Lectures & Colloquia
Friday, March 18
Mikhail Belkin, University of California San Diego - Philosophy Colloquium
Talk Title: Fit without fear: remarkable phenomena of deep learning through the prism of interpolation
3:05 – 4:25pm EST via Zoom
Abstract: In the past decade the mathematical theory of machine learning has lagged far behind the triumphs of deep neural networks on practical challenges. However, the gap between theory and practice is gradually starting to close. In this talk I will attempt to show some of the new insights into the nature of generalization brought to light by the practice of deep learning. The key theme of the talk will be the idea of interpolation, fitting the data exactly, a counter-intuitive notion from the point of view of classical statistics. As we will see, just as a physical prism separates colors mixed within a ray of light, the figurative prism of interpolation helps to disentangle properties within the complex picture of modern Machine Learning.
Friday, February 11
Prof. Quill Kukla, Georgetown University - MAP (Minorities and Philosophy chapter at CMU) Talk
Talk Title: The Pragmatics and Ethics of Gender Ascriptions
Abstract: What are we doing when we ascribe gender---when we announce, "I am a man," tell someone "You are not a woman," or use gendered pronouns to describe someone? And why does such gendered language matter? What should people care what pronouns are used to refer to them? In this paper, I give a pragmatic analysis of the structure of gender ascriptions. I argue that gender ascriptions are not first and foremost truth claims of any sort, but rather serve a different primary pragmatic function. I try to show why gender ascriptions, including pronoun uses, are ethically and politically important, including why it is harmful to ascribe to someone a gender that they reject. My more specific goal is to understand why gender ascriptions matter by understanding how they function at the level of linguistic pragmatics. I show that first, second, and third personal gender ascriptions have quite different pragmatic forms and functions, and that the ethics of their utterance are likewise quite different. I argue that first personal gender ascriptions are almost always entitled and deserve respect, and that contradicting someone's gender self-ascription is a more serious form of social violence than has previously been appreciated by philosophers.
Friday, January 28
Sabine Hossenfelder, Research Fellow at the Frankfurt Institute for Advanced Studies
Talk Title: Problems in the Foundations of Physics
(note the time has been changed from the usual 3:05pm start in order to accommodate Sabine’s time zone)
Abstract: In the foundations of physics, we have long-standing open problems such as the nature of dark matter and dark energy, the hierarchy problem, or the missing quantization of gravity. Yet physicists have not made progress on solving those problems for 50 years. In this talk, I discuss how physicists ended up in a vicious cycle of theorizing and null results, and what we can do to overcome the present phase of stagnation.
Friday, October 8
Wesley H. Holliday, University of California, Berkeley - Philosophy Colloquium
Talk Title: Splitting Cycles, Breaking Ties
Remote via Zoom
Abstract: In this talk, I will report on joint work with Eric Pacuit aimed at solving the problems for democratic voting posed by Arrow’s Impossibility Theorem, the Paradox of Voting, and related results. We diagnose what went wrong with Arrow’s assumptions, explain how to resolve cycles in the majority relation of an election in arational way, and how to fairly break ties should they arise. Logical methods, including SAT solving and interactive theorem proving, have been employed in this research, as well as computer simulations. Finally, I will demo a website that makes our new voting methods available to all for practical use.
Friday, October 29
Teddy Seidenfeld, Herbert A. Simon University Professor of Philosophy and Statistics, Carnegie Mellon University - Philosophy Colloquium
Talk Title: Strategic aspects of coherent decision making: some challenges for interpreting “rational” behavior
based on joint work with M.J.Scherivsh and J.B.Kadane: Exchange Rates, Phil. Sci. (2013) 80: 504-532
Remote via Zoom
Abstract: A common view among decision theorists is that if an agent is disposed to choose in accord with any one of several, well known normative theories of Subjective Expected Utility [SEU], then those choices have a unique representation in terms of a probability-utility pair, where: probability quantifies the agent’s degrees of belief over states of uncertainty, and utility quantifies the agent’s values for outcomes.
In this talk I discuss challenges to that view based on strategic decision making that is underwritten by the same normative SEU theories.
To illustrate the challenges, I use de Finetti’s two, seminal frameworks:
- coherent pricing of variables (a generalization of coherent betting), and
- coherent forecasting of variables.
These two frameworks are linked by a common principle of dominance, which forms the core of his criterion of rational coherence.
Strategic pricing is a familiar problem for de Finetti’s attempt to give an operational treatment of personal probability. Where the decision maker gives strategic prices, those may prevent identifying the agent’s degrees of belief, as I will review. This concern was important for de Finetti’s adoption of his second framework: forecasting variables subject to Brier-score. Because Brier-score is a strictly proper scoring rule, that is thought to be a sufficient reason for the absence of strategic forecasting.
We will see why that conclusion is not correct, and why strategic forecasting creates at least as serious challenges for identifying the agent’s degrees of belief and values as are created by strategic pricing.
Friday, November 12
Christian List, Professor of Philosophy and Decision Theory & Co-Director, Munich Center for Mathematical Philosophy, LMU Munich - Philosophy Colloquium
Talk Title: Special-science counterfactuals
Abstract: On the standard analysis, a counterfactual conditional such as “If P had been the case, then Q would have been the case” is true in the actual world if, in all nearest possible worlds in which its antecedent (P) is true, its consequent (Q) is also true. Despite its elegance, this analysis faces a difficulty if the laws of nature are deterministic. Then the antecedent could not have been true, given prior conditions. So, it is unclear what the relevant “nearest possible worlds” are. David Lewis suggested that they are ones in which a local breach of the laws occurred: a “small miracle”. Others have suggested that they are ones in which the initial conditions were different (“backtracking”). I propose another response. It builds on the idea that the special sciences, where counterfactual reasoning is most common, operate at a higher level of description from fundamental physics, and that the world may behave indeterministically at higher levels even if it behaves deterministically at the fundamental physical one. The challenge from determinism can then be bypassed for many special-science counterfactuals