Analysis of social learning strategies when discovering and maintaining behaviours inaccessible to incremental genetic evolution

It has been demonstrated that social learning can enable agents to discover and maintain behaviours that are inaccessible to incremental genetic evolution alone. However, previous models investigating the ability of social learning to provide access to these inaccessible behaviours are often limited. Here we investigate teacher-learner social learning strategies. It is often the case that teachers in teacher-learner social learning models are restricted to one type of agent, be it a parent or some fit individual; here we broaden this exploration to include a variety of teachers to investigate whether these social learning strategies are also able to demonstrate access to, and maintenance of, behaviours inaccessible to incremental genetic evolution. In this work new agents learn from either a parent, the fittest individual, the oldest individual, a random individual or another young agent. Agents are tasked with solving a river crossing task, with new agents learning from a teacher in mock evaluations. The behaviour necessary to successfully complete the most difficult version of the task has been shown to be inaccessible to incremental genetic evolution alone, but achievable using a combination of social learning and noise in the Genotype-Phenotype map. Here we show that this result is robust in all of the teacher-learner social learning strategies explored here.