Causal and Probabilistic Reasoning
print


Breadcrumb Navigation


Content

Program

Room Arrangement

DateTimeAddress, Room
18 June 09:00 - 12:15
13:45 - 18:00 
Geschwister-Scholl-Platz 1, Room M010
Prof. Huber Platz 2, Room W201 
19 June 10:00 - 17:45 Prof. Huber Platz 2, Room V005
20 June 09:00 - 17:45 Prof. Huber Platz 2, Room W101

18 June

TimeTopic
08:30 - 09:00 Registration
09:00 - 09:15 Welcome
09:15 - 10:30 Keynote Lecture: Spohn, Wolfgang: Fifteen Dimensions of Evaluating Theories of Causation. A Case Study of the Structural Model and the Ranking Theoretic Approach to Causation
Chair: Gregory Wheeler
10:30 - 10:45 Coffee Break
10:45 - 11:30 Leuridan, Bert/Beilaen, Mathieu: A Logic for the Discovery of Causal Regularities
11:30 - 12:15 Fan, Da/Wang, Linton: Paradigmatic Causation and Multitudes of Trumping
12:15 - 13:45 Lunch
13:45 - 14:30 Danks, David: Graphical Models, Cognitive Representations, and Semantic Heterogeneity
14:30 - 15:15 Poellinger, Roland/Hubert, Mario: Bell’s Theorem and Non-Markovian Network Models
15:15 - 15:30 Coffee Break
15:30 - 16:15 Mayrhofer, Ralf/Waldmann, Michael: Agents and Causes: Dispositional Intuitions as a Guide to Causal Structure
16:15 - 17:00 Näger, Paul: The Causal Markov Condition and Non-Screening-Off Common Causes
17:00 - 17:15 Coffee Break

19 June

TimeTopic
10:00 - 11:15 Keynote Lecture: Hahn, Ulrike: Causal Argument
Chair: Karolina Krzy┼╝anowska
11:15 - 11:30 Coffee Break
11:30 - 12:15 Sydow, Momme von: Logical Inclusion Fallacies Within One Polytomous Dimension
12:15 - 13:45 Lunch
13:45 - 14:30 Lukits, Stefan: A Natural Generalization of Jeffrey Conditioning
14:30 - 15:15 Betz, Gregor: The Veritistic Merit of Probabilistic Degrees of Justification and Doxastic Conservativism in Belief Revision
15:15 - 15:30 Coffee Break
15:30 - 16:15 Horne, Zachary/Livengood, Jonathan: Ordering Effects, Updating Effects, and the Specter of Global Skepticism
16:15 - 17:00 Gyenis, Balazs: Do Genuine Ordering Effects Exist?
17:00 - 17:15 Coffee Break
17:15 - 18:00 Stenning, Keith/Martignon, Laura: A Qualitative Intensional Approach to Causal Reasoning and Decision in Uncertainty
18:00 - 18:45 Stenning, Keith/Varga, Alexandra: Normative Gloves That Fit Agents’ Hands: Consequences of a Multiple Logics-Framework for Reasoning
20:00 Conference Dinner (Cafe Reitschule)

20 June

TimeTopic
09:00 - 10:15 Keynote Lecture: Hertwig, Ralph: Navigating the Twilight of Uncertainty: Decisions from Experience
Chair: Michael Waldmann
10:15 - 10:30 Coffee Break
10:30 - 11:15 Carter, Sam: Probabilistic Judgement about Indicative Conditionals: A Formal Model-Based Theory
11:15 - 12:00 Pfeifer, Niki/Stöckle-Schobel Richard: The Probability of Indicative and Counterfactual Conditionals in Causal and Non-causal Settings
12:00 - 13:30 Lunch
13:30 - 14:15 Meder, Bjoern/Mayrhofer, Ralf/Waldmann, Michael R.: Structure Induction in Diagnostic Causal Reasoning
14:15 - 15:00 Brössel, Peter: On the Rationale of Reverse Inference in Neuroscience
15:00 - 15:15 Coffee Break
15:15 - 16:00 Sprenger, Jan/Colombo Matteo: Graded Causation and Explanatory Power, Explicated Probabilistically
16:00 - 16:45 Weinberger, Naftali: Probabilistic Causality, Structural Equations, and Causal Intermediaries
16:45 - 17:00 Coffee Break
17:00 - 17:45 Harinen, Totte: On the Need to Model Proportionality

