This paper introduces three protocols that define Social Learning on Jason, the Java-based implementation of AgentSpeak(L). The implementation of these protocols is based on the Instructive stage of a classification correlated with the developmental stages of a child. Agents defined in these protocols are able to share both the inputs and the output of a learning algorithm in order to improve the Multi-Agent System (MAS) performance. Compared with environments without social interaction, the performance of the MAS improves when any of these protocols is implemented. The results shown in this paper open the door to Social Protocols based on Collaborative Learning.