How intermittent breaks in interaction improve collective intelligence

Ethan Bernstein, Jesse Shore, and David Lazer
PNAS
115 (35) 8734-8739
August 13, 2018

Abstract

People influence  each other when they interact to solve problems. Such social influence  introduces both benefits (higher average solution quality due to exploitation  of existing answers through social learning) and costs (lower maximum  solution quality due to a reduction in individual exploration for novel  answers) relative to independent problem solving. In contrast to prior work,  which has focused on how the presence and network structure of social  influence affect performance, here we investigate the effects of time. We  show that when social influence is intermittent it provides the benefits of  constant social influence without the costs. Human subjects solved the  canonical traveling salesperson problem in groups of three, randomized into  treatments with constant social influence, intermittent social influence, or  no social influence. Groups in the intermittent social-influence treatment  found the optimum solution frequently (like groups without influence) but had  a high mean performance (like groups with constant influence); they learned  from each other, while maintaining a high level of exploration. Solutions  improved most on rounds with social influence after a period of separation.  We also show that storing subjects’ best solutions so that they could be  reloaded and possibly modified in subsequent rounds—a ubiquitous feature of  personal productivity software—is similar to constant social influence: It  increases mean performance but decreases exploration.

Related publications