Pedagogical agents for social music learning in Crowd-based Socio-Cognitive Systems

This paper considers some of the issues involved in building a crowdbased system for learning music socially in communities. The effective implementation of building such systems provides several fascinating challenges if they are to be sufficiently flexible and personal for effective social learning to take place when they are large number of users. Based on our experiences of building the infrastructure for a crowd-based music learning system in Goldsmiths called MusicCircle we address several some of the challenges using an agent based approach, employing formal specifications to articulate the agent design which can later be used for software development. The challenges addressed are: 1) How can a learner be provided with a personalised learning experience? 2) How can a learner make best use of the heterogenous community of humans and agents who co-habit the virtual learning environment? We present formal specifications for an open learner model, a learning environment, learning plans and a personal learning agent. The open learner model represents the learner as having current and desired skills and knowledge and past and present learning plans. The learning environment is an online platform affording learning tasks which can be carried out by individuals or communities of users and agents. Tasks are connected together into learning plans, with pre and post conditions. We demonstrate how the personal learning agent can find learning plans and propose social connections for its user within a system which affords a dynamic set of learning plans and a range of human/ agent social relationships, such as learner-teacher, learner-learner and producer-commentator.