The present invention is directed to a method and apparatus for learning call routing information in a communication system. A switching architecture is presented. The switch may be implemented in a centralized architecture or a distributed architecture. In addition, the switch may be implemented in a variety of networks such as a circuit-switched network or a packet-switched network. The switch includes a policy server and a self-learning application server. The policy server includes routing information for routing calls across the switch. The routes for incoming and outgoing calls are analyzed as they pass through the switch. The self-learning application server uses artificial intelligence techniques and caching algorithms to learn new more-efficient routing paths or initial routing paths based on the incoming and outgoing calls. The self-learning application server then updates the policy server with the new learned routes.