Research

Concurrently, I am working on a book-length analysis of the distinction between rule and case-based reasoning. Recently it has been argued that while rule and case-based reasoning “should both be seen as special cases of a more general Bayesian learning framework,”1  there might be non-computational explanatory reasons why we might want to treat them as different reasoning methodologies. I explore in detail what these non-computational differences are and argue that case-based reasoning is a more fundamental cognitive heuristic than is rule-based reasoning.

The following two observations play a central role in motivating my excitement in this case-based/rule-based analysis:

1. There is a good deal of interest in many different communities – AI, linguistics, philosophy -- in the idea of analogical reasoning, how we use metaphors, and so on, but there isn’t much at the theoretical level that explains how this kind of reasoning connects up with other kinds of reasoning. Not surprisingly, we find that in analogical machine-learning systems, the active constraints are methodological and performance-based rather than theoretical. I suspect that a theory of case-based reasoning will be foundational for work in analogical reasoning since, I think, analogical reasoning is a special kind of CBR. Trying to create a system that reasons analogically using a rule-based approach is putting the cart before the horse, so to speak; because we have to figure out the constraints for the types of things that can be compared and the ways in which they can be compared before we know what we are interested in comparing, our rules will always fall short of our requirements in a way precisely analogous to the way in which RBR in general falls short in those domains in which CBR is better situated. In this sense, then, analogical reasoning is a special case of CBR.

2. My hunch is that an implicit appeal to these two fundamentally different ways of reasoning underlies the polarising debates in many different disciplines – linguistics, cognition, ethics – and that by making these commitments explicit, the debates might be resolved through discussion of these more fundamental issues. Ultimately, I’m interested in shedding light on the debate between the cognitivist and the embodied theorist in cognitive science, but I’ve begun this project by writing a paper on what I take to be the parallel debate in ethics between moral generalism and particularism.

Project

I’d like to test out the conclusions in this paper by developing two systems, one cased-based and one rule-based, both trained to make moral judgements. Anyone interested in doing collaborative work in this or a similar area should e-mail me.

1  Tenenbaum, J. 2000. Rules and Similarity in Concept Learning. In S. A. Solla, T. K. Leen, and K.-R. Muller, eds., Advances in Neural Information Processing Systems 12, 59–65. MIT Press.