FAX2 Conference

Formal and Experimental Philosophy Workshop

In recent years there has been an explosion of work in both formal and experimental philosophy, as well as a growing number of researchers who are working at their intersection. This includes a broad tent of topics: the psychology of decision and moral judgment; the rationality of reasoning under uncertainty; the role of bias in human and algorithmic inferences; the epistemology and psychology of imagination; the structure of social networks; and so on. We believe that beneath the differences are a unified class of methods, models, and approaches.  The conference will bring researchers across these growing areas together to start fruitful conversations about these unified themes.

Conference Schedule:

Saturday, October 8th
9:30am – Coffee available in 1119
10-11am – Talk 1, Thomas Icard, Stanford University, “What do we want from a theory of explanation?””
11:30am – 12:30pm – Talk 2, Tania Lombrozo & David Kinney, Princeton University, “Building Compressed Causal Models of the World”
12:30 – 2pm: Break for lunch
2-3pm – Talk 3, Edouard Machery, University of Pittsburgh, “Factivity across languages”
3:30 – 4:30pm: Talk 4, Melissa Fusco, Columbia University, “Scope Considerations in Free Choice Permission”
5 – 6pm: Talk 5, Jonathan Philips, Dartmouth College, “Modal spaces as a foundation for high-level cognition”

Sunday, October 9th:
8:30am – Coffee available in 1119
9-10am – Talk 6, Hannah Rubin, University of Notre Dame, “Diversity and homophily in group inquiry”
10:15 – 11:15am – Talk 7, Johanna Thoma, London School of Economics, “Diversity and homophily in group inquiry”
11:30am – 12:30pm: Talk 8, Sina Fazelpour, Northeastern University, “Fairness in sociotechnical machine learning systems”
12:30 – 1:30pm: Break for lunch
1:30–2:30: Talk 9, Aydin Mohseni, “Why are the Human Sciences Hard?” (with Daniel Herrmann)
2:30pm : Informal chatting/coffee

Register here

Titles and Abstracts:

Thomas Icard
Title: What do we want from a theory of explanation?
Abstract: Theoretical accounts of explanation range from largely descriptive — focusing, e.g., on which explanations ordinary people find satisfying or convincing — to entirely normative, often aiming at analyses in wholly non-psychological terms. The latter tend to be formal and non-experimental, while the former are typically experimental though sometimes less formal. In this talk we plan to explore the possibility that the most productive mode for theorizing about explanation may be somewhere in between these extremes. Specifically, we pursue a version of the hypothesis that explanation is best understood in *communication-theoretic* terms. Such a perspective invites formal and normative underpinning, while also being strongly subject to empirical test.

Johanna Thoma
Title: Cost-Effectiveness Analysis of Risky Health Interventions: Moving Beyond Risk Neutrality
Cost-effectiveness analysis for health interventions is traditionally conducted in a risk neutral way, insensitive to risk attitudes in the population, which are potentially non-neutral. While the standard metric of quality-adjusted life years (QALYs) aims to be deferential to people’s own valuations of health states, cost-effectiveness of risky interventions using the QALY metrics is not similarly deferential to people’s risk attitudes. I argue that there is no good justification for this practice. Non-neutral attitudes to risk, especially where they concern individually life-changing interventions need not be irrational, and so imposing non-neutrality is not justifiable as a way of debiasing preferences. Many common justifications for deference to preferences regarding health states either extend to risk attitudes, in which case methods for measuring risk attitudes separately and incorporating them into cost-effectiveness analysis should be used more widely. But even if reasons for deference don’t extend, the imposition of risk neutrality as opposed to any other rationally permissible risk attitude is under-motivated as default practice. In that case, a richer set of information should be presented to political decision-makers, to enable them to decide how to take into account the individual risks faced by the population, on top of aggregate effects on health.

Melissa Fusco
Title “Scope Considerations in Free Choice Permission”
Free Choice Permission (Kamp, 1973) is the apparent natural-language validity of the inference from May(p or q) to the conjunction of May p and May q. Much recent work on Free Choice has interacted with two questions: (i) whether the inference is pragmatic or semantic; (ii) whether the felt force of the inference obtains when the disjunction in the premise has wide (rather than narrow) scope. In this presentation, I will take a page from the syntax literature to explore both questions. For (i), I claim that the time-honored “I don’t know which”-riders on Free Choice sentences, traditionally taken to show that the effect is pragmatic, are sensitive to scope. Careful attention to such riders suggests that these sluices do not show cancellation on Free Choice antecedents in which disjunction scopes narrower than the modal. This constrains, but does not completely settle, the answer to (ii).

