A Case Study of Quality-Diversity Search in Human Computation

Abstract

Human computation, applying human problem solving to computational problems, has shown promise in numerous applications. In some applications of human computation, it may be useful to find not just a single best solution, but a variety of good solutions with different properties that can be used for further analysis. Recent work in quality-diversity search, such as MAP-Elites, has developed techniques that aim to find a variety of solutions. Thus, in this work, we explore the potential of combining quality-diversity and human computation approaches. We ran a crowdsourced study of the Traveling Salesperson Problem in which some participants were provided with a visualization of their MAP-Elites archive and some were not. We did not find a difference in the quality of the best solutions found by participants between the two groups. However, we did find that participants provided with the archive visualization searched more of the MAP-Elites behavior space than those without the visualization. This demonstrates potential for quality-diversity approaches to lead to finding a wider variety of solutions in human computation search.