Keynote Abstracts

Hahn, Ulrike: Causal Argument

The talk outlines the range of argument forms involving causation that can be found in everyday discourse. It also survey empirical work concerned with the generation and evaluation of such arguments. This survey makes clear that there is presently no unified body of research concerned with causal argument. The benefits of a unified treatment both for those interested in causal cognition and those interested in argumentation are highlighted, alongside the key challenges that must be met for a full understanding of causal argumentation.top

Hertwig, Ralph: Navigating the Twilight of Uncertainty: Decisions from Experience

In many of our decisions we cannot consult explicit statistics telling us about the relative risks involved in our actions. In lieu of explicit statistics, we can search either externally or internally for information, thus making decisions from experience (as opposed to decisions from descriptions). Recently, researchers have begun to investigate choice in settings in which people learn about options by experiential sampling over time. Converging findings show that when people make decisions based on experience, choices differ systematically from description-based choice. Furthermore, this research on decisions from experience has turned to new theories of decision making under uncertainty (ambiguity), “rediscovered” the importance of learning, and suggested important implications for risk and precautionary behavior. I will review these issues.top

Spohn, Wolfgang: Fifteen Dimensions of Evaluating Theories of Causation. A Case Study of the Structural Model and the Ranking Theoretic Approach to Causation

The point of the talk is not to defend any exciting thesis. It is rather to remind you of all the dimensions theories of causation must take account of. It explains 15 such dimensions, not just in the abstract, but as exemplified by the structural model and the ranking theoretic approach to causation, which, surprisingly, differ on all 15 dimensions. Of course, the subcutaneous message is that the ranking theoretic approach might be preferable. However, the main moral is to be: Keep all these dimensions in mind, and don't think that any one of these dimensions would be settled! Even if working at a specific issue, you are never on secure ground.top

Abstracts

Betz, Gregor: The Veritistic Merit of Probabilistic Degrees of Justification and Doxastic Conservativism in Belief Revision

There are at least two ways in which epistemic agents can be conservative in belief revision: (A) all novel beliefs acquired through evidential learning are entailed by the evidence and prior beliefs (cf. AGM axiom K*3); (B) when contracting their beliefs, agents attempt to give up as few beliefs as possible. This paper presents a model for assessing the veritistic merit of type-B conservativism while assuming that agents are type-A conservative. The basic idea is that epistemic agents choose a new belief set in light of (i) the alternative belief sets' degree of justification and (ii) their content loss compared to the old belief set. Different parametrizations of this choice function correspond to different doxastic strategies, whose veritistic merit can be assessed. The paper brings together work on verisimilitude and belief revision, ideas from Spohn's ranking theory, and argumentation-theoretic models of debate dynamics.top

Brössel, Peter: On the Rationale of Reverse Inference in Neuroscience

Reverse Inference is a popular inference schema in neuroscience, typically used to link evidence about brain activation with hypotheses about the elicitation of cognitive processes. However, there are many reasons to consider this inference schema fallacious. First, I argue that Machery's (2014) attempt to defend it by analyzing it in likelihoodist terms is unsuccessful. Second, I demonstrate that the rationale behind it is the same as behind the schema \Inference to the Best Explanation and that both inference schemas can be framed in Bayesian terms. Finally, I provide a vindication for both inference schemas by showing that they are truth-conducive.top

Carter, Sam: Probabilistic Judgement about Indicative Conditionals: A Formal Model-Based Theory

