with Dr. Alex Broadbent
Tuesday, March 3, 2020
12:30 – 2:00 PM
5 Washington Place, Rm 101
New York, NY, 10003
Lunch will be served.
Abstract: Machine learning (ML) appears to offer near-magical abilities to make predictions from large, messy data sets, without the usual headaches associated with (or caused by?) working out which exposures are associated with, or cause, the outcome or outcomes of interest. This is a shock for epidemiology, since applications of ML to population and public health are still relatively new and rare. It threatens epidemiological expertise, which appears not to be necessary for the machines. And it threatens epidemiological methodologists, many of whom insist that causal inference is an essential prerequisite for good predictions, and in particular for effective interventions. Advocates of the potential outcomes approach (POA) are especially vocal in this regard. This talk explains the potential and limitations of ML for epidemiological research, and what is needed to realise the potential benefits for public health. It advocates moving beyond the “catalogue” approach of repeatedly checking whether associations are causal – which, from a philosophical perspective, is neither necessary and sufficient for acting on a prediction. Instead it draws a distinction, familiar from philosophy of science, between context of discovery and justification, and argues that, for justification, a wider theoretical understanding of a domain must be built up, using both descriptive and analytic methods.
Speaker Bio: Dr. Alex Broadbent is Professor of Philosophy, Executive Dean of Humanities, and founding Director of the African Centre for Epistemology and Philosophy of Science at the University of Johannesburg. He has interests in philosophy of science, medicine, epidemiology, and law.