Coevolutionary theories describe the probability distribution of interacting proteins in terms of a Boltzmann statistical model. As a result of selective pressures, that distribution is expected to sharply deviate from uniformity by featuring a relatively small number of highly probable sequences across the entire sequence space. While that statement must be true for interolog systems in general, their sequence distributions may have not been fully shaped by selective pressures opening the possibility that novel interolog sequences could be selected from artificially generated lower entropy distributions. The goal of this work was to investigate the physical meaning of selected sequences from artificial fitness landscapes. For that, we explore a Genetic Algorithm, which solves the distributions by maximizing the statistical coupling, starting from the native multi-sequence alignments and exploring the space of scrambled multi-sequence alignments. We also solve a distribution by minimizing statistical couplings through random shuffling of multiple sequence alignment. Once likely artificial sequences are selected from maximized and minimized distributions, their binding free-energies at a fixed native bound state are evaluated according to free energy calculations based on the MM/PBSA method. To evaluate the physical meaning of native and artificial sequences, we calculated the selection temperature in relation to random sequences of the same composition. Our results indicate that it is possible to select new non-similar artificial sequences at colder or warmer selection temperatures than the native selection temperature, and that the selected artificial sequences show differences only in specificity, but not in relation to binding affinity. It is possible to conclude that the molecular evolution of the interaction of non-obligate dimers can be restricted only by specificity, since the affinity must only be a consequence of the amino acid composition determined by the folding restrictions. In addition, it is possible to find new protein-protein interactions with characteristics equal to or better than the interactions existing in nature.