This paper defends an internalist perspective of selection based on the hypothesis that considers living evolutionary units as Maxwell's demons (MD) or Zurek's Information Gathering and Using Systems (IGUS). Individuals are considered as IGUS that extract work by means of measuring and recording processes. Interactions or measurements convert uncertainty about the environment (Shannon's information, H) into internalized information in the form of a compressed record (Chaitin's algorithmic complexity, K). The requirements of the model and the limitations inherent to its formalization (...) are discussed. This approach offers an alternative view to the causes of evolutionary variations which goes beyond the classical Lamarckian-Darwinian controversy. I argue that random variations only apply near-to-equilibrium at the time organisms have attained structural closure, and that a speed up of mutation rates that facilitates the production of directed variations occurs far-from-equilibrium due to organisms' openness to the surrounding conditions. However, real organisms are located somewhere between the above two cases and thus, operate at an intermediate stage where there is a maximum efficiency of H/K conversion. In consequence, IGUS keep their autonomy and evolving capacity by compromising between external circumstances and inner constraints. This compromise is made possible by closure regulation. Likewise, this model explains why nature has favored the selection of agents capable of selectively recording a partial description of their environment. (shrink)