Workshop: Recent Work in the Philosophy of Probability and Conditionals (18 May 2018)
Idea and Motivation
This workshop will serve as an opportunity for philosophers working on the philosophy of probability and the philosophy of conditionals to convene with each other, and to discuss new trends in both areas.
- Jean Baccelli (LMU/MCMP)
- Benjamin Eva (University of Konstanz/MCMP)
- Alan Hájek (Australian National University)
- Stephan Hartmann (LMU/MCMP)
- Karolina Krzyżanowska (University of Amsterdam/MCMP)
- Rush Stewart (LMU/MCMP)
- Reuben Stern (LMU/MCMP)
- Reuben Stern (LMU/MCMP)
|09:15 - 09:30||Welcome|
|09:30 - 10:30||Reuben Stern (with Stephan Hartmann): The Ineffable Learning Handbook|
|10:45 - 11:45||Rush Stewart (with Michael Nielsen): Obligation, Permission, and Bayesian Orgulity|
|12:00 - 13:00||Jean Baccelli: The Problem of State-Dependent Utility: A Reappraisal|
|13:00 - 14:30||Lunch Break|
|14:30 - 15:30||Karolina Krzyżanowska: Learning that P is a Reason for Q: Indicative Conditionals, Testimony, and Probabilistic Relevance|
|15:45 - 16:45||Benjamin Eva: Revising Confidence Orderings and The Comparative Ramsey Test|
|17:00 - 18:15||Keynote: Alan Hájek: Ω|
This paper focuses on the problem of state-dependent utility, i.e., the challenges posed by state-dependent utility to the behavioral identification of beliefs. It examines two views that are well established in the literature. The first view is that expected utility and non-expected utility are equally exposed to the problem of state-dependent utility. The second view is that any choice-based solution to this problem must involve moral hazard, i.e., situations in which the decision-maker can influence the resolution of the uncertainty to which she is exposed. I show that these two views must be rejected at once. Non-expected utility is less exposed than expected utility to the problem of state-dependent utility, and there are choice-based solutions to this problem that do not involve moral hazard. In light of these new conclusions, I update the standard methodological interpretation of the problem of state-dependent utility.top
Contemporary formal epistemology is dominated by the analysis of two kinds of epistemic attitude: qualitative 'full' belief and numerically graded belief (or 'credence'). However, there is a third kind of epistemic attitude that appears to play a similarly fundamental role in everyday reasoning and decision making, but which is largely neglected in the current literature – namely, comparative confidence judgements of the form 'I am more confident in the truth of p than I am in the truth of q', or 'I am equally confident in the truth of p and q'. While some progress has been made in identifying the synchronic coherence constraints that determine the rationality of an agent's comparative confidence at a given time, practically nothing has been said about how a rational agent should change their comparative confidence judgements over time in the face of new evidence. In this paper, I aim to fill this lacuna by developing a general framework for the revision of comparative confidence judgements. Specifically, I show how the Ramsey test for conditionals and belief updating can naturally be translated into a comparative setting, forward some novel principles governing comparative confidence judgements involving indicative conditionals, and map out how the resulting revision procedure relates to its counterparts in the theories of qualitative and numerically graded belief.top
Probability theory is the dominant approach to modeling uncertainty. We begin with a set of possibilities or outcomes, usually designated ‘Ω’. We then assign probabilities—real numbers between 0 and 1 inclusive—to subsets of Ω. Nearly all of the action in the mathematics and philosophy of probability for over three and a half centuries has concerned the probabilities: their axiomatization, their associated theorems, and their interpretation. I want instead to put Ω in the spotlight.
Ω is a set of possibilities, but which possibilities? While the probability calculus constrains our numerical assignments, and its interpretation guides us further regarding them, we are entirely left to our own devices regarding Ω. What makes one Ω better than another? Its members are typically not exhaustive—but which possibilities should be excluded? Its members are typically not maximally fine-grained—but how refined should they be? I will discuss both philosophical and practical problems with the construction of a good Ω. I will offer some desirable features that a given Ω might have, and some heuristics for coming up with an Ω that has them, and for improving an Ω that we already have.top
Karolina Krzyżanowska: Learning that P is a Reason for Q: Indicative Conditionals, Testimony, and Probabilistic Relevance
Our beliefs change in response to what we learn, and a lot of what we learn comes from the testimony of other people. The extent to which our beliefs change may depend on how many people are the sources of given information, or how reliable their expertise makes them be. It is not entirely clear, however, what is it exactly that people learn when the testimony has the form of an indicative conditional, and how it is affected by the reliability or the number of speakers asserting it. On a recently revived view on conditionals, a conditional, “If p, then q,” does not only convey that the probability of q given p is high, but also that p is a reason for q, or, more formally, that p raises the probability of q. Consequently, learning a conditional should be expected to increase the so-called probabilistic relevance. In this talk, I will present empirical data supporting this hypothesis and discuss its significance for the conditional reasoning research.top
There are cases of ineffable learning—i.e., cases where an agent learns something, but becomes certain of nothing that she can express—where it is appropriate for the agent to update her credences by Jeffrey conditionalization. But there are also cases of ineffable learning where Jeffrey conditionalization leads the agent astray, and where Bayesians standardly offer no advice about how to update. In this paper, we tackle these cases, first, by distinguishing cases where the agent’s error occurs prior to her application of Jeffrey conditionalization from cases where the agent errs only upon applying Jeffrey conditionalization, and, second, by using the distance-based approach to Bayesian learning to develop new updating procedures that apply to the latter cases.top
This talk is a corrective to an increasingly popular way to misunderstand Belot's Orgulity Argument. The Orgulity Argument charges Bayesianism with defect as a normative epistemology. For concreteness, we reply to Cisewski et al.'s recent rejoinder to Belot's argument. The conditions that underwrite their version of the argument are too strong and Belot does not endorse them on our reading. A more compelling version of the Orgulity Argument than Cisewski et al. present is available, however---a point that we make by drawing an analogy with de Finetti's argument against mandating countable additivity. Moreover, we show that Elga's strategy of appealing to finitely additive probability to meet the challenge posed by the Orgulity Argument can be extended considerably to meet variations of the challenge.top
Attendance is free, but registration is required. If you want to register, please send notice to Reuben.Stern@lrz.uni-muenchen.de.
Munich Center for Mathematical Philosophy
If you need help finding the venue or a particalur room, you might consider using the LMU's roomfinder, a mobile web app that lets you display all of the 22.000 rooms at the 83 locations of the LMU in Munich.