Within the last decade or so theories of inductive learning in games have increasingly become the primary approach in the construction of models for explaining how agents may resolve repeated coordination problems as well as the emergence social conventions at the more general level. However, looking closer at a paradigm case of such models, the Dirichlet model, this paper argues that such models only work for explaining emergence if presupposing pre-tailored and ad hoc conceptualizations of the recurrent decision problem faced by the agents. It then argues that such conceptualization itself rest on convention and thus that the models only work by begging the question they were thought to answer. Finally, the paper points to the possibility that a non-circular solution to the problem of conceptualization may be found in an understanding of the way agents reason about coordination, when the idea of repeated coordination itself acts as a frame for conceptualizing their mutual decision problem as a coordination problem.
Logkca-10: Proceedings of the Second Ilcli International Workshop on Logic and Philosophy of Knowledge, Communication and Action, 2010, p. 227-254
Game theory; Coordination problem; Learning in games; Social convention; Spontaneous order