Decision Impact Forecasting and Modeling

What

Interface where participants input policy proposals and receive impact models, showing effects on both targeted and related outcomes with confidence intervals, assumptions, and interactive exploration capabilities.

Why

Enable informed decision-making ↑ by participants with limited system knowledge, allow exploration of novel solutions ↑, improve policy effectiveness ↑, and reduce expert consultation costs ↓ 

Problem Definition

Citizens in deliberative processes must make consequential policy recommendations without access to sophisticated modeling tools that professional policymakers use. They rely on static expert presentations and intuition to understand complex cascading effects. Current expert-led impact assessments are expensive, take weeks to months, and are often inaccessible to lay audiences. This leads to recommendations with unintended consequences or missed opportunities for more effective interventions.

Definition of Success

Deliver accurate impact assessments for low cost (<$5,000 per process). For a Citizens’ Assembly: Thorough approach potentially costs $100k+ (detailed modeling, adjustments, presentation in understandable format) and so is almost never utilised. This tool: $5k setup with unlimited queries would enable exploration of many more policy variations, potentially saving billions through better policy outcomes (e.g., 5% more effective poverty policy = $500M+ saved in mid-sized state).

Requirements

• Interface: natural language input; visual dashboard showing multi-dimensional impacts; real-time parameter adjustment with updates; side-by-side policy comparison; and uncertainty visualization and confidence intervals.
• Challenges to address: communicating uncertainty without undermining confidence; preventing result manipulation; and handling long-term impacts.

Existing Limitations

Currently relies on: (1) Pre-prepared expert assessments delivered as 40-100 page reports and expensive, (2) Expert testimony limited to 20-30 min presentations, (3) Simplified scenario cards showing only first-order effects, (4) Facilitator-led discussions based on intuition. No real-time modeling, variation testing, or systematic identification of unintended consequences. Existing policy modeling tools (MIT En-ROADS, RAND CHOICE, Urban Institute DYNASIM) are domain-specific and require expertise. AI forecasting research exists but isn’t integrated for citizen use.

Milestones

1. Basic single-dimension calculator (e.g., tax revenue only) demonstrating core value.
2. Multi-dimensional modeling across economic, social, environmental impacts.
3. Successful assembly pilot with expert validation of recommendations.
4. Accurate prediction of implemented policy outcomes.

Starting Points

1. Partner with 3 upcoming assemblies for A/B testing (traditional vs. tool-assisted tracks).
2. Build on existing modeling frameworks: Policy Priority Inference, En-ROADS, and DYNASIM.
3. Integrate AI forecasting research from Metaculus and ForecastBench.
4. Collaborate with CBO and think tanks for model validation.
5. Weekly iteration with facilitators and participants.