Britt Anderson

Contact Information

Office:  E7 6328 / PAS 4039
Phone: 519-888-4567 x43056
Email: britt@uwaterloo.ca

Appointed to Psychology in Cognitive Neuroscience

Education

PhD, Brown University (2006)
MD, University of Southern California (1984)
BS, University of Arizona (1980)

Research

Led by the work demonstrating that a lot of what is called attention can be more directly accounted for in terms of stimulus probabilities, I have become (writing in 2021) pre-occupied with how it is we represent uncertainty. It is relatively easy to say that our degrees of belief are simply probabilities, but that can't be entirely right for a lot of reasons. We seem to have beliefs about events that have never happened and never could happen. That is why people can argue about counter-factual alternate histories. And even for more common circumstances the events we are making decisions for are non-repeatable. Deciding to have a first date with person X is not something you can model based on frequencies. But you do decide, and you have feelings that are stronger or weaker about whether it is a good idea. In all the Bayesian and similar accounts of decision making under uncertainty that are on offer to account for these types of decisions there are probability distributions embedded. Do we think those distributions (or some simplification) are generic or are they rather modality and situationally specific? How do we compare the similarity and differences of distributions when deciding if things have changed? Measuring belief is a tricky technical issue, and I do work on that, but also the lab tries to think more conceptually about what this space of representation for uncertainty looks like and what are the calculi for comparison. How can they be instantiated in neural hardware?

You can't read papers with all these different kinds of models without wondering in what fundamental ways they are different. The language may imply that two approaches are very similar (or very different), but the wording used to describe can be a long way from the mathematics that underlies them and both may not be reflected in the code written to simulate them. How can we do a better job of communicating our models and theories formally? What is the right language to facilitate comparisons and contrasts? I am a new enthusiast to applied category theory and am looking for my first opportunity to use it in some meaningful way. As probability theory can be developed from a category theory perspective (in contrast to the traditional measure theoretic one) I am hopeful there may be a way to braid these two strings together (there is a pun in there for any category theorists that find their way here).