The Lab: Work Stuff

James McAllister's 2003 article, ``Algorithmic randomness in empirical data'' claims that empirical data sets are algorithmically random, and hence incompressible. We show that this claim is mistaken. We present theoretical arguments and empirical evidence for compressibility, and discuss the matter in the framework of Minimum Message Length (MML) inference, which shows that the theory which best compresses the data is the one with highest posterior probability, and the best explanation of the data.

Computer-based argument mapping greatly enhances student critical thinking, more than tripling absolute gains made by other methods. I describe the method and my experience as an outsider. Argument mapping often showed precisely how students were erring (for example: confusing helping premises for separate reasons), making it much easier for them to fix their errors.

This paper presents an attempt to integrate theories of causal processes---of the kind developed by Wesley Salmon and Phil Dowe---into a theory of causal models using Bayesian networks. We suggest that arcs in causal models must correspond to possible causal processes. Moreover, we suggest that when processes are rendered physically impossible by what occurs on distinct paths, the original model must be restricted by removing the relevant arc. These two techniques suffice to explain cases of late pre&eumlaut;mption and other cases that have proved problematic for causal models.

We introduce and discuss the use of Bayesian networks for causal modeling. Despite their growing popularity and utility in this application, numerous objections to it have been raised. We address the claims that Chickering's arc reversal rule undermines a causal interpretation and that failures of Reichenbach's Common Cause Principle, or again failures of faithfulness, invalidate causal modeling. We also argue against Pearl's deterministic interpretation of causal models. Against these objections we propose new model-building principles which evade some of the difficulties, and we put forward a concept of causal faithfulness which holds when faithfulness simpliciter fails. Finally, we particularize our account of type causal relevant to token causal relevance, providing an alternative to the recent deterministic accounts of token causation due to Hitchcock and Halpern & Perl.

We present a probabilistic extension to active path analyses of token causation (Halpern & Pearl 2001, forthcoming; Hitchcock 2001). The extension uses the generalized notion of intervention presented in (Korb et al. 2004): we allow an intervention to set any probability distribution over the intervention variables, not just a single value. The resulting account can handle a wide range of examples. We do not claim the account is complete — only that it fills an obvious gap in previous active-path approaches. It still succumbs to recent counterexamples by Hiddleston (2005), because it does not explicitly consider causal processes. We claim three benefits: a detailed comparison of three active-path approaches, a probabilistic extension for each, and an algorithmic formulation.

The investigation of probabilistic causality has been plagued by a variety of misconceptions and misunderstandings. One has been the thought that the aim of the probabilistic account of causality is the reduction of causal claims to probabilistic claims. Nancy Cartwright (1979) has clearly rebutted that idea. Another ill-conceived idea continues to haunt the debate, namely the idea that contextual unanimity can do the work of objective homogeneity. It cannot. We argue that only objective homogeneity in combination with a causal interpretation of Bayesian networks can provide the desired criterion of probabilistic causality.

A quick introduction to Bayes nets and probabilistic causation, followed by an analysis of causal interaction. Based on the Causal Interaction papers and the more general papers and talks not yet posted here. I gave a version of this talk at AAHPSSS in Sydney in July, and this version at Indiana University in Bloomington in September 2002.

For philosophers and computer scientists. Gives a general account of causal interaction using Bayes nets. Shows that noisy-OR and logit (two common models used in Bayes net research) are special cases. Shows that previous accounts in philosophy have been badly mistaken. Suggests a way to measure causal interaction. 11 double-column pages.

For computer scientists mostly. A shortened version of TR 2002/118 (fewer examples, no philosophy), and more details about measuring the interactions. Some results. 8 single-column pages

For philosophers mostly. A shortened version of TR 2002/118 (barest mention of noisy-OR and logit models), with a better discussion of Ellery Eells and the causal interaction debate in philosophy of science. Superseded by "A criterion for probabilistic causality" (above), Nov. 2003.

No, the Maya were not on the road to modern astronomy, so far as we can tell. The question is, what can we tell besides that? Can we evaluate and study scientific traditions which are not on our own branch of the intellectual phylogeny? Hopefully this paper has provided a clear analysis of various aspects of science which may be in mind when we ask whether Maya astronomy was scientific, and explored those features of Maya astronomy which have a bearing on the question.

We present a general hierarchical Bayesian model where Intelligence Sources make Reports about events or states in the world, which we call Hypotheses. The underlying multi-entity Bayes net for even a simple scenario has hundreds of nodes. We hide the details via Wigmore diagrams and a Google Maps GUI. Our application domain is Intelligence data fusion in asymmetrical warfare (terrorism). Some Hypotheses ­ like whether a village is a threat ­ may be abstract or unobservable. For these, we define Indicators ­ more observable Hypotheses whose value has some bearing on the target Hypothesis. The hierarchy can be arbitrarily deep, and Reports can provide evidence at any level. Furthermore, all Sources have credibility models. Traditional Sources are physical sensors with well-known error models. Non-traditional Sources include humans, websites, news, etc. For these Sources, our credibility models include Hypotheses about unknown factors like objectivity, competence, accuracy, reliabil- ity, and veracity. Every Report by a Source provides evidence about those factors. So, for example, successful ad hominem attacks against one Source can undermine his assur- ances that a village is safe, and lead us to believe it is hostile after all.