about my research, generally: I use formal methods to address questions in epistemology, philosophy of science, and the evolution of human behavior and cognition. Much of this work uses the framework of self-assembling games to model the emergence of social, behavioral, and cognitive phenomena from more basic learning and evolutionary processes. All of it has to do with learning, in one way or another.
about the dissertation, specifically: There is a tradition at least as old as David Hume's treatment of the "problem of induction" that understands our inductive beliefs and methods primarily as products of natural adaptive processes. Today, we have powerful scientific resources for studying the natural foundations of our inductive practices. On the empirical side, we have the results of over a century of laboratory experiments on human and animal learning. On the theoretical side, we have well-understood formal models of learning and evolution from biology, psychology, and economics. My work marshals these resources to consider how successful methods of learning might emerge from more basic evolutionary and learning processes. The strategy is to take basic components of empirical inquiry (e.g. inductive biases, learning rules, linguistic devices) and consider how they might arise out of simple trial-and-error processes like reinforcement learning and evolution by natural selection. I call this the self-assembly of learning. My dissertation focuses on self-assembly with respect two basic aspects of learning: first, projectibility and, second, the explore-exploit tradeoff. The guiding questions are: How might agents learn or evolve to track the regularities in their environments that are relevant for learning and action? and How might agents learn or evolve to balance exploration of untested options against exploitation of options that have succeeded in the past? Material from (2), (3), (4), and (5) below appears in the dissertation.
(1) Janina Hosiasson and the Value of Evidence
Studies in History and Philosophy of Science
shows that I.J. Good's classic result concerning the pragmatic value of learning is prefigured in a 1931 paper by Janina Hosiasson
(2) Learning How to Learn by Self-tuning Reinforcement
with Jeff Barrett, Synthese
develops a model of self-tuning reinforcement learning that captures a well-known experimental finding on learning to learn in rhesus macaques
(3) Learning to Forget
with Jeff Barrett, forthcoming in British Journal for the Philosophy of Science
considers how reinforcement learners in Lewis-Skyrms signaling games might learn to adopt methods of learning well-adapted to the communicative problems they repeatedly face
(4) Learning in Crawford-Sobel Games
with Jeff Barrett, Cailin O'Connor, and Brian Skyrms, forthcoming in British Journal for the Philosophy of Science
bridges a gap in the game theoretic literatures on communication in philosophy and economics by modeling simple forms of trial-and-error learning in Crawford-Sobel games and comparing findings to related results for Lewis-Skyrms signaling games
(5) Task-Switching and Natural Projectibility
under review
articulates a version of Goodman's "new riddle of induction" that applies very generally to learners in nature; develops a model showing how that problem might be solved by means of a simple form of reinforcement learning in the context of an experiment on task-switching in rhesus macaques
(6) Learning as a Mechanism in the Evolution of Fairness
under review
uses the "indirect evolutionary approach," developed in economics, to model the evolution of inequity aversion in the context of the Nash demand game
(7) Alexander Bain on Learning and Control
draft available on request
explicates the theory of learning underlying Bain's account of the development of voluntary control; argues that the account presages important later developments in the psychology of learning