Kearns Speaks on Social Computation Research
Professor Michael Kearns detailed six years of his experiments in social computation last week when he came to CSAIL to speak as part of the annual Dertouzos Lecturer Series. Kearns, the founding director of the University of Pennsylvania’s new program in Market and Social Systems Engineering, explained that the overriding mission of his research is to study strategic and economic human interactions in social networks to help build a hybrid model for crowdsourcing.
“If we can imagine a theory of social computation, like the theory of computation, what would that look like?” said Kearns of the underlying goal of his research.
According to Kearns, there are currently no functional design principles that allow humans and machines to work together in an organized, efficient and beneficial manner. His group’s experiments, which brought different groups of individuals into his lab to perform computer tasks in physical isolation, examined whether the “network structure and nature of a task influenced individual performance and behavior.” The studies allowed Kearns and his colleagues to gather copious amounts of data on human behavior, to which they have applied machine learning techniques as a means of predicting how humans operate as part of a network.
Participants in Kearns’ studies had no physical contact with the other participants, but, through their computers, were allowed a local view of their neighbors. Through several rounds of experimentation, participants were asked to perform a variety of different tasks, from graph coloring to reaching to a unanimous consensus on a group color, with real financial incentive for solving the assigned problem.
Drawing from these studies, Kearns showed that people tend to have strong collective performance when working as part of an online social network, and that it is easier to get people to disagree than agree. Additionally, the studies showed the overwhelming influence of the minority, as in study after study people in the minority were able to overwhelming convince their neighbors to change their stance.
Additionally, Kearns noticed a new trend: when people were allowed a comprehensive look at all of the participants in a study and their actions within the network, they were not as successful at completing the assigned task. Further work is needed, according to Kearns, but initial results suggest that humans may not be the best at building their own networks.
Abby Abazorius, CSAIL