Abstracts

Bas van Fraassen

Saturday, June 1      09:45 - 10:30

Probabilities of Conditionals: Three Easy Pieces

I will argue for three claims. First, that lack of closure under conditionalization is ubiquitous and not a fault. Second, that the CCCP (Stalnaker’s Thesis) has consequences for our basic concept of the relation of logical consequence. Third, that Alan Hajek’s view, that conditionals are true only if the corresponding conditional probability equals 1, can receive a tenable formulation by relinquishing possible world semantics.

Xueyin (Snow) Zhang

Saturday, June 1      10:30 - 11:15

Deference done multiple ways

We defer to experts about questions of their expertise. But how should our opinion be constrained by information about the opinion of an expert? This presentation has two goals. First, I argue that there are many kinds of epistemic experts; how one should defer to a given expert E depends on the kind of expert E is. Second, I propose one way of classifying experts based on (i) the kind of practical decision problems that it is rational for the agent to coordinate with the expert and (ii) the manner of their coordination. One upshot of this taxonomy is that it uncovers some new deference principles and open questions that have not been investigated in the literature.

Xinhe Wu

Saturday, June 1      11:15 - 12:00

Vague Identity: A Uniform Approach

There are numerous apparent examples of vague identity, i.e. examples where two objects appear to be neither determinately identical nor determinately distinct. Philosophers disagree on whether the source of vagueness in identity is semantic or ontic/metaphysical. In this talk, I explore the use of Boolean-valued models as a many-valued semantic framework for identity. I argue that this semantics works well with both a semantic and ontic conception of vague identity. I also discuss, in the context of Boolean-valued logic, responses to the Evans’ argument under the two conceptions.

Jacqueline Harding

Saturday, June 1      12:15 - 12:45

A New Pragmatics of Explanation

Explanations are acts of communication: a speaker explains something to a listener. Indeed, we spend an awful lot of our lives giving and receiving explanations. Read too much philosophy of explanation, though, and you’d probably forget this (a few honourable mentions notwithstanding). The philosophical problem of explanation is taken to be a matter of locating the right explanatory relations in the world, with human interests and capacities offered at best a marginal role. In particular, philosophers have paid basically no attention to psychological work on explanation. I'll present work in progress (with Tobias Gerstenberg and Thomas Icard) in which we attempt to rectify this. Our claim is extremely simple: explanation is communication of dependence information. We make this precise using tools from computational cognitive science, showing that our proposal accounts for a lot of what philosophers and psychologists say about explanation.

Ahmee Christensen

Saturday, June 1      12:45 - 13:15

Logics of knowledge and knowability

Knowability has received attention from a number of perspectives in logic, beginning with the introduction of Fitch's paradox and more recently coming from dynamic epistemic logic. In this talk, however, we will take knowability to be a primitive modal operator in a nondynamic language. After motivating a particularly natural class of epistemic temporal models and defining a semantics, we will establish sound and complete axiomatizations of two logics: one of knowledge and knowability and one of pure knowability.

Balder ten Cate

Saturday, June 1      15:00 - 15:45

Van Benthem Theorems for Algebras of Binary Relations

Several areas of computer science (databases, knowledge representation, program verification) feature algebraic languages in which one can define complex binary relations from basic ones, through a certain set of operations (e.g., composition, intersection, inverse, …). Examples include various dynamic logics, as well as recently ISO-standardized SQL extensions for querying graph databases. Generalizing from these, we can define a “binary relation algebra” to be any algebraic signature consisting of finitely many operations on binary relations. Tarski's relation algebra is another classic example of such an algebra of binary relation.

Preservation theorems link semantics properties of formulas to their syntactic shape. The most famous example is the Los-Tarski theorem, which states that a first-order formula is preserved under induced substructures (i.e., remains true if one passes from a structure to an induced substructure) if and only if it can be written without using any existential quantifiers (assuming negation normal form). Another famous example is Johan van Benthem’s celebrated bisimulation characterization theorem, which states that a first-order formula is preserved by bisimulations if and only if it can be written as a modal formula.

In this talk, we will review some recent results (both positive and negative) about the existence of preservation theorems for algebras of binary relations. The starting point for this is yet another classic result of Johan van Benthem, namely the Bisimulation Safety Theorem, which characterizes the first-order operations on binary relations that preserve bisimulations.

Fenrong Liu

Saturday, June 1      15:45 - 16:30

Johan's Journey: Bridging Cultures and Inspiring Minds Through Logic

In this talk, I will highlight the profound impact Johan has had on logic education in China over the past 20 years. Beginning with his initial visit in 2004, Johan's lectures and collaborative efforts have significantly advanced the field, connecting scholars and students across continents. I will discuss his contributions and illustrate how his dedication has fostered a new generation of logicians in China and beyond.

Shane Steinert-Threlkeld

Saturday, June 1      16:30 - 17:15

Unnatural Language Semantics

This talk will survey a range of research projects stemming from / inspired by Johan van Benthem's pioneering work on natural language semantics in the 80s and 90s, focusing on the themes of fine structure and of universal constraints on possible meanings.

