Is Holism A Problem For Inductive Inference? A Computational Analysis

We investigate whether holism presents a problem for inductive inference by examining the relationship between the size of a Bayesian network that represents human conceptual knowledge and the computational complexity of probabilistic inference in that network. We find that, despite prior claims, holism may not be a problem for inductive inference, as computational cost does not increase exponentially as the network grows. While the network we analyze is holistic, it has a modular organization and grows in a way that potentially makes efficient inductive inference possible.