Annual Lecture Series (in-person)
Lunchtime Talks (in-person)
- This event has passed.
ALS: C. Buckner
April 22 @ 3:30 pm - 5:30 pm EDT
Cameron Buckner, Department of Philosophy, University of Houston
The new DoGMA of Empiricism: Deep Learning and Domain-General Faculties
Abstract: Deep learning is a research area in computer science that has over the last ten years produced a series of transformative breakthroughs in artificial intelligence—creating systems that can recognize complex objects in natural photographs as well or better than humans, defeat human grandmasters in strategy games such as chess, Go, shoji, or Starcraft II, create “deepfake” images or bodies of text that are sometimes indistinguishable from those produced by humans, and predict how proteins will fold more accurately than human microbiologists who have devoted their lives to the task. In this talk, I argue that these achievements were inspired by and in turn vindicate a moderately empiricist approach to cognition. However, this particular form of moderate empiricism has not been carefully articulated and defended in the computer science it has inspired. This has allowed deep learning’s rationalist critics—such as Gary Marcus, Melanie Mitchell, and Judea Pearl, as well as (in the previous generation of neural network criticism) Steven Pinker, Jerry Fodor, and Zenon Pylyshyn—to define the rules of the debate on their own terms. In this talk, I rebut this caricature by explicating the empiricism which undergirds most of these achievements. In particular, I interpret them in terms of a Domain-General Modular Architecture (DoGMA) for cognition centered on active, general-purpose, innate faculties like perception, memory, imagination, attention, and empathy. This empiricism echoes not the “radical” forms associated with behaviorism and logical positivism, but rather the more plausible forms defended by empiricist-leaning faculty psychologists such as Aristotle, Ibn Sina (Avicenna), John Locke, George Berkeley, David Hume, Adam Smith, and Sophie De Grouchy. If the debate between rationalism and empiricism is to remain a useful framework for critique of deep learning—even for rationalists who might still think it ultimately wrong-headed—we must stop framing it as a sequel to the oft-recited fable in which the valiant young Chomsky vanquished the dark lord Skinner, and see it instead as a novel empirical exploration of the most ambitious ideas offered by historical faculty empiricists.
To attend this lecture via Zoom please visit: https://pitt.zoom.us/webinar/register/WN_M_dQJZZaRS6oK371bkcE1A
Please Note: Non-Pitt individuals who want to attend our in-person talks must send an email in advance to Shoshi Burd-Baugh (email@example.com) requesting Guest Building Access, or you will not be able to enter the Cathedral of Learning.