This paper aims to offer an explanation of otherwise unaccounted for results regarding individuals’ judgements about the probability of indicative conditionals. Studies into such judgements have found a bimodal distribution in participants’ responses, with a third pattern of response emerging under particular conditions. Both Bayesian theories (such as Evans & Over (2004), Oaksford & Chater (2000, 2007)) and mental model based theories (such as Johnson-Laird & Byrne (1991, 2002)) of conditional cognition fail to explain this data. I develop recent work in Koralus & Mascarenhas (2013) to provide a model which succeeds where these theories have failed.top

Danks, David: Graphical Models, Cognitive Representations, and Semantic Heterogeneity

Probabilistic and causal graphical models are spreading rapidly in cognitive science, particularly as models of cognitive representations. I first argue that a significant, though typically implicit, assumption of graphical models—semantic homogeneity of the edges—presents a prima facie objection to this spread: Many cross-domain or cross-cognition tasks require semantically heterogeneous representations, so semantically homogenous graphical models will be insufficient. I then directly overcome this objection by characterizing the licensed inferences in semantically heterogeneous graphical models. I conclude by using this generalized framework to explain some classic psychological data as normatively defensible inferences on semantically heterogeneous graphical models.top

Fan, Da/Wang, Linton: Paradigmatic Causation and Multitudes of Trumping

The objective of this paper is to propose and defend a synthesis of Hitchcock's two theses of token causation: first, that C is a paradigmatic cause of E just in case E counterfactually depends on C and C provides a satisfactory explanation for E; second, that from the perspective of contrastive causation, trumping cases in the literature are not simply cases of preemption, but are cases of non-redundancy or overdetermination depending on the choice of contrastive pairs. To accomplish the objective of the synthesis, we argue that, in both the contrastive and non-contrastive frameworks, Hitchcock's proposal for how an event C provides a satisfactory explanation of E, based on his notion of self-containment, is not yet fully satisfactory. Furthermore, based on Hitchcock's proposal, we develop an alternative notion of satisfactory explanation—the notion of integrity—which is suitable for both the contrastive and non-contrastive frameworks.top

Gyenis, Balazs: Do Genuine Ordering Effects Exist?

We present a disjunctive model of Bayesian learning which, to the extent permitted by the limitations of discrete probabilistic models of subjective belief revision, seems to provide a better mathematical representation of evidence than its better known alternatives. We base the model on results concerning the equivalence of a variety of models of learning. In particular we show that Jeffrey learning is merely a particular type of Bayesian learning if we accept some well motivated assumptions. The disjunctive model suggests that in the probabilistic learning models under consideration genuine ordering effects do not exist.top

Harinen, Totte: On the Need to Model Proportionality

Ordinary speakers' intuitions about scenarios of causal selection have received a lot of attention from the psychological and modeling perspective. This is because such intuitions are practically important and also because they are sufficiently uniform. In this paper, I argue that there is another type of intuition--the proportionality intuition--that is similarly practically important and sufficiently uniform, but which has nevertheless received very little attention from the psychological and modeling perspective. I argue that this state of affairs should be corrected and sketch some initial psychological factors influencing the proportionality intuition.top

Horne, Zachary/Livengood, Jonathan: Ordering Effects, Updating Effects, and the Specter of Global Skepticism

One widely-endorsed argument in the experimental philosophy literature maintains that intuitive judgments are unreliable because they are influenced by the order in which thought experiments prompting those judgments are presented. We show that the argument from ordering effects is defective. On one reading of what it is to be influenced by ordering, the empirical observation is well-supported, but the normative principle is false. On an alternative reading, the argument leads to global skepticism, since every basic source of justification is influenced by ordering in the relevant respect. We consider how the argument from ordering effects might be resisted.top

Leuridan, Bert/Beilaen, Mathieu: A Logic for the Discovery of Causal Regularities

We present a logic for the discovery of deterministic causal regularities starting from empirical data. Our approach is inspired by Mackie's theory of causes as inus-conditions and makes use of the adaptive logics framework. Our knowledge of deterministic causal regularities is, as Mackie noted, most often gappy or elliptical. The adaptive logics framework is well-suited to explicate both the internal and the external dynamics of the discovery of such gappy regularities. After presenting our logic, we consider some criticisms of the inus-account and how they affect our approach; and we compare our logic with a recent algorithm by Michael Baumgartner.top

