Hajo Greif (Technical University of Munich)
There are various ways in which models and simulations in Artificial Intelligence (AI) are designed to relate to cognitive phenomena. In order to get a systematic grasp of these approaches, a taxonomy is developed here. It is based on the coordinates of the distinctions between formal and material analogies and theory-guided and pre-theoretic models in science. These distinctions have relevant analogies in the computational versus mimetic aspects of simulations as well as in the distinction between analytic, mathematical and synthetic, exploratory types of computer simulation.
On this conceptual background, one can distinguish between at least four approaches that figured prominently in the early and classical stages of AI, and have partly developed into distinct research programmes: first, purely phenomenal, theory-free simulations (e.g., Turing's "imitation game"); second, simulations that are used to explore general-level isomorphisms between computational processes and patterns of reasoning (e.g., logic-based AI); third, simulations as material models, where computational concepts directly inform theoretical hypotheses concerning cognitive processes (e.g., Marr's stages of visual processing and connectionism); and fourth, simulations as strictly formal models, where computational processes are taken to be expressions of the propositions of a theory of cognition (strong symbolic AI).
In continuation of pragmaticist views of the modes of modelling and simulating world affairs (Humphreys, Winsberg), this taxonomy of approaches to modelling in AI will serve to elucidate how available computational concepts and simulational resources contribute to the modes of representation and theory development in AI research – an what made that research programme uniquely dependent on these resources.
