EpImp Concluding Conference 2024
Posted on February 22, 2025 by Kevin Blackwell (edited by Arthur Van Camp)[ go back to blog ]
The final conference marking the end of Jason Konek’s ERC-funded Epistemic Utility for Imprecise Probabilities project was held in Bristol on the 18th and 19th of December. As with the EpImp Inaugural Conference in 2021, one of our core aims for the conference was to continue deepening ties between philosophers who use probabilistic models (especially in formal epistemology and decision theory) and IP researchers from the SIPTA community.
The conference was hosted in the stately Old Council Chamber of Wills Memorial Building. Several of our speakers have worked in Bristol, or visited for extended periods of time, but some of those who haven’t might have gotten a false impression of the typical grandeur of the lecture spaces here; whether or not it’s representative, it’s a lovely venue for a conference.
Day 1: Wednesday 18 December
Although the conference was only two days, there was still a great diversity of topics. Wednesday, the first day, started with some opening comments from Jason about his vision for the project and the conference. The first talk was from Francesca Zaffora Blando, “How to de-idealize: Computability and Imprecision”. The main idea of the talk was to connect results in the theory of algorithmic randomness to themes in traditional Bayesian epistemology, like the idea that we can expect opinions of agents to converge; computability is perhaps a more reasonable standard for rationality than the extremely strong demands of the traditional Bayesian paradigm (which requires logical omniscience, among other things). Although much of the talk was focused on results about algorithmic randomness in the precise context, Francesca also highlighted some of the recent work on this in the imprecise context. Next, we had Branden Fitelson on “Elimination Counterexamples: State of the Art”. Branden discussed several of the main varieties of responses to elimination counterexamples (cases where elimination experiments yield a credal state with worse accuracy than the prior, according to some scoring rule), settling on finding a proper scoring rule which is immune to these counterexamples as the most promising response. This is not trivial! But Branden closed the talk with an example of a scoring rule of this kind formulated by Bernhard Salow. Jason Konek and Giacomo Molinari gave the first of two tag-team talks in the conference: “An Accuracy Argument for Self Trust”. Jason’s portion was largely setting up the framework of how our team thinks about accuracy in the context of imprecise probabilities; then Giacomo used this background to present accuracy arguments for versions of trust principles that he has been developing. (It’s worth mentioning: Giacomo’s work on “Deference Principles for Imprecise Credences” won the 2024 Levi Prize!) Gregory Wheeler’s talk, “Beyond Expectations”, focused on extending desirability models to various cases beyond the simple case of linear utility. In addition to the formal results, Greg argued that attempts to reify objects like probabilities, or even expectations/previsions, miss out on much of what imprecise probabilities can describe and do for us. Gert de Cooman closed Wednesday with “Unifying Classical and Quantum probability: It’s All About Symmetries” – a talk giving an extremely general justification for coherence axioms for desirability, as well as functionals corresponding to prices/valuations; as the title suggests, this story does indeed unify both sets of desirable gambles in various classical contexts and probabilities in quantum mechanics in a very elegant way. For me, and I’m sure for many others, this talk was one of the highlights of the conference.
Day 2: Thursday 19 December
On Thursday, the first talk was “Structured Elicitation of Epistemic Uncertainty in Decision Problems” from Matthias Troffaes; this included a theoretical result about bounds on expert elicitation, and a practical application: an effective procedure to elicit useful information (to a particular decision problem) about an expert’s uncertainty in as few queries as possible. What Matthias’s talk wasn’t about was structured decision problems. (Matthias had sent me his slides as a back-up, and I was also introducing/chairing Dmitri’s talk – more on that later. I told Matthias that his talk on structured decision problems would be very interesting to philosophers; he was visibly confused, but apparently changed the slides to include talking about this process as a kind of structured decision in thinking about how to elicit uncertainty! Once I revealed my mistake, Matthias lightly ribbed me for the confusion at the start of his talk.) Next up, Catrin Campbell–Moore presented some of her continuing work on “Probability Filters”; as I said in my introduction, this is something that many of the people in the room were already very excited about – for anyone who didn’t already know, it was time to get excited. (It’s worth mentioning: at ISIPTA 2021 “in” Granada, Catrin’s poster on probability filters shared the gold prize.) I also had the privilege of introducing the next speaker, J. Dmitri Gallow; Dmitri and I overlapped as grad students at Michigan. His talk, “Structured Deliberation”, was not on how to effectively elicit expert uncertainty through choosing good queries, but on rationality failures that can arise in decision problems if you try to break them down into sub-problems that you resolve first. Dmitri’s talk sparked a very spirited Q&A, with a lot of very interesting discussion about the foundational assumptions of decision theory. The penultimate talk of the day was the final tag-team talk of the conference, given by the dynamic duo of Arthur Van Camp and Kevin Blackwell (me!). Unlike Giacomo and Jason’s talk, this one wasn’t really unified, except that both Arthur and I were talking about choice/rejection functions. Arthur’s talk, “Marginal Extension for Choice Functions”, was about how to combine marginal information about one variable with conditional information about how another variable depends on the first. (Psst! This simultaneously generalizes both Jeffrey and Adams conditioning to the context of choice functions.) My talk, “Coherent Rejection Functions for Arbitrary Things”, is pretty much what it says in the title: I give coherence axioms for rejection functions in a very general setting where all we assume is that there’s some background order on preferences which the agent treats as required; the objects in question can be any kind of things. The final talk of the conference was Kevin Dorst’s “Ambiguity Drives Hindsight Bias”. The main formal result of this talk was showing that hindsight bias is rational for Bayesian agents when, and only when, the agent’s prior is ambiguous and they trust their priors; Kevin also presented some empirical surveys he had conducted which seem to show that this is actually predictive of when people experience hindsight bias.
Social events
In addition to the formal exchange of ideas through the many great talks, there were many opportunities for more informal discussion. These included very nice conference dinners at Nutmeg and Bayt, and drinks at both Gin & Juice and Milk Thistle – two excellent places to get a cocktail in Bristol. Although not officially part of the conference itinerary, Branden Fitelson organized karaoke, which I was one of the enthusiastic participants in. I think events that Jason organizes, in addition to leading to productive academic conversations, are also outstanding in making the speakers feel welcome in Bristol and supporting the kinds of after-talk interactions that we all know are such a valuable part of conferences.
Conclusion
I think the Concluding Conference was a rousing success, by every metric. I certainly learned a great deal, as I’m sure everyone in attendance did. I think Jason’s project of helping to bridge the gap between those of us who come from a philosophical background and IP researchers coming from, roughly, the SIPTA community has been shockingly successful. And the EpImp team (Jason, Arthur, Giacomo, and me) is really, I think, a great example in microcosm of just how useful this kind of closer collaboration between the communities can be. It’s been a great honor and privilege to have been part of this project; to learn so much from Jason, Arthur, and Giacomo; and to meet so many other people doing amazing work, whether they think of themselves as philosophers or IP researchers. It is, of course, also a bit bittersweet. This amazing team that Jason put together, this collaborative research environment that is much rarer in philosophy than it should be, is coming to an end. For one shining moment, the gang was back together, and it was glorious. But I also believe the spirit of collaboration and community-building that was one of the cornerstones of the EpImp project will live on.
About the author
Kevin Blackwell is currently a Lecturer in the Department of Philosophy at the University of Bristol; until very recently, he was one of the two postdocs on Jason Konek’s EpImp project. He earned his PhD in Philosophy from the University of Michigan; his advisor was Jim Joyce. Kevin’s research focuses on formal epistemology, decision theory, and imprecise probabilities.