We introduce a network formation model based on the idea that individuals engage in production (or decide to participate in an action) depending on the similar actions of the people they observe in the society. We differentiate from the classical models of participation by letting individuals to choose, non-cooperatively, which agents to observe. Observing behavior of others is a costly activity but provides benefits in terms of reduction in cost of production for the observing agent, which we take it as learning. In this non-cooperative setting we provide complete characterization of both Nash stable and socially efficient network configurations. We show that every society can admit a stable network. Moreover, typically there will be multiple stable configurations that will be available for a society. While all stable networks will not be efficient, we show that every efficient network will be stable.