Ignoring the title and main purpose of the paper (29 page PDF) for a moment, I encourage people to look at the first section for a really useful discussion of the various types of models that have been proposed through the years. Models might "bear observable and pre-theoretic similarities to their target systems." Or maybe they "are structures that express interpretations of a formal theory." Or maybe they " are structures that bear a formally defined relation of isomorphism to their target system," as Bas van Fraassen suggested in 1980. Whatever your definition (semantic, syntactic, or other) "models are epistemically more central to science than theories."
All of this becomes really important if you think of minds as being (or containing) models of knowledge, the world, or whatever, whether they are representations, reflections or whatever. This in many ways is the core element of constructivism, as it recognizes the central role played by models, and focuses on ways to build them and manipulate them (as opposed to instructionist theories, which simply stack fact upon fact until you have a big pile of knowledge). Anyhow, what makes modern AI (and modern theories of learning) different (in my view) is that some of the models they propose are not representational; the models are connective structures that may produce a characteristic output (ie., recognition) but not an interpretation either of phenomena nor of formal theory. Image: Kotu & Deshpande.
Today: 2 Total: 7 [Direct link] [Share]
