This is not completely new. The “empirical turn” of economics has been well documented for quite a while now. It is explained by a conjunction of factors: the low- hanging fruit of economic theory has largely been picked, access to large datasets has been democratized, and computational power has been exponentially increasing. These factors have contributed to raising the opportunity costs of theoretical innovation to such a level that it now hardly pays off for junior economists to specialize in theoretical economics. However, what Cowen is describing is more than that. He’s describing a process whereby economists are progressively forgetting the marginalist roots of their discipline, roots that are still taught in introductory economics classes. It is a process in which economists are substituting machine-learning methods to build incredibly complex models (up to 360,000 factors, as Cowen documents) for economic—marginalist— reasoning based on basic concepts such as supply and demand and opportunity cost. In a way, we seem to be back to the old “measurement without theory” debate of the 1940s, but in a radically different context with more access to data and more computing power.