In this work we present some computational considerations about the nature of concepts. After a first distinction between "per se" concepts and the epistemic or psychological aspects of concepts we focus our attention on this latter by linking their understanding to the formalization of automatic classification problems as developed in machine learning field. We show how the problem of categorization and of formation of concepts representing categories, has to be considered a computationally intractable problem that hence can be faced only with heuristic strategies. In this perspective the different cognitive theories proposed in cognitive science to explain categorization processes (e.g. the prototype-theory by E. Rosch), can be considered as different heuristic strategies (computationally tractable solutions) to face the categorization problems and that the human mind uses in order to prevent an unsustainable cognitive-computational load. Finally, we frame these ideas in the cognitive naturalism and in the current viewpoint that considers the most of the human reasoning as heuristic solutions to intractable problems. Our thesis is that concepts can be considered as heuristic and “perspective” solutions that any intelligent system, such as human mind, finds in order to represent categories —using limited resources and capacities— by which it organizes and gives a sense to the large variety of reality that surrounds it.