Friday, April 18, 2025 | 3:30-4:50pm ET | Baker Hall A36
Sarah Moss, University of Michigan

Talk Title: Justification and Grounding

Abstract: Bayesians have had a lot to say about justification. Bayesians have endorsed claims about when evidence gives you justification to believe a hypothesis, and some endorse particular measures of how much justification it provides. Bayesians defend various theories of why hypotheses can be justified by old evidence. In more recent literature, some theorists have developed Bayesian accounts of when justification for a hypothesis counts as being defeated by further evidence. Unfortunately, many of these theories are false. Bayesians have failed to appreciate that propositional justification involves grounding. As a result, some Bayesians conflate relations of justification with relations of probabilistic dependence.

This talk offers a course correction. After discussing examples that serve to sharply distinguish propositional justification from probabilistic dependence, I argue that a number of significant epistemic relations involve grounding, including relations that hold when evidence supports, gives you justification to believe, or gives you reason to believe a hypothesis. Then I apply my account of justification to some existing debates, developing problems for recent Bayesian accounts of epistemic defeat and for several classic responses to the problem of old evidence. Finally, I develop a novel measure of the justification that evidence provides for a hypothesis, where this measure can help us address counterexamples for existing theories of justification, defeat, and old evidence.

Friday, October 25, 2024 | 3:30-4:50pm ET | Baker Hall A36
Simon Huttegger, University of California, Irvine

Talk Title: Inductive Logic: Its Philosophy and Contemporary Significance

Abstract: Rudolf Carnap spent much of the last 25 years of his career developing an inductive logic: a logic of reasoning from the known to the unknown that is derived from first principles and, at the same time, faithful to how scientists evaluate hypotheses and make predictions based on observations. In my talk, I will review what I take to be the main contribution of Carnap's inductive logic. I will then connect it to developments in Bayesian statistics, in particular probabilistic symmetries and invariance principles and developments in predictive inference, and suggest ways in which it can be enriched to supply a more comprehensive account of scientific inference. I will end with some philosophical reflections on the kind of model of scientific reasoning that inductive logic gives rise to.

Friday, November 8, 2024 | 3:30-4:50pm ET | Baker Hall A36
Spencer Breiner, National Institute of Standards and Technology

Talk Title: Systems theory from the relative point of view, or what systems shouldn't not be

Abstract: What is a system, and why should we care? The first question is a surprisingly slippery, especially the issue of whether systems (and related features like boundaries) exist "in the world" or "in our heads". I won't try to pin down an answer here, but I will introduce a mathematical perspective of systems theory that sheds some light on this issue, and use it to question a prominent extant definition. We start from the idea that systems can be probed by "local models", partial characterizations that emphasize certain aspects and suppress others. Consequently, systems are modeled by systems of models. These models are bound together through overlapping architectures and processes (shared syntax) as well as common data and parameters (shared semantics). Furthermore, I will argue that certain features of the structural approach to systems theory, particularly Grothendieck's "relative point of view", speak to the second question, offering some hope that better definitions of systems and their attendant concepts might help to defray the ever-increasing cost and complexity of building and maintaining systems in the modern world.

Bio: Spencer Breiner is a research scientist at the National Institute of Standards and Technology (NIST) and a graduate of the CMU Department of Philosophy (Ph.D. 2013).

Friday, November 22, 2024 | 3:30-4:50pm ET | Baker Hall A36 Adamson Wing
Kun Zhang, Carnegie Mellon University

Talk Title: Revealing the Hidden Causal World

Abstract: Our long-term goal is to develop a platform for automatic scientific discovery that takes all available measured data as input and outputs hypotheses regarding hidden entities and causal influences. This platform aims to continually expand human knowledge and make us more resourceful. How can we achieve this? Causal discovery (or causal representation learning) involving causally-related hidden variables plays a pivotal role. In this talk, I will report our recent algorithmic advances in this endeavour, along with their applications to real problems in psychology, machine learning, and computer vision. I will demonstrate how the “modularity” property of causal systems, paired with suitable simplicity assumptions, makes it possible to recover the underlying causal process from observational data with identifiability guarantees: under appropriate assumptions, the learned representations are consistent with the underlying causal process.