 |
| home |
| modal logic |
| type theory |
| rational choice |
| probabilistic logic |
| uncertainty frameworks |
| probability, causation and confirmation |
instructors:Frederick Eberhardt (Washington University in St. Louis, USA) course proposal:This course is an introduction to causal Bayes networks and we demonstrate the power of these techniques by addressing an open problem in formal epistemology: the relationship between epistemic coherence and confirmation. The coherence theory of justification maintains that, other things being equal, we have more reason to believe a body of propositions that ‘cohere’ than we do a body which does not. There are many situations in which epistemic coherent evidence seems to psychologically invite belief in a hypothesis, confirm a common cause, or lend credence to a unifying explanation. For example, after arriving in the study and finding that the electronic clock is blinking 12:17, that the computer has automatically rebooted and is asking for a password, and that the microwave is also blinking 12:17, it is irresistible to avoid hypothesizing that there was a power outage that affected the whole house. But the problem is that there are just as many cases in which coherence does not invite belief. For example, arriving home late and finding a messy kitchen, one hears, in turn and independently, one's oldest, middle, and youngest child attest, with maximal coherence that the cleaning lady forgot to do the kitchen. Nobody has been able to account for why coherence helps in some circumstances but not in others, and recent impossibility results (Bovens and Hartmann 2003; Olsson 2005) have suggested that probability is of little help. However, recent work suggests that pure probabilistic accounts of epistemic coherence under-constrain the problem by not making explicit the causal structure regulating the relationships between evidence and hypothesis (Wheeler 2009, Wheeler and Scheines, to appear). Once assumptions about causal structure are fixed, then stable relationships between probabilistic association and confirmation emerge. Further, since we now know a lot about how to tell, from data, whether or not a set of measured variables are indeed effects of an unmeasured common cause and otherwise causally independent (Junker and Ellis 1997, Glymour 1998, Silva et. al. 2006, Eberhardt 2008, Eberhardt to appear), these methods can be combined with probabilistic measures to provide useful indicators of epistemic coherence. course objectives:This course offers an introduction to the techniques of causal inference using the framework of causal Bayes nets, and uses recent result within formal epistemology linking probabilistic association, incremental confirmation, and causal structure, as an organizing theme. The course starts by introducing the foundational connections between probabilistic and causal knowledge, and shows how causal knowledge can be learned from probabilistic data using simple algorithms. The explicit formal representation of causal relations is then used to analyze the recent impossibility theorems for measures of evidential coherence and theories of confirmation to show that they depend upon particular assumptions of causal structure that are not general. The course is designed to meet two main objectives. First, introduce students to the causal Bayes net formalism and its foundations. Second, introduce students to a current research topic within formal epistemology-- measures of evidential coherence -- and illustrate how to make progress on this topic with the theory of causal Bayes nets. tentative course outline:
selected bibliography:
|