Artificial Neural Networks (ANNs) are large collections of interacting entities. Certain conditions of behavior and of nonlinear coupling between the entities enable self-organization of the system with emergent properties of associative memory, abstraction and generalization. Therefore, it should not be surprising that the mathematical tools of irreversible thermodynamics and evolution are very relevant and useful in the analysis of these interesting "computational networks". The behavioral resemblance between ANNs and certain dynamical physical systems is not coincidental.