Large-Scale Social Simulations for Understanding Information Flows, Cultural Patterns, Collective Adaptation, & Emergent Structures
Alexis Drogoul, Institut de Recherche pour le Développement au Vietnam; Kiran Lakkaraju, Sandia National Labs; Santiago Nuñez-Corrales, University of Illinois at Urbana-Champaign; Samarth Swarup, Virginia Tech; Alex Yahja, NCSA
There is tremendous interest in computational modeling, simulation, and analysis of social, socio-technical, and socioenvironmental systems, in many different application areas including theoretical social science, government/policy, and commerce/industry. Even at a small scale, social simulation has proven useful for basic research supporting human/social aspects of policymaking, disaster management, climate change, disease epidemics, environmental management, urban development, and more. Simulations can provide tremendous educational opportunities to explore and visualize alternative theories and models of complex social systems, and to study how they change. Thus, agent based simulation (ABS) has become a standard tool for computational social science (CSS). ABS/CSS models generally comprise 3 levels: (1) a physical world or GIS model, (2) a collection of active "agents" with simple or detailed knowledge-rich cognitive architectures, and (3) a communication layer for information exchange. Virtually all current ABS/CSS models employ very simple layers run on single machines. Scaling scientific ABS/CSS applications remains challenging for theoretical and pragmatic reasons including: a) Information complexity of models; b) gaps in our knowledge of how patterns in real data can optimize large ABS/CSS models; c) lack of integrated modeling, software, hardware, and scientific workflow infrastructures for large ABS/CSS models; and d) missing practice and experience engineering such complex multi-level ABM/CSS models for HPC environments. Having tools that scale in and across each of the three layers is crucial for grounding theoretical ideas in real social science data and problems. This proposed work will run two computation- and data-intensive ABM/CSS experiments: one that models transmission and coherence of cultural knowledge and one that models the emergence of organizational structure among sophisticated actors. These experiments examine different facets of how information flow and interpretation patterns, conditioned by physical constraints and population size, influence social structure and coherence over time. As a side effect the experiments are designed to test and gain experience with new software tools, scaling ability, cross-level integration, and the pragmatics of science workflows for ABS/CSS models. Our request of Blue Waters resources is justified by four elements: 1) the computational and communication complexity of large-scale models studying interesting new research problems in social sciences, 2) the ability to test batch implementations of such simulations in a leadership-class supercomputer, 3) the need for data storage allowing for complete reconstruction of histories 4) the CPU power required for modeling large collections of complex agents. Our research interest is twofold: (1) understanding how large-scale multi-agent systems may provide a window to many problems in social science and (2) testing and extending supercomputing infrastructures towards a goal of building what has been called a "social science supercollider."