Artificial Government uses meta-learning and unsupervised learning to coordinate agents in heterogeneous multi-agent systems. It acts as a mediator between two classes of agent: producers and consumers. Producers produce solutions to problems in some environment. Consumers can reuse these solutions for their own purposes. Artificial Goverment proposes a new class of agent: the mediator. It uses feedback from the consumers to direct the producers toward solutions that will better satisfy the consumers. The thesis describes the components of Artificial Government and an experiment that demonstrates its effectiveness in a multi-agent reinforcement learning domain.