World Modelers

The World Modelers program aims to develop technology that integrates qualitative causal analyses with quantitative models and relevant data to provide a comprehensive understanding of complicated, dynamic national security questions. The goal is to develop approaches that can accommodate and integrate dozens of contributing models connected by thousands of pathways—orders of magnitude beyond what is possible today.

World Modelers analyses are intended to be timely enough to recommend specific actions that could avert crises. The program seeks to develop technologies that will provide clearly parameterized, quantitative projections within weeks or even hours of processing, compared to the months or years it takes today to understand considerably simpler systems.

World Modelers technologies will be applied to increasingly varied use cases as they mature through the phases of the program. Questions for analysis will typically be framed at subnational scales and look one to five years into the future, although the factors that influence outcomes of interest might operate on larger spatial and temporal scales. This subnational focus reflects the changing nature of conflict and security, which, increasingly, plays out in cities and districts. The first use case of World Modelers is food insecurity resulting from the interactions of multiple factors, including climate, water availability, soil viability, market instability, and physical security.

Causemos

Causemos is a collaborative analytical platform, developed by Uncharted Software, that serves as the main human-machine interface (HMI) for the World Modelers program. Its qualitative and quantitative analysis workflows are designed to allow analysts—from generalists to subject matter experts—to leverage integrated knowledge, data, and models to better understand complex multi-domain issues.

• Create custom models: Build causal analysis graphs to represent your mental model of the complex system you are analyzing. Augment the model with system-generated suggestions based on evidence from uploaded scientific literature. Use the model to visualize the cascading effects of change from different scenarios.
• Analyze data across time and space: Quantify your mental model with system suggestions or upload your data. Use comparative analysis capabilities to understand the historical precedents, regional analogs and correlations between variables.
• Identify risk: Create your own composite index of risk based on the weighted key drivers identified in your causal analysis graph. Easily explain why a region's population is considered most at risk.
• Plan interventions: With more visible levers in the causal analysis graphs, think creatively about interventions, their challenges, and the unanticipated second and third order impacts. Complement with integrated sector-specific expert models to test scenarios and interventions at higher spatial resolution.
• Build briefings: Capture insights throughout the analysis process and export findings to support report generation and briefings.

Corpus Ingestion and Assembly

DART, developed by TwoSix Labs, is the technology pipeline for uploading and extracting text and analytics from documents. It also provides capabilities to enable rapid ontology curation. Three machine readers (Eidos from University of Arizona, Hume from BBN and Sofia from Carnegie Mellon university) extract causal statements from the corpus of documents. INDRA, developed by Harvard Medical School, assembles the causal statements into knowledge bases that can be leveraged by analysts in Causemos to build qualitative models backed by evidence from the literature.

Qualitative Analysis

The qualitative (“top-down”) analysis space in Causemos enables causal analysis graph (CAG) assembly and scenario analysis using three inference engines: DySE from University of Pittsburgh, Delphi from University of Arizona, and Sensei from Jataware. Qualitative models define representation and reasoning about continuous aspects of entities and systems in a symbolic, human-like manner (Forbus 2008). They allow analysts to:

• Facilitate collaboration by making their knowledge of a complex problem explicit.
• Make sense of and communicate complexity as stories.
• Get a holistic picture not constrained by availability of quantitative data.
• Develop effective interventions and see unintended consequences.

Supported qualitative analysis types in Causemos include:

• Identifying from literature the main drivers and relationship evidence for target concepts of interest
• Capturing core feedback loops causing a problem
• Identifying most influential drivers/impacts
• Identifying most influential pathways
• Performing current situation / baseline trends assessment
• Comparing scenarios

Quantitative Analysis

The quantitative (“bottom-up”) analysis space in Causemos enables comparative analysis across and within datacubes and quantitative models. Quantitative models incorporate higher-resolution data and mathematical representations to describe the performance of a system for different inputs and initial states. This valuable source of knowledge is typically inaccessible to analysts, as they usually lack the means for finding, calibrating and running these expert models. Dojo, developed by Jataware, enables quantitative model data preparation and registration, which makes these valuable assets accessible to analysts in Causemos.

Quantification makes qualitative models less subjective; Causemos uses the data to make inferences about the nature of the relationships. The system provides a head start on quantification by automatically matching datacubes to concepts. However, human judgment is needed to validate the defaults and resolve any data quality issues.

Supported qualitative analysis types in Causemos include:

• Regional analog (spatial)
• Historical analog (temporal)
• Categorical breakdown
• Scenarios comparison
• Region ranking
• Correlation of variables
• Similar models/data comparison