Lukits, Stefan: A Natural Generalization of Jeffrey Conditioning

When we come to know a conditional, we cannot straightforwardly apply Jeffrey conditioning to gain an updated probability distribution. Carl Wagner has proposed a natural generalization of Jeffrey conditioning to accommodate this case (Wagner conditioning). The generalization rests on an ad hoc but plausible intuition (W). Wagner shows how the principle of maximum entropy (M) disagrees with intuition (W) and therefore considers (M) to be undermined. I argue that under assumptions that are natural to proponents of (M), (M) elegantly generalizes (W) (just as it generalizes standard conditioning and Jeffrey conditioning), far from being inconsistent with it.top

Mayrhofer, Ralf/Waldmann, Michael: Agents and Causes: Dispositional Intuitions as a Guide to Causal Structure

Violations of the Markov condition, a key constraint of Bayes nets, are prevalent in human reasoning. This tendency has even been observed with abstract scenarios, which weakens the possibility of background knowledge about hidden variables. However, although Bayes nets are popular to explain reasoning with variables, there are other causal tasks that have inspired different theories. These competing conceptualizations may crosstalk, and constrain each other. We will demonstrate this phenomenon by showing how dispositional assumptions about causal agents and patients influence the parameterization of Bayes net representations. Various experiments demonstrate how this hybrid account explains Markov violations in reasoning.top

Meder, Bjoern/Mayrhofer, Ralf/Waldmann, Michael R.: Structure Induction in Diagnostic Causal Reasoning

Many theories of diagnostic reasoning assume that inferences from effect to cause should reflect the observed conditional probability of cause given effect. We argue that this assumption is myopic, as it neglects uncertainty regarding alternative causal structures that may underlie the sample data. According to the structure induction model of diagnostic reasoning, diagnostic judgments should not only reflect the empirical probability of cause given effect, but should also depend on the reasoner’s beliefs about the existence and strength of the cause-effect relation. We confirmed this prediction empirically, demonstrating that the model accounts better for human diagnostic judgments than alternative theories.top

Näger, Paul: The Causal Markov Condition and Non-Screening-Off Common Causes

The causal Markov condition, which is a generalisation of Reichenbach’s principle of the common cause, is the central principle of causal explanation. In a non-technical way it says that every correlation has to be explained by a causal connection. While the principle provably holds for deterministic systems, van Fraassen (1980: The Scientific Image, 1982: Rational Belief and the Common Cause Principle) and Cartwright (1988: How to Tell a Common Cause, 1989: Nature’s Capacities and Their Measurement) have argued that the principle might fail in indeterministic systems, when common causes do not screen off, because they produce their effects not independently but in pairs. These cases pose the dilemma that one either has to deny that such systems are causal (van Fraassen’s horn) or one denies that the theory of causal Bayes nets adequately captures causal facts (Cartwright’s horn). In this talk I shall re-investigate this alleged failure and discuss scenarios for a via media, which upholds basic ideas of the theory of causal Bayes nets AND understands indeterministic systems (including the peculiar quantum cases) in a causal way.top

Pfeifer, Niki/Stöckle-Schobel Richard: The Probability of Indicative and Counterfactual Conditionals in Causal and Non-causal Settings

Conditionals and reasoning about conditionals are basic for human reasoning. As probabilistic conditionals allow for representing causal information, they are also basic for causal cognition. Our talk presents two experiments, which for the first time systematically compare how people reason about indicative conditionals (Experiment 1) and counterfactual conditionals (Experiment 2) in causal and non-causal task settings ($N=80$). The main result of both experiments is that conditional probability is the dominant response pattern and thus a key ingredient for modeling causal, counterfactual, and indicative conditionals. In the talk, we will give an overview of the main experimental results and discuss their relevance for understanding causal and probabilistic cognition.top

