How can one model growth such that, in general, poor countries and rich countries do not converge in average income, but some poor countries can have Korea and Taiwan-style growth miracles? Lucas contends that on-the-job learning is critical. In World War II, a fantastic natural experiment on ship construction shows learning-by-doing improving productivity initially as much as 40% per year – though it is an open question whether this reflects learning-by-doing embedded in a worker, in managers, or in the aggregate firm. Since reasonable models of learning-by-doing show the effect going to zero as workers become more productive, growth miracles must involve nations shifting production into higher and higher technologies, gaining new learning-by-doing effects each time. This model suggests why already-rich countries do not have growth miracles (no higher tech to move to) and why import substitution is ineffective in the long run (in autarky, firms produce for domestic consumption, and therefore can only move up the value chain at the rate at which domestic consumption demand changes – there are no Korea-style export contests to be had). This paper contains a discussion of the early learning-by-doing literature as well (Stokey 88, Young 91, Helpman-Grossman 91).