COMPASS
Cognitive Opinion Modeling for Population Analysis and Scenario Simulation
When Complex Scenarios Evolve Faster Than Traditional Surveys Can Measure Population Response
At a Glance
COMPASS is a general-purpose AI tool that predicts population behavioral responses to complex, rapidly evolving scenarios by training large language models on longitudinal survey data and integrating them with structured psychological models. Traditional surveys are expensive to field, take weeks or months to collect, and deliver results when conditions have already changed—creating dangerous blind spots during crises. COMPASS solves this by generating real-time predictions for how specific demographic groups will respond to intricate "what-if" situations, enabling rapid policy testing impossible with human respondents. Our FluPaths/COVIDPaths dataset (8 years, 2000+ respondents, 16 waves) provides initial validation for pandemic behavioral modeling, demonstrating proof-of-concept that the tool can be extended to any longitudinal panel data and any domain where understanding population response is critical but traditional surveys cannot keep pace with changing conditions.
The Challenge
Policymakers, researchers, and organizations face a critical paradox during rapidly evolving situations: the behavioral data they need most urgently—how populations perceive risk, whether they'll adopt new policies, how quickly they'll change behaviors—arrives too slowly to inform decisions. Traditional surveys are expensive to field (often $50K-$200K+), take weeks or months to design, collect, and analyze, yet conditions shift daily as crises unfold. By the time survey results arrive, the situation has changed and the data is already outdated. This creates dangerous blind spots across domains.
Existing Approaches Fall Short:
Survey Speed & Cost Problem
Surveys are increasingly expensive with declining response rates. By the time results arrive (weeks to months), conditions have already changed—whether pandemic spread, economic shocks, or policy environments.
Counterfactual Impossibility
Complex "what-if" questions ("How would you respond if unemployment doubled AND inflation spiked?") overwhelm human respondents but are essential for scenario planning across all policy domains.
Resource Trade-offs
Limited budgets force researchers to choose between frequent simple questions or infrequent complex assessments—neither captures rapidly evolving behavioral dynamics.
The result? Decision-makers across domains—public health during pandemics, economic policy during financial crises, climate adaptation planning, social program design—make critical assumptions about population behavior without empirical validation. Policy interventions launch without understanding public response, and officials discover behavioral barriers only after interventions fail. Organizations need behavioral intelligence that keeps pace with change, couples with ongoing data collection, and deploys rapidly when crises strike.
The Solution: AI-Augmented Behavioral Intelligence
COMPASS does what traditional surveys cannot: generates high-fidelity predictions of how specific population segments will respond to evolving scenarios—in real-time, at scale, and for complex counterfactual situations impossible to ask human respondents—by training AI models on longitudinal behavioral data. The tool couples with ongoing data collection efforts and deploys rapidly when conditions change, whether during pandemics, economic shocks, climate events, or policy transitions.
Three Game-Changing Capabilities
1. Real-Time Behavioral Forecasting at Crisis Speed
- Generates predicted responses to new scenarios in seconds instead of weeks
- Enables continuous behavioral intelligence as conditions evolve—economic shocks, policy changes, environmental events, health crises
- Supports rapid "what-if" analysis: test dozens of intervention scenarios in hours
- Couples with ongoing surveys: AI maintains predictions between survey waves, flags when new data collection is urgently needed
Why it matters: When conditions change rapidly—new policies announced, economic indicators shift, crises emerge—decision-makers need to know how populations will respond NOW, not three weeks from now when survey results arrive
2. Complex Scenario Analysis Beyond Human Cognitive Limits
- Evaluates intricate counterfactuals with multiple interacting factors (economic conditions × social networks × institutional trust × past experiences × current policies)
- Captures population heterogeneity: different demographic groups respond differently to identical scenarios
- Provides probabilistic forecasts with uncertainty quantification, not point estimates
- Cross-domain scenario testing: How do health, economic, social, and environmental factors interact to shape behavioral responses?
Why it matters: Real policy decisions involve complex trade-offs—COMPASS can predict how "economically stressed households with low institutional trust" will respond differently than "secure households with high civic engagement"
3. Transparent AI Grounded in Psychology
- Not a black box: Combines language models (understanding context and nuance) with structured psychological frameworks (explaining WHY people make choices)
- Explains predictions through intermediate causal factors (perceived risk, institutional trust, social influence, past experiences)
- Allows experts to inspect, validate, and adjust the reasoning process
- Continuously calibrated to real human survey data to prevent AI drift
- Domain-extensible framework: Validated with pandemic behavioral data; applicable to any longitudinal survey panel and any domain
Why it matters: Decision-makers need to understand WHY predictions indicate certain responses to design effective interventions—and they need confidence the AI reflects real human psychology, not training biases
What COMPASS Enables That Traditional Surveys Cannot
For Policy Research Organizations
- Rapid policy impact assessment across multiple scenarios without fielding new surveys
- Cross-domain scenario testing: How do health, economic, and social factors interact to shape behavioral responses?
- Real-time crisis intelligence when conditions evolve faster than traditional data collection
- Test intervention messaging strategies: Which communication frames resonate with different populations?
