WIPS - Lindsay Brainard, Washington University in St. Louis

Abstract: The odds are against scientists who investigate systems for which few physical details are observable. In such epistemically bleak contexts, one increasingly popular path forward is to reverse engineer a target system by simulating its observable features. For instance, the fact that astrophysicists are not able to observe the big bang and the early evolution of the universe motivates the need for cosmological computer simulations. I offer an account of how reverse-engineering simulations function in which I identify five heuristics that specify reasonable indicators of their evidential value. I validate some uses of these heuristics by showing how they represent good scientific reasoning. To that end, I present a case study of simulation in which scientists attempt to uncover the structure of human learning by designing computer programs in which simple, ideal, adaptive agents attain mastery of a task through practice. By understanding this simulation project in terms of the account I develop this paper, it becomes clear why reverse-engineering simulations are scientifically valuable and why the scientists who perform these simulations are justified.