AI can help government work better — if we use it wisely

For decades, government has struggled with the same basic challenge: how to deliver services efficiently, fairly, and at the scale Americans expect. 

The obstacles have changed — recessions, pandemics, political polarization — but the problem has not. Today, however, two powerful forces are converging in a way that could break the pattern: the rapid advance of artificial intelligence and growing pressure to do more with fewer public dollars.

Together, they create a rare opportunity to rethink how government actually works — not by cutting corners, but by redesigning everyday processes so they are faster, more accurate, and more humane.

This moment didn’t arrive overnight.

For nearly four decades, I’ve worked with federal and state agencies to improve public programs through data and evidence. For much of my early career, data were scarce, studies took years, and results often arrived long after policy decisions were made. Programs were judged in blunt terms — effective or not — with little ability to improve in real time.

About 15 years ago, digital records and administrative data changed that. Agencies gained the ability to monitor programs continuously and adjust operations as they learned what worked, bringing evidence into day-to-day management rather than relying solely on post-hoc assessments.

Today, we are on the verge of another shift. Modern AI systems can scan massive data sets, summarize documents, and generate first-pass analyses in minutes. Increasingly, they can also support how work flows through agencies by routing information, flagging issues early, and connecting systems that have long operated in silos. These technologies won’t replace public servants, but they will change how their work gets done.

That matters in an era of tight budgets. If government is going to maintain service quality while managing fiscal constraints, it must find ways to expand capacity without expanding payrolls. Used well, AI can help agencies do exactly that.

Consider something familiar to millions of Americans: applying for public benefits such as Medicaid, the Supplemental Nutrition Assistance Program, or housing assistance. Eligibility workers often spend hours reviewing scanned pay stubs, leases, and medical records — sometimes submitted as blurry phone photos — then reentering the same information into multiple systems. The process is slow, frustrating, and prone to error.

AI offers a practical alternative. Imagine an “eligibility copilot” that works alongside the caseworker. As applicants upload documents, the system automatically identifies them, extracts key information, and checks it against program rules. It flags only the cases that require human judgment and drafts clear, plain-language notices for staff to review and send. Every recommendation is reviewable, and the caseworker remains in control, but their time shifts from data entry to decision-making.

States are already piloting pieces of this approach. For example, in King County, Washington, agencies are using AI to automatically classify and extract information from large volumes of incoming documents, reducing the need for staff to manually sort and review paperwork. Officials there estimate that their AI redaction pilot safely removes sensitive personal information from senior property tax exemption applications, reducing the time spent per application from 30 minutes to less than five seconds. Routine cases move faster, and staff can spend more time applying their expertise to complex situations. The technology doesn’t replace human judgment; it frees experienced workers from rote tasks so they can focus where they add the most value. 

Scaled nationally, tools like this could help millions of households get assistance more quickly — especially during economic shocks or natural disasters — while allowing overburdened caseworkers to focus on resolving complex cases rather than transcribing paperwork.

AI can also help agencies clear backlogs, anticipate service bottlenecks, evaluate programs in real time, and strengthen program integrity without increasing burdens on the vast majority of people who follow the rules.

But these benefits are not automatic. The agencies seeing the strongest results are adopting AI with clear governance — transparency, rigorous testing, continuous monitoring and evaluation for bias and unintended consequences, and strong human oversight — so AI supports human judgment and earns public trust. As I often tell students entering public service, the future belongs to professionals who know how to work alongside AI, not defer to it blindly.

If government gets this right, today’s disruption can become a turning point. Public servants are deeply committed to doing right by the people they serve. What they lack is time, capacity, and modern tools. When a parent waits weeks for a benefits decision or a veteran can’t get timely answers on a disability claim, the consequences are deeply personal.

Used responsibly and for clearly defined purposes, AI can give public servants the breathing room and insight they need to deliver services the way they’ve always wanted to — faster, fairer, and with greater dignity for the people on the other side of the counter. 

The choices agencies make now — where to start, how to govern these systems, and how to keep humans firmly in the loop — will determine whether AI becomes another missed opportunity or a catalyst for a stronger, more trustworthy public sector. I believe it can be the latter, and that government should seize this moment while it still can.

Paul Decker is the president and CEO of Mathematica.