Wesley Holliday

Sunday, June 2      09:30 - 10:15

A new impossibility theorem in social choice theory

I will explain a new impossibility theorem in social choice theory (https://doi.org/10.1016/j.econlet.2024.111589), discuss some connections of this work to logic, and raise some implications for practical reform of electoral systems.

Ashish Goel

Sunday, June 2      10:15 - 11:00

New Directions in Social Choice: Deliberation, Representation, and Complexity

In this talk, we will touch upon three aspects of social choice that have gained additional salience in the digital age: deliberation, representation, and voting in complex domains. We will first describe the Stanford Online Deliberation platform, which is a video conferencing platform for civic deliberations that incorporates an automated moderator. We will provide empirical results from this deliberation platform. We will also formally describe a sequential negotiation process that has provably good properties, and outline new research directions. We will then illustrate the difficulty of running a civic feedback process that is representative of the underlying population using a budgetary feedback process that we ran with the city of Austin as an example. This feedback process saw dramatic changes in the number of participants and how they voted over its lifetime. Our experience leads to both theoretical and normative questions regarding sortition as well demographic reweighting. We will provide preliminary progress towards some of the theoretical questions and leave the normative questions for general discussion. Time permitting, we will motivate the design of more complex menus for Participatory Budgeting, and describe some recent results and open problems.

Fenrong Liu

Sunday, June 2      11:00 - 11:45

Knowing is winning: An epistemic approach to the hide and seek game

The game of hide and seek serves as a robust testbed for addressing numerous computational challenges within pursuit-evasion scenarios. In this presentation, I will introduce the Epistemic Logic of Hide and Seek Game (ELHS), a formal framework specifically developed to define key aspects of the game, such as players' positions, observational capabilities, knowledge, and winning conditions. Utilizing ELHS, I will present a novel method for tracking player interactions and managing information updates during gameplay. This is supported by a newly defined dynamic operator, whose properties will be explored. In addition, I will provide a comparative analysis of this method against traditional dynamic epistemic logic (DEL) updates, highlighting its distinct advantages and identifying new technical challenges for future research. This is joint work with Dazhu Li and Sujata Ghosh.

Guillaume Massas

Sunday, June 2      12:00 - 12:30

Possibility Semantics and van Benthem's "Possible World Semantics for Classical Logic"

Possibility semantics is a semantic framework in which formulas are evaluated at partial possibilities rather than at maximally determined possible worlds. Although they have many applications to the semantics of non-classical propositional logics, possibilities also provide an alternative semantics for classical first-order logic which arguably presents some conceptual and foundational advantages over standard Tarskian semantics. In this talk, I will review some recent results on the model theory of possibility structures for first-order logic, drawing connections to several key ideas and insights from Van Benthem's 1981 manuscript "Possible Worlds Semantics for Classical Logic”.

Mirac Suzgun

Sunday, June 2      12:30 - 13:00

On the Epistemological Failures of Modern Language Models

Sanjit Seshia

Sunday, June 2      14:30 - 15:15

Oracle-Guided Learning of Logic Specifications and Programs

Formal verification and verified program synthesis rely on having a mathematical specification of correctness, typically in a suitable logic. In this talk, I will discuss the problem of learning formal specifications from multi-modal inputs, including examples, counterexamples, demonstrations, etc., with a particular focus on temporal logics. I will discuss a framework called oracle-guided inductive synthesis (oracle-guided learning) that captures most common approaches to specification learning. I will further discuss how the paradigm of oracle-guided learning can be extended to verified program synthesis, including with large language models.

Stuart Russell

Sunday, June 2      15:15 - 16:00

Unifying logic and probability: The BLOG language

Logic and probability are ancient subjects whose unification holds significant potential for the field of artificial intelligence. The BLOG (Bayesian LOGic) language provides a way to write probability models using syntactic and semantic devices from first-order logic. In modern parlance, it is a relational, open-universe probabilistic programming language that allows one to define probability distributions over the entire space of first-order model structures that can be constructed given the constant, function, and predicate symbols of the program. I will describe the language mainly through examples and cover its application to monitoring the Comprehensive Nuclear-Test-Ban Treaty.

Josh Rule

Sunday, June 2      16:00 - 16:45

Efficient learning of rule-based concepts via metaprogram search

Humans acquire a variety of logically and algorithmically rich rules, often given minimal data and computation. Symbolic models based on program learning successfully explain rule-learning in many domains, but performance degrades quickly as program complexity increases. I will describe how symbolic search over the space of metaprograms—programs that revise programs—dramatically improves learning. On a behavioral benchmark of 100 algorithmically rich rules, this approach fits human learning more accurately than alternative models while also using computation consistent with conservative estimates of human resources during learning. Metaprogram search may thus help scale symbolic rule-learning methods to model human performance in challenging domains.