A Minimal Model of the Interaction of Social and Individual learning
Social, supervised, learning from others might amplify individual, possibly unsupervised, learning by individuals, and might underlie the development and evolution of culture. We studied a minimal model of the interaction of individual unsupervised and social supervised learning by interacting agents. Agents attempted to learn to track a hidden fluctuation "source", which, linearly mixed with other masking fluctuations, generated observable input vectors. Learning was driven either solely by direct observation of inputs (unsupervised, Hebbian) or, in addition, by observation of another agent's output (supervised, Delta rule). To enhance biological realism, the learning rules were made slightly connection-inspecific, so that incorrect learning sometimes occurs. We found that social interaction can foster both correct and incorrect learning. Useful social learning therefore presumably involves additional factors some of which we outline.