Dawn Drescher Proposes Tactical AI Governance Models for Robust Preparedness
Exploring thousands of scenarios to design adaptive policies for AI governance.
Dawn Drescher's proposal, published on LessWrong, calls for a shift from traditional strategic modeling in AI governance to computational exploratory modeling on the tactical and operational levels. While existing work (e.g., Intelligence Rising, MAIM) focuses on strategic scenarios, Drescher argues that dynamic systems require robust, adaptive strategies rather than predictions of likely outcomes. Tools like the EMA Workbench and Dynamic Adaptive Policy Pathways enable preparing for a wide range of futures, a method already used in fields like water management but not yet in AI safety.
Drescher outlines three implementation options: building a reusable software model for think tanks, creating a consultancy that tailors the model to individual organizations, or founding a new think tank dedicated to this approach. The proposal emphasizes cost-effectiveness and adaptability, suggesting that even a basic model could help policymakers test policies against thousands of simulated AI development paths. This could reduce governance blind spots and improve resilience amid deep uncertainty about AI timelines and risks.
- Computational exploratory modeling uses EMA Workbench and Robust Decision Making to generate thousands of scenarios, not just a few likely ones.
- Drescher distinguishes three levels: strategic (current focus), operational, and tactical — the latter two are largely overlooked in AI governance.
- Implementation paths include a general software model, a specialized consultancy, or a dedicated think tank, with the software model being the cheapest first step.
Why It Matters
Moves AI governance from reactive forecasting to proactive, adaptive planning for a wide range of possible futures.