Consequential decisions often have unique features, requiring human judgment and discrimination. When people disagree, one needs a method of resolution, either by formal algorithm or by group discussion. An advantage of algorithms is that they can involve many dispersed individuals and preserve their independence. Prelec will describe an aggregation approach based on a new definition of the best answer to a non-verifiable question, as the one given by those who would be “least surprised” by the true answer if that answer were revealed. Since this definition is of interest only when the true answer is unknown, algorithmic implementation is nontrivial. A solution can be constructed using a secondary input, namely, respondents’ predictions of how other people will respond.
 
Based on joint work with Sebastian Seung (Princeton) and John McCoy (MIT)