Resource discovery is an important problem in unstructured peer-to-peer networks as there is no centralized index where to search for information about resources. The solution for the problem is to use a search algorithm that locates the resources based on the local information about the network. Efficient data sharing in a peer-to-peer system is complicated by uneven node failure, unreliable network connectivity and limited bandwidth. A well-known technique for improving availability is replication. If multiple copies of data exist on independent nodes, then the chances of at least one copy being accessible are increased. Replication increases robustness. In this paper, we present a novel technique based on Q-learning for replicating objects to other nodes.