Prof. Ludwig examines how machine learning can be used to improve human decisions. Using data from a large American city, he shows how an algorithm was trained to predict defendants' future behavior, and concludes releasing defendants using the predictions of an algorithm can achieve less crime and fewer people detained in jail, while reducing racial disparities. He discusses one key to this analysis - overcoming a censoring problem: we do not observe what jailed defendants would have done had they been released. One methodological implication is that adapting the techniques of machine learning to this domain must be a joint activity between the design of prediction algorithms and the development of an economic framework that focuses on payoffs, decisions and selection biases.
This event is free and open to the public. Registration is required.