P.J. Lamberson: University of California, Los Angeles

The challenge of balancing between exploiting existing solutions and exploring for new ones spans problem domains from animals foraging for food to businesses searching for new revenue streams. In this talk I will discuss a simple version of this tradeoff: When tackling a complex problem, is it better for problem solvers to work in-sequence, where each problem solver exploits the progress of those that have come before them, or in-parallel where each individual explores for new solutions independently? Using a novel mathematical model of collective problem solving we prove that for a class of especially difficult problems that we call unstructured, problem solving in-parallel performs better in expectation. We then use computational simulations to examine how violating the unstructured assumption can lead in-sequence problem solving to be superior. This is joint work with ULCA postdoctoral fellow John Lang.