Large Language Models Are Transforming Evidence-Based Policy-Making

The views expressed are those of the author and do not necessarily reflect the views of ASPA as an organization.

By Dennis McBride
August 4, 2025

Artificial Intelligence, specifically, Large Language Models (LLMs), will thrust evidence-based policy-making to a new level. LLM technology will serve as a powerful facilitator—a smart, even live collaboration capability connecting public administrators with one another and crucially with stakeholders—from individuals and interest groups, government agencies, legislators. Perhaps even with the judiciary.

Imagine a live public forum where concerned citizens or think tank specialists question or request a change in the very assumptions that underlie a prospective policy and view almost immediately the likely consequences, or at least the sensitivities of that policy, in real time. AI will become a very useful “what if and why” co-pilot in the multi-layered, multiple chessboard challenge that public administration faces every day… if it includes a key component: the ability to explain itself. Let’s look at a trio of examples from local, state and national public administration scenarios.

Scenario 1: A city planner is tasked with vetting a proposed new zoning ordinance. With LLM co-piloting, the planner uploads planning documents, demographic data and the proposed ordinance text. The administrator seeks expertise and data from specialty sources including traffic engineering, GIS dynamics, sociology and other relevant specialties. LLM “understands” all of this complex material, including what it doesn’t know. It goes online to pull down needed data, which it reveals and discloses its sources. LLM then generates an array of “what-if and why” outcomes. It provides a plain English explanation of each. For the administrator, there are expected outcomes, but there are surprises, but with explanation, the surprises become serious possibilities or even likelihoods.

That afternoon, in live public forum, as community members convene, the administrator provides summaries of likely outcomes. With each live challenge, LLM instantly runs micro-simulations, making visual the immediate impact of a suggested change on, for example, traffic or school capacity, potential impacts on neighborhood housing and local retail economics. The administrator asks: “Okay, based on feedback from Neighborhood Association X about density and Developer Y’s proposals for green space, what are the three most likely traffic pinch points this could create and what mitigation efforts, within our budget, would be most effective?” LLM immediately generates interactive maps and projections and suggests policy adjustments that attempt to reconcile conflicting interests within budget.

Scenario 2: A state Department of Health faces the challenge of preparing for an anticipated influenza outbreak. Professionals need to understand infection patterns, optimal vaccine distribution and hospital capacity strain. The AI-powered platform ingests large volumes of historical disease data, population density and local travel maps along with healthcare infrastructure resource details. LLM stays awake and integrates real-time updates from county health departments, hospital networks, federal agencies and vetted citizen reports.

During the live event (or a simulated event), the public health administrator inputs live updates from hospital dashboards and emergency dispatches. LLM immediately assesses bed availability and vaccine distribution bottlenecks. The administrator prompts: “Given current hospital admissions from County Health Department A and projections from Federal Agency B, how many ICU beds will be needed in each county at peak? Based on real-time supply chain data from pharmaceutical partners, what’s the optimal strategy for distributing antiviral medication to reduce hospitalizations, considering logistical support from the National Guard and Coast Guard Auxiliary?” The AI runs multiple scenarios, visualizes contagion dynamics and offers data-driven recommendations. It acts as a central hub, synthesizing disparate health data and external partner inputs. It generates clear, actionable briefing material for emergency management agencies and drafts public advisories that resonate with specific demographics, all vetted by public health communication experts for immediate release.

Scenario 3: National policymakers are grappling with complex economic challenges including trade agreements and energy policy shifts, requiring consideration of broad economic ripple effects. A federal agency administrator uses a vetted LLM-integrated economic model to explore “what-if” scenarios. Experts upload details of a proposed trade tariff, including economic forecasts from think tanks, impact assessments from industry associations and concerns from labor unions. It churns, and in minutes, informs administrators about the most likely courses the new policy would likely take—leaving to superior human intellect the important vetting of courses that it would very likely not take.

During the afternoon’s congressional hearing, as lobbyists and union representatives present their views, LLM processes live testimony. The administrator tasks the AI: “Estimate the impact of a 15% tariff on imported steel on domestic manufacturing jobs, consumer prices for automobiles and overall GDP growth in the next two fiscal years, incorporating the real-time projections from Think Tank X and the employment concerns raised by Labor Union Y.” LLM declares a range of potential outcomes, each showing sensitivity to critical variables and to permutations of multiple interactive variables. In this case, though, LLM declines to give a single, summary forecast. It advises the administrator: “we will need to look for changes a, f and y, in variables 4, 7 and 13, and then we might know…”

The path forward isn’t without its challenges—data quality, AI bias and rigorous human oversight remain paramount. The integration of diverse, sometimes live, stakeholder inputs introduces new complexities, demanding vigilance against amplifying existing biases and ensuring the AI remains a neutral facilitator, not an arbiter, of human deliberation. Yet, the opportunity to democratize modeling and simulation, making it accessible and actionable for public administrators while fostering dynamic, collaborative interactions across sectors, is a game-changer. By empowering more public administrators with these advanced tools, we can move closer to a future where evidence truly guides policy, leading to more informed, effective and resilient public administration for all.

---

Author: Dr. Dennis K. McBride is President, Institute for Regulatory Science, adjunct professor at Georgetown University and Florida Tech, and former editor-in-chief for Review of Policy Research. He recently served as Director of the Office of the Secretary of Defense, Acquisition Innovation Research Center, a policy game-changing consortium of two-dozen universities, including colleges of policy, business, law, science, engineering, and technology. Contact information:  [email protected]; cell/text:  703-635-5285

(No Ratings Yet)
Loading...