David Kinney and Tania Lombrozo
Title: Building Compressed Causal Models of the World
Abstract: In both cognitive science and formal epistemology, agents’ representations of the causal structure of their environment can be formalized using a Bayesian network. Specifying a Bayesian network requires that we choose a level of granularity at which variables are defined. Here, we develop a formal theory according to which variable choices reflect a trade-off between compression and informativeness, where the optimal trade-off depends on the decision-theoretic value of information for a given agent in a given context. This theory predicts that, when a particular decision context is not salient, agents prefer causal models that are as compressed as possible without sacrificing too much informativeness. However, when a particular decision context is salient, our theory predicts that agents will favor compressed causal models that sacrifice informativeness with respect to the underlying dynamics of some target system, as long as the information lost is not relevant to the salient decision context. This framework allows us to unify two dimensions of causal claims – proportionality and stability – that have received significant attention from philosophers of science. Over the course of four experiments, we show that preferences over causal models are in keeping with the predictions of our theory. We conclude by discussing some potential implications of our work for the use of social categories in cognition.

Hannah Rubin
Title: Diversity and homophily in group inquiry
Abstract: Diversity of identities has been argued to be important to inquiry. Yet, the potential benefits do not always materialize and efforts to promote diversity can backfire, both impeding inquiry and resulting in increased inequity. We argue that one reason for the gap between potential and realized benefits of demographic diversity is limitations to the inferences we can draw based on experimental results in which small groups are shown to benefit from decreased group-based conformity. We use formal models to show that factors that may be beneficial to group inquiry in restricted experimental settings (e.g., where everyone is talking to everyone) may be detrimental when we consider them in the context of how larger groups interact and share information. In particular, we focus on homophily – the tendency of individuals to associate with similar others – and show that its impact on inquiry depends both on which dimension of similarity is salient (e.g., social demographics, beliefs, or values) and how it manifests in different structural and behavioral effects (e.g., forming connections or trust relations).

Edouard Machery
Title: Factivity across languages
I will review the results of the geography of philosophy project about factivity across languages for epistemic, perceptual, emotive, and declarative verbs and examine how they bear on semantic vs. pragmatic theories of factivity presupposition.

Sina Fazelpour
Title: Fairness in sociotechnical machine learning systems
The increasing use of machine learning (ML) algorithms in consequential decision pipelines has made it imperative to carefully evaluate their impact and to design appropriate socio-technical responses. Importantly, however, current practices of evaluating and improving algorithm-informed decisions tend to narrowly focus on the algorithms in isolation from organizational and social contexts in which they are embedded. Focusing on issues of justice and fairness, in this talk, I first outline how this failure to incorporate the dynamics of deployment critically undermines our ability to anticipate the impact of algorithm-informed decisions in context. I will then discuss some of our recent and ongoing works that take steps towards broadening ethical assessments to include key aspects of deployment dynamics. I will end by considering how these works motivate a fundamental re-orientation of our normative thinking.

Jonathan Phillips
Title: Modal spaces as a foundation for high-level cognition
Abstract: High-level cognition relies on the ability to determine what the relevant possibilities are in the face of incomplete information. For example, to judge that someone is morally responsible for a given action requires assessing what other actions were available to the agent at the time. Defining the set of relevant possibilities in a given decision context—what we call the contextual `modal space’—is computationally intractable both because such contexts are not well defined and because there are indefinitely many possibilities that one could consider. Intriguingly, the high-level judgments that require reasoning over such possibilities are often made quickly and effortlessly, suggesting that they may recruit a heuristic or implicit representation of the contextually relevant modal space. Investigating this possibility, I will present work that (1) introduces a sampling method for empirically measuring contextual modal spaces, (2) finds that modal spaces are constrained by `normality’, (3) demonstrates that they are domain-general and recruited across four distinct forms of high-level reasoning, (4) shows that manipulating modal spaces changes high-level judgments, and (5) provides evidence that heuristic representations are used rather than on-line sampling of relevant alternatives.

Aydin Mohseni
Title: Why are the Human Sciences Hard? (with Daniel Herrmann)
Abstract. The human sciences, it is often claimed, are hard. Most explanations of this claim locate the hardness in some intrinsic property of the human subject—complexity, context-sensitivity, adaptivity, and the like. While we do not disagree with these hypotheses, we present a distinct way to conceptualize the challenge of the human sciences. We attempt to formalize one sense in which a science may be ‘hard’—in terms of performance in prediction tasks—and use this to articulate two novel, higher-level hypotheses as to why the human sciences may be particularly hard.

Register here