Automation, Partial and Full

Abstract

When some steps of a complex, multi-step task are automated, the demand for human work in the remaining complementary sub-tasks goes up. In contrast, when the task is fully automated, the demand for human work declines. Partial automatability of complex tasks leads to a bottleneck of development (where further growth is constrained by the scarcity of essential human work) which is removed once the tasks become fully automatable. Theoretical analysis using a two-level nested CES production function specification demonstrates that the shift from partial to full automation generates anon-convexity: humans and machines switch from complementary to substitutable, and the share of output accruing to human workers switches from an upward to a downward trend. This process has implications for inequality, the risk of technological unemployment and the likelihood of a secular stagnation.