approach to causation, on which knowing the causal role of a property X is a matter of knowing something abou how one could manipulate other variables by manipulating X, and how one could mannipulate X by manipulating other properties. Formulating this approach correctly demands that we have in place the relation between a person's grasp of what an intervention is in general, and the person's own capacity for manipulative agency.
I argue that grasp of a concept, so conceived, is more fundamental that either knowledge of what it is for a predicate to be true of an object, or knowledge of how to tell that the predicate applies to an object.
Alison Gopnik (Psychology, Berkeley): Causal maps, spatial maps and Bayes nets: Bridging covariation, intervention and mechanism.
Traditionally, there has been an opposition between views of causation that emphasize covariation and those that emphasize mechanism. More recently, philosophers have stressed the central role of intervention in understanding causation and the formal relations between intervention and covariation. This set of ideas has been formalized in the causal Bayes net framework. I will suggest a way of integrating ideas about mechanism into this framework. The idea is that abstract causal relations in the world are assumed to both constrain patterns of covariation and intervention and to constrain spatial and physical patterns. Another way of putting this is to say that causal maps and spatial maps are mutually constraining, in the way that for example, tactile and visual maps of objects may be mutually constraining. I will present evidence from two sets of studies with four year olds exploring the relations between covariation and intervention, on the one hand, and spatial information on the other. In one set of studies children were equally willing to postulate causal relations when objects acted at a distance on other objects and when they made contact with those objects. In the other studies children assumed that objects with similar causal powers had similar internal structure and vice-versa.
Brian Scholl (Psychology, Yale): The Nature of Causal Perception
Long traditions in both philosophy and psychology have concerned themselves with the nature and origins of our notions of cause and effect, precisely because these notions seem to lie at the root of nearly all aspects of reasoning. In addition, however, even relatively hardwired visual processes seem to traffic in representations of causality, a la Michotte. We have been exploring the nature of causal perception, and I will describe and demonstrate several of our results, with an eye to relating these generalizations to those employed in studies of higher-level causal reasoning. Some specific discoveries include: (1) The rules which mediate the detection of causal relationships in visual processing are more nuanced than suggested by Michotte's research; (2) Causal perception interacts richly with other visual processes such as attention and perceptual grouping; (3) The detection of causal relationships in perception is not an epiphenomenon, but can also influence the perception of other visual properties; and (4) The perception of causality can be determined postdictively, such that new information which is obtained can effectively rewrite the immediate past in our conscious awareness of causal relationships. We conclude that nearly all aspects of causal perception can be explained by appeal to what is perhaps the single most fundamental principle of visual perception: coincidence-avoidance. Collectively these results emphasize that perception concerns not only a recovery of the physical structure of the world, but also a recovery of its causal structure.
Laura Schulz (Brain and Cognitive Sciences, MIT): Naïve Physics, Savvy Science? Experimental design, causal determinism and preschoolers.
Much of the work on causal reasoning in infancy and early childhood has focused on children's naïve physics. The work is consistent with a philosophical tradition suggesting that understanding causal mechanisms is fundamental to causal inference. However, children's ability to learn causal relations may extend beyond their understanding of causal mechanisms. In one set of studies, I will show that in the absence of mechanism knowledge and differential spatiotemporal information, preschoolers can distinguish causal chains, common cause and interactive causal structures. In another set of studies, I will suggest that preschoolers can override known, observable, causal mechanisms to make complex inferences about unobserved causes. I will suggest that this type of learning might be better explained by reference to other traditions in philosophy of science, including recent 'interventionist' assumptions about causal structure and Enlightenment assumptions about causal determinism.
Su-Hua Wang (Psychology, University of California at Santa Cruz): Intuitive physics in infancy: Effects of priming and teaching
As adults, we possess a great deal of knowledge about how objects move and interact. This physical knowledge guides us to encode key information about physical events, and as such enables us to better predict their outcomes. Prior research shows that infants' physical knowledge becomes richer as they identify more and more variables that are relevant for predicting the outcomes of the events. Converging evidence has revealed an interesting learning pattern: infants sometimes identify the same variable at different ages in different event categories. For example, the variable height is identified at about 3.5 months in occlusion events, 7.5 months in containment events, and 12 months in covering events. Consequently, infants begin to represent height information at different ages when watching events from different categories. Here I will present an account that attempts to explain the above pattern, and report two sets of studies that ask whether infants' physical representation can be enhanced through appropriate experiences in the laboratory.
In one set of studies, infants watched two consecutive events from different categories; the variable height has been identified as relevant for the first but not the second event. After watching the first event, infants as young as 4.5 months of age were induced or primed to represent height information in the second event, suggesting that infants? representation of the physical world is dynamic and flexible. In the other set of studies, infants watched teaching events that are designed to guide them to identify a new variable for an event category. After a brief teaching session, infants at 9 months of age identified height as a variable for predicting covering events and included height information in their representation of a covering event. In addition, negative results were obtained when the infants were not allowed to establish causal links in the teaching events. These findings suggest that infants' physical understanding can be enhanced through deliberate exposure to appropriate observations. The role of causal analysis in this learning process will also be discussed.
James Woodward (Philosophy, Caltech): Interventionist theories of causation in psychological perspective
This talk will explore a number of issues concerning the adequacy of interventionist accounts of causation of the sort developed in my recent book Making Things Happen: A Theory of Causal Explanation (Oxford, 2003) as empirical psychological theories of human and animal causal learning and judgment. Very roughly, according to such theories C causes E if and only if under some interventions on (ideal manipulations of) C (and perhaps other variables as well) changes in C are associated with changes in E. Interventionist accounts like mine were originally intended primarily as a normative accounts rather than as a descriptive accounts of the causal judgments that subjects actually make . Still one can certainly ask how such accounts fare as a descriptive theories and this is what I propose to do. Among the issues I will discuss are those having to do with "psychological reality" of the various notions that figure centrally in the interventionist account. Do people (adults, children) reason about causation in a way that respects the connection between causation and intervention? Do people respect the contrast between intervening and observing which is so crucial to the interventionist account? Do people reason in normatively appropriate ways in simple examples involving combinations of interventions? To what extent, if any, do people behave as if they make the default assumption that their voluntary actions have the characteristics of interventions? Do subjects connect causal claims and counterfactuals in the way that interventionst accounts say they should?
Other questions to be explored in connection with non-human animals: what is the relationship between instrumental conditioning and casual learning? How should we characterize the apparent differences in causal understanding between non-human primates and human children? In connection with this last question, I will propose a distinction between three levels of causal/instrumental understanding.
a) A purely egocentic viewpoint in which the agent grasps (or behaves as if he grasps) that his manipulations have stable effects (that there is an invariant relationship between his actions and certain outcomes) but stops at this point, not recognizing that the same invariant relationship can be present when other agents or nature intervene to produce similar effects. I take this to be characteristic of instrumental conditioning
b) An agent causal viewpoint in which the agent grasps that the very same invariant relationship that he exploits in intervening also can be present when other agents act.
c) A fully causal viewpoint in which the agent grasps that the very same invariant relationship that he exploits in intervening also can be present in nature, in the absence in any interventions at all, as in the salt chamber example. This last step involves thinking of causal relationships as relationships that are invariant not just under interventions but also under other sorts of changes as well. This is characteristic of human causal understanding. I will suggest that we may associate these with progressively stronger forms of causal learning. Humans, including small children, can learn from observing the interventions of others By contrast, non-human animals have a great deal of difficulty with this.