An application of choice under ambiguity to social learning. In traditional social learning models, there is often as assumption that past individual choice was essentially at random; that is, the realized outcomes of individuals choosing some action x is the same as the counterfactual outcomes that individuals who chose something else would have achieved had they chosen x. Manski applies results from the partial identification literature, showing that even if individuals do not make this choice-at-random assumption, but only assume some stationarity to the distribution of outcomes conditional on actions, then the identification region of non-dominated actions will nonetheless shrink over time: individuals will learn. The actual action they choose will depend on their rule for decisionmaking under uncertainty. An application to the acceptance of new innovations shows that, depending on this rule, a new innovation could be accepted quickly then decline, or be accepted slowly but rise over time, or many other possibilities. Without more experimental knowledge of choice under ambiguity, it is not possible to predict which of these rates of acceptance is more likely.