Parasitic Pedagogy: Learning Lessons from the Common Cold
How might the structure of a learner's environment affect its ability to successfully learn? If we could answer this question, perhaps we could structure learning environments in helpful ways, exposing learners to specially chosen learning trials - those most likely to result in genuine learner improvements. This approach immediately confronts the problem of how to characterise and exploit what Vygotsky would call the learner's zone of proximal development - how are we to know, at any point in time, which learning experience would be most likely to facilitate successful learning? The research presented here addresses these issues obliquely from within the paradigm of evolutionary computation - a class of search algorithms modelled loosely on evolution by natural selection. More specifically, we explore coevolutionary algorithms, in which populations are pitted against each other as predator and prey, or parasite and host. Are there ways of designing these algorithms such that one coevolving population (the "teachers") consistently encourages the other (the "learners") to improve towards some target level of performance? In nature, the relationships between parasites and their hosts vary in many ways. We will concentrate on one aspect - virulence. While some parasites are deadly (e.g., killing diseases) others are merely a mild irritant (e.g., the common cold). This variation exists for a reason. Some parasites require their hosts to remain alive in order to successfully reproduce, others are able to spread effectively from a host's corpse and have less interest in keeping their temporary home alive. In contrast, artificial parasites are almost always maximally virulent - perhaps this is thought to exert "maximum selection pressure" on the host population. Could it be that artificial parasites might serve our interests better if they were less than maximally virulent? We examine how varying parasite virulence in an artificial coevolutionary system influences the ability of the system to encourage successful learning by automatically structuring the learning environment. Some tentative pedagogical implications are drawn.