Social learning is defined here as the imitation of behaviors exhibited by other members of a population. In a population of humans, we might be talking about memes, or more broadly, culture. The authors mention animal examples such as the way birds learn their songs, and more abstruse examples such as the spread of enzymes in a host population via contagion through symbiotic bacteria. The common thread is the relationship between host fitness for survival and a non-genetic process for acquiring a relevant attribute.

The paper presents an abstract model of a population in which the phenotype of an individual, that is, its set of expressed survival-related attributes, is simply a bit string that may probabilistically be a copy of either its own genotype or of some other individual’s phenotype. Genotypes are inherited from the parent. Genotypes and phenotypes are subject to mutations and copy errors, respectively. Selection for single-parent reproduction, and also survival, are probabilistic functions of the weight (number of ones) of the phenotype.

In discrete-time simulations with a constant population size, the total weights of phenotypes and genotypes each converge. If they converge to different values, social learning is said to have fixated, in the sense that a stable subpopulation of social learners has evolved. The extent to which that happens depends on the relative settings of a few parameters characterizing the probability distributions.

The authors are careful to place the model into perspective with prior work, motivate the design choices, and interpret the simulation results intuitively. The results support the novel conclusion that social learning can fixate without a specific decision-making capability for preferring some phenotypes. It is still difficult to judge how insights obtained from such a model apply in a given natural setting.