For Public Health Agencies
- Pandemic preparedness & response: Predict population behavioral responses during outbreaks
- Intervention design: Test vaccine communication strategies and protective behavior campaigns
- Rapid behavioral intelligence when outbreak conditions evolve faster than survey cycles
- Validate epidemic model assumptions about behavioral responses to disease dynamics
For Computational Modelers & Researchers
- Inform agent-based models with realistic, heterogeneous behavioral responses across domains
- Test model sensitivity to behavioral assumptions impossible to measure empirically
- Generate synthetic behavioral datasets for model development and validation
- Accelerate modeling cycles: simulation → behavioral prediction → refined simulation in hours, not months
For Social Policy & Economic Researchers
- Economic policy response modeling: How do populations adapt to inflation, unemployment, or policy reforms?
- Climate adaptation planning: Predict behavioral responses to environmental changes and mitigation policies
- Social program design: Test intervention strategies before expensive rollouts
- Test theoretical predictions in counterfactual scenarios unavailable in real data
Innovation Grounded in Established Expertise
Pioneering Research Track Record
- Longitudinal behavioral data from FluPaths and COVIDPaths studies spanning 2016-present (8+ years, 16 survey waves, 2000+ respondents)—providing initial validation dataset for pandemic behavioral modeling
- Initial prototype development through UCLA IPAM RIPS program (Research in Industrial Projects for Students)—summer collaboration with undergraduate research team validating feasibility of AI-augmented behavioral forecasting
- Established agent-based models coupling disease transmission with behavioral adaptation—demonstrating proof-of-concept for integrated modeling
- Published methods for extracting behavioral parameters from survey data to inform simulation models
AI Architecture: Combining Language Understanding with Psychological Theory
Language Model Integration
Large language models understand context and generate situation-specific concerns (fears, motivations, social influences) from individual profiles—capturing nuanced reasoning impossible with traditional rule-based approaches
Psychological Framework
Structured decision models weight AI-generated concerns according to individual values to predict behavioral choices—ensuring predictions are grounded in psychological theory, not just statistical patterns
Transparent by Design
Experts can inspect intermediate reasoning, not just final predictions—enabling validation, adjustment, and trust-building with decision-makers
Continuous Human Calibration
COMPASS doesn't replace surveys—it augments them with AI-generated predictions between survey waves. Regular validation against new human data prevents AI drift and maintains accuracy.
- In-sample validation: predict held-out responses from existing longitudinal data
- Out-of-sample validation: predict responses to new survey questions never seen during training
- Prospective validation: forecast future behavioral shifts and verify against subsequent surveys
- Fairness evaluation: ensure predictions remain accurate across demographic subgroups
- Active learning: identifies scenarios where AI confidence is low, triggering targeted human surveys
Flexible Development Pathways
COMPASS development follows a phased approach ensuring each stage delivers validated capabilities while building toward a general-purpose behavioral forecasting platform applicable across domains.
| Phase | Investment | Timeline | Key Deliverables |
|---|---|---|---|
| Phase I: Core Model Development | $270K–$345K | 12–15 months |
• Neuro-symbolic architecture integrating LLMs with causal behavioral models • Training on FluPaths/COVIDPaths data (8 years, 16 waves) • Validation on held-out survey responses • Working prototype generating predictions for COVID-19/influenza behaviors |
| Phase II: Full Platform Development & Deployment | $270K–$345K | 12–15 months |
• Extension to complex counterfactual scenarios with multiple interacting factors • Additional validation testing predictions on novel scenarios • Integration with agent-based transmission models • Closed-loop system: ABM simulations → behavioral predictions → refined ABM parameters • Real-time prediction interface for rapid "what-if" analysis • Web-based decision support tool for public health planners • Complete production-ready platform linking disease dynamics with behavioral forecasting |
Investment Options
Phase I Only
Core neuro-symbolic architecture with validation; working prototype for pandemic-related protective behaviors
Full Integration (All Phases)
Complete platform with ABM integration and web-based decision support; production-ready tool for behavioral forecasting and epidemic scenario planning
Custom Applications:
$75K–$200K for specific domain applications (e.g., climate policy, economic interventions, specific health contexts), geographic regions, or policy scenarios requiring tailored development
Why Now
AI Capabilities Reached Critical Threshold
Recent advances in large language models enable nuanced understanding of complex behavioral scenarios—previous generation models couldn't handle this task
Established Behavioral Data Foundation
Our 8-year longitudinal dataset provides empirical grounding for initial validation—demonstrating proof-of-concept for pandemic applications with extensibility to other domains
Urgent Policy Need Exposed by COVID-19
The pandemic revealed catastrophic gaps in behavioral intelligence during crises—we cannot afford similar failures in future emergencies
Behavioral Intelligence Needs Transcend Pandemics
Economic shocks, climate events, social disruptions, technological shifts—all require rapid understanding of population response. Traditional surveys cannot keep pace. AI-augmented forecasting is essential for evidence-based policy in fast-moving environments
COMPASS transforms a fundamental limitation of crisis response across domains—the impossibility of rapidly assessing population behavioral responses to novel, complex scenarios—into a solvable engineering problem. For decision-makers facing pandemics, economic shocks, climate events, or policy transitions, this is the difference between evidence-based intervention design and educated guessing about how populations will respond.
Ready to Discuss COMPASS for Your Organization?
Whether you're a public health agency, policy research organization, federal agency, or behavioral science team, we'd like to explore how COMPASS can provide the rapid behavioral intelligence your work requires.
CausalPaths Analytics LLC
Advancing expert modeling through AI-augmented capabilities and 20+ years of domain expertise