Introduction: Simulation, Visualization, and Scientific Understanding

Only a decade ago, the topic of scientific understanding remained one that philosophers of science largely avoided. Earlier discussions by Hempel and others had branded scientific understanding a mere subjective state or feeling, one to be studied by psychologists perhaps, but not an important or fruitful focus for philosophers of science. Even as scientific explanation became a central topic in philosophy of science, little attention was given to understanding. Over the last decade, however, this situation has changed. Analyses of scientific understanding that do not treat it as a subjective state or feeling have been offered and debated, and both the epistemic value and the pitfalls of purported psychological aspects of understanding have been discussed. The topic of scientific understanding has emerged as one meriting—and receiving—significant attention from philosophers of science.

During the same period, the topic of computer simulation also burst on the scene; what had been a trickle of philosophical papers became a steady stream, if not a veritable flood. Computer simulation is now widely recognized as a central and vital tool of contemporary science, and philosophers are exploring both how this tool is used in practice as well as how its addition to the methodological landscape puts pressure on traditional analyses of how science works. One issue that has not received much attention so far, however, is how simulation contributes to scientific understanding; since scientists themselves not infrequently claim that simulation is a valuable tool in this regard, there is work to be done here.

Visualization is a cross-cutting theme, periodically identified as important or valuable in discussions of both scientific understanding and computer simulation. Yet exactly why visualization is important or valuable often goes unstated, leaving a number of unanswered questions. Is visualizability of a process or phenomenon or theory a precondition for scientific understanding of it? How do advanced visualization tools enhance the utility of simulation results for various purposes, including understanding? And so on.

The papers in this special issue address questions at the intersection of these three topics: scientific understanding, computer simulation and visualization. They are a subset of the papers presented at a workshop on scientific understanding held at the Lorentz Center (Leiden, The Netherlands) in 2010. Spanning a wide range of scientific fields—from sociology to biology to climate science to fundamental physics—as a group they both reveal common threads and serve as a reminder of the diversity of practices in science, including thought experiments, theoretical analysis on paper, computer simulations, and data-intensive research employing online databases.

Petri Ylikoski highlights links between simulation and understanding in his paper, “Agent-Based Simulation and Sociological Understanding.” Agent-based simulation is a methodology in which the represented behavior of numerous individual agents is made to evolve over time according to rules that consider at each time step both properties of the agent (e.g., goals, preferences, beliefs, etc.) as well as the agent’s environment. Agent-based simulation is a relatively new approach to research in the social sciences but, as Ylikoski argues, one that seems particularly promising as a means of advancing understanding of sociological phenomena.

Ylikoski holds an inferential conception of understanding, on which understanding a phenomenon consists in the ability to make what-if inferences about that phenomenon. Understanding a macro-level sociological phenomenon involves not just identifying correctly the mechanism by which the phenomenon emerges from the micro-level behaviors of individual agents but also being able to infer what would happen if conditions were different in various ways. Agent-based simulation, he argues, increases inferential understanding of sociological phenomena in at least two ways. First, it allows scientists to make a wider range of what-if inferences, since it can reveal the implications of modeling assumptions—including implications that cannot be derived analytically. Second, it allows scientists to make these inferences more reliably (i.e., to reach correct conclusions more often) than would be achieved by unaided human reasoning.

To make his discussion of agent-based simulation more concrete, Ylikoski considers a classic example: Schelling’s model of residential segregation. Since its...