Networks and Conditional Reasoning

We will deal with the (i) representation of conditionals in neural-like networks and with the question (ii) how such networks are able to reason on the basis of these representations. One can prove that if the representation of conditionals in networks is "distributed", each of the systems of nonmonotonic logic in Kraus, Lehmann, and Magidor (1990) turns out to be adequate with respect to a particular network semantics. We will interpret this result in the context of the debate on "symbolic vs. distributed representation", we will investigate which properties of the world, if any, are tracked by such conditional representations in networks, and we will end up with an open question about possible extensions of these results to learning networks and their description in terms of systems of inductive logic.