Combining Human and Machine Intelligence

An extensive literature in psychology, economics, statistics, operations research and management science has dealt with comparing forecasts based on human-expert judgment vs. (statistical) models in a variety of scenarios, mostly finding advantage of the latter, yet acknowledging the necessity of the former. Although computers can use vast amounts of data to make predictions that are often more accurate than those by human experts, humans are still more adept at processing unstructured information and at recognizing unusual circumstances and their consequences. Can we combine predictions from humans and machines to get predictions that are better than either could do alone? We used prediction markets to combine predictions from groups of people and artificial intelligence agents. We found that the combined predictions were both more accurate and more robust in comparison to those made by groups of only people, or only machines. This combined approach may be especially useful in situations where patterns are difficult to discern, where data are difficult to codify, or where sudden changes occur unexpectedly. Thesis Supervisor: Thomas W. Malone Title: Patrick J. McGovern Professor of Management, MIT Sloan School of Management; Director, MIT Center for Collective Intelligence