A Symbiotic Agent-Based Network Platform Linking Expert Knowledge and Machine Learning for Systemic Risk Mitigation

Significant increases in interdependencies and interactions dominate modern societies, fundamentally transforming critical systems like the global food network and human-mobility dynamics. The pace and magnitude of these changes have severely limited the capacity of decision makers to respond effectively to new and unanticipated events. Here we propose a symbiotic platform to rapidly anticipate and assess shock events for food security and human migration. By linking human-expert judgment and machine learning, a software platform will provide enhanced identification and early warning of resultant impacts. This enables the implementation of sound and responsive socio-political or economic interventions in order to reduce instability. The symbiotic platform approach moves beyond present-day, piecemeal attempts with advancing decision-maker response capacity by integrating a network perspective into the software architecture. This network perspective, i.e. the awareness and implementation of cross-links among domains, is essential, considering that "we will never understand complex systems unless we develop a deep understanding of the networks behind them" (Barab‡si & P—sfai, 2016). The platform will contain multiple models on food provision, use, trade, and price dynamics as well as migration Ð developed by subject-area experts Ð that have the capacity to represent system dynamics at multiple scales and are connected via a network-based software that allows for interaction between the models, with an agent-based trade model at its core. We will start with threats to the global food system and then extend our approaches to their downstream pressures on human mobility. At its core, the AB-EMP will have a multiscale agent-based dynamic network model of the global food-resource system, which simulates short-term dynamics (< 1 year) associated with shocks and identifies regions with heightened vulnerabilities in subsequent years. This core component will be linked with state-of-the-art expert models of household-level food security, agricultural production, and commodity logistics. The architecture of the software platform will be based on a symbiotically derived ontology of food security. Key outputs from this model ensemble will include global commodity prices and food-security metrics (hunger, import/export totals & pathways, rural economic metrics etc.).