Poellinger, Roland/Hubert, Mario: Bell’s Theorem and Non-Markovian Network Models

The main implication of Bell's famous theorem is the non-locality of nature. In particular, every physical theory has to violate the Bell inequalities. A paradigmatic system showing non-local effects are two particles in an entangled state. Mermin (1985) emphasizes that no common cause can reproduce the statistical patterns of the experimental results, suggesting a non-local “connection” between the particles not inferable from classical physics and not embeddable in a standard Bayes net causal model obeying the causal Markov condition. We propose to understand the correlations as generated by physical entanglement proper and devise non-Markovian network models utilizable for causal inference.top

Sprenger, Jan/Colombo Matteo: Graded Causation and Explanatory Power, Explicated Probabilistically

Determining the strength of causal relations is a central problem for causal reasoning in science (e.g., in Bayesian Networks), and for assessing the efficacy of an intervention in a causal model. However, there is a surprisingly small amount of literature that tries to quantify the degree to which C is a cause of E. We contribute to this important research question by asking how causal strength differs from explanatory power, and we do so within a probabilistic framework. Our analysis combines the mathematical method of representation results with experimental work, based on simple vignettes where participants give causal and explanatory judgments.top

Stenning, Keith/Martignon, Laura: A Qualitative Intensional Approach to Causal Reasoning and Decision in Uncertainty

We present an experiment extending [Cummins, 1995] which shows that a Logic Programming (LP) network implementing fast and frugal decision heuristics can give a good model of subjects’ graded nonmonotonic naive causal reasoning, judgement and decision about familiar causal conditionals, on the basis of rich general knowledge. We com- pare this mainly qualitative proposal to ones based on Bayes Nets, both as cognitive models of subjects’ actual reasoning and as normative philosophical proposals, arguing that having a cognitively plausible intensional system for reasoning, judgment and decision raises questions of mutual interest to both philosophers and psychologists.top

Stenning, Keith/Varga, Alexandra: Normative Gloves That Fit Agents’ Hands: Consequences of a Multiple Logics-Framework for Reasoning

Philosophers currently mostly adopt probability as a single framework for normative studies. Some cognitive scientists have questioned probability for understanding how agents reason decide and act e.g. Simon’s bounded rationality ([Simon, 1972]); [Tversky and Kahneman, 1974] ‘Heuristics and Biases’ program. The ABC group’s fast and frugal decision heuristics [Todd et al., 1999] under some conditions outperform probability. [Stenning and van Lambalgen, 2008] explores Logic Programming for cognitive modelling of a broad sample of intensional reasoning.

This talk will argue, using the example of goals, that the choice of normative framework is not so easily decoupled from the framework being used by agents for making the decisions and actions under scrutiny.top

Sydow, Momme von: Logical Inclusion Fallacies Within One Polytomous Dimension

The discussion of conjunction fallacies (CFs) or, more generally, inclusion fallacies (IFs), is usually limited to dyadic relationships with dichotomous events (e.g., von Sydow, 2001). Here we address the more basic issue of probability judgments about nested hypotheses concerning mutually exclusive classes on a single dimension and spell out a corresponding variant of Bayesian logic (BL). We conduct two frequency-based experiments using material from the prominent Linda tasks (one concerned with jobs, the other with political attitude). The results corroborate the systematic occurrence of Ifs, and they are at odds with extensional probability and a confirmation account of inclusion fallacies.top

Weinberger, Naftali: Probabilistic Causality, Structural Equations, and Causal Intermediaries

I elucidate the relationship between structural equations approaches to causality and earlier probabilistic approaches by arguing that only the former are able to account for path-specific effects. In cases where there are several directed paths between a cause and its effect, path-specific effects indicate the contributions of particular paths to the total effect. Eells (1991) acknowledges that probabilistic approaches cannot account for path-specific effects, but questions whether there is any account to be given. Recent methods (Pearl, 2001) do enable one to define such effects. The inability of probabilistic accounts to give a similar analysis is a genuine limitation.