The Promise and Perils of Simulation in the Social Sciences

Moreover, humans—unlike atoms—are amenable to many other kinds of study methods like opinion surveys, document analyses, or participant observations. Such empirical methods have in the past proved to be more productive than simulation for increasing our knowledge on social life and the promotion of social theory.

The success of simulation across many other scientific domains has nevertheless invited growing interest in using computers in social research, particularly for supporting political decision making. It is conceivable, for example, for a government to want to use simulation when planning for social programs a country might need in 10 or 20 years. But troubling questions about simulation's abilities remain: Are today's computational social sciences mature enough to make such predictions reliably? Would it make a difference if we gained more and better data, more computing power or better algorithms? What questions can we expect simulation to answer in the future? And what problems will always be too complex for simulation to solve?

Recently, members of the High-Performance Computing Center Stuttgart (HLRS) Department of Philosophy of Science and Technology of Computer Simulation, under the direction and initiative of HLRS Director Prof. Michael Resch, launched a new collaboration with investigators in the FAU Institute for Sociology focused on the intersection of simulation and the social sciences. On March 9–10, 2018, the group held its inaugural symposium in Erlangen to begin identifying social science problems that simulation could help address and developing a better understanding of simulation's capabilities and limits. The meeting was the first in a series of annual symposia that will be held alternately in Stuttgart and Erlangen in the coming years.

Titled "Simulation in the Social Sciences and the Sociology of Simulation," the symposium brought together two key communities of social scientists—those who use simulation to do social research and those who study how computational social research is interpreted and used in society. To provide additional context, the event also featured talks by simulation scientists from HLRS who discussed scientific uses of simulation and its relevance for society.

What simulation can do, and what it can't

Simulation in the social sciences typically falls into two categories, says Prof. Dr. Nicole J. Saam, who specializes in methods of empirical social research at FAU and was a co-organizer of the symposium. In one category, simulation has been used to test "toy" models developed using social theory. In game theory, for example, scientists have used simulation for several decades to investigate hypotheses about how trust, cooperation, and social norms evolve during human interactions and decision-making.

In another category, simulation can explore "microscale" models based on large collections of empirical data—for example, demographic data representing key events in a person's lifespan. Such models might be used, for example, to predict trends in population change, workforce participation, or needs for long-term care as people age.

Participants in the symposium pointed out that in both macroscale and microscale models, simulation will always face limitations, as input conditions must be dramatically simplified to focus on specific problems. Microscale models, for example, can not reflect the emergence and effects of things like public opinion, laws, and different cultural backgrounds. In addition, no model can predict the effects of unforeseen events like wars, mass migrations, or sudden political realignments like Brexit, which have major impacts on societies.

"Are social scientists condemned to only developing toy models?" Saam asked in a follow-up interview. "This is an important question that we as social scientists need to discuss. Some people think it should be possible to simulate pretty much anything and come to us with completely unrealistic expectations. We need to understand the foundations of simulation better, particularly what kinds of simulation models can't be developed. This will help us to learn what models we can indeed develop, what characteristics these models have, and what kinds of knowledge they can produce."

The social context of simulation

Whenever simulations are developed—whether in the social sciences or physical sciences—the context in which they originate is often obscure to those who did not participate. In the natural sciences, for example, simulations often arise out of collaborations among large groups of investigators. For social scientists, understanding the social processes through which such projects grow also presents interesting questions.

“Considering the size of the research groups and the number of different disciplines involved in the development of technologically and mathematically complex computer simulations, it is understandable that it would be difficult even for simulation scientists to understand them completely," says Dr. Andreas Kaminski, leader of the HLRS Department of Philosophy of Science and Technology of Computer Simulation. Philosophers of simulation refer to a condition called "epistemic opacity," the fact that not only nature but also the scientific method itself can appear opaque.

“Experts in the field of simulation sciences are beginning to become aware of this problem," Kaminski explains. "Now it has also become very important that those who want to use computer simulations from a social science perspective or in the political field understand the scientific method behind simulation, as well as its limitations.”

Also important to simulation results is how they are received and integrated into society and political processes. As HLRS Director Michael Resch related, several earth scientists were found guilty of manslaughter when their models of seismic activity near the town of L'Aquila, Italy, in 2009 did not lead them to warn local residents about imminent danger—an earthquake killed approximately 300 people there at that time. The controversial judgment was later reversed, though it still indicates that society has high expectations for scientific modeling. Resch suggested that social scientists can help to improve communication between scientists and society to promote better understanding of what scientific models can do and what they can't predict.

The Erlangen workshop was one step toward this goal. By bringing together critical perspectives from the social sciences, philosophy of science, and simulation science, the group aims to chart a path forward for simulation in social research. And indeed, the March 2018 event was a prelude to a series of workshops that will take place in the future. The next workshop will take place at HLRS in spring 2019, with the location of the meeting alternating between Stuttgart and Erlangen in the following years.

— Christopher Williams