Roughly one in six children in New Jersey live in households that struggle to afford enough food. Since 2024, the federal government has given families $120 per eligible child for summer groceries. Children can qualify through several safety net programs — but because each benefit is run by a different agency, it is easy to miss someone. So New Jersey’s Office of Innovation, which I began leading after its founding in 2018, built an artificial intelligence tool to match records across systems and find eligible children who hadn’t yet been reached. Benefits were extended to 106,000 more children in the first year, and 96,000 the following year. No family filled out a form. The state used information it already had.
When the government repeatedly fails to listen to people, solve their problems and carry out basic tasks, it weakens confidence in democracy itself. I believe we can use A.I. to repair our broken institutions — and people’s cratered trust in them — so they can better address today’s challenges. There is no question that the potential misuse and reckless deployment of A.I. must be addressed to prevent future catastrophe. But we should not let the need to put guardrails around autonomous systems stop us from using A.I. to strengthen the civic institutions that keep our democracy strong and responsive.
The evidence for A.I.’s ability to improve government services goes beyond New Jersey. After the 2025 Los Angeles wildfires, California asked survivors what recovery should look like, and used A.I. to parse through more than 1,500 responses to determine the most pressing concerns and priorities. That has led to new plans by the state to bury power lines and strengthen emergency communications, and Gov. Gavin Newsom just signed a law making the A.I. effort permanent. Last year, the British government estimated that its A.I. tool Consult enabled it to sort more than 50,000 public comments in about two hours — something that would otherwise have taken weeks.
It is striking how much those examples reduce the friction people experience with government services. Now we must go a step further and put A.I. behind every public servant. That transformation can be achieved with three key investments.
Governments should train their entire work force in how to use A.I. to listen, learn and deliver better to the nation’s constituents. Last year, Pennsylvania trained 175 state employees to use ChatGPT in their work; the state found it saved each one an average of 95 minutes a day on drafting, researching and summarizing dense documents.
Training also helps to avoid risks. A child protection worker in Victoria, Australia, who apparently had no training in A.I. used ChatGPT to draft a report on a case of abuse for the Children’s Court, and failed to spot that the chatbot had misread severe danger signs. How can governments regulate these technologies intelligently if their own staff members don’t understand them?
Public workers should get secure access to multiple models at once. That reduces dependence on a single company’s technology and contract terms and lets workers compare which tools work best for which tasks.
Governments can also use their enormous purchasing power to demand more from A.I. companies and to build more A.I. with them. In return for providing government funding, South Korea is securing contracts with multiple tech and telecom companies to give A.I. access to all of its 51 million citizens. The companies will keep the basic version of the service (which includes access to three chatbots) free and unlimited, and run it mostly on Korean-built models. In Spain, public money has been used to develop and deploy a family of open-source A.I. models. States and cities in the United States should buy access to A.I. models jointly and build open models together, giving governments, communities and universities transparent, accountable and high quality A.I.
Even when governments rent commercial A.I. models, they can own the products public servants build using them. At Adelaide Botanic High School in South Australia, a teacher, Rebecca Weber, used the school system’s language model to build a tool that assesses students’ English proficiency in 52 seconds, compared with about 30 minutes, which the traditional way took. The tool is now being rolled out statewide for trained teachers. South Australia operates a $20 million fund that supports such pragmatic public sector experiments.
American states and cities are also beginning to share successful products. My students at Northeastern University, where I lead the A.I. for Impact fellowship, spend six months building A.I. tools with government agencies. Some have worked with the Massachusetts Federal Funds and Infrastructure Office to create a free tool that translates dense federal funding notices and walks staff through writing an application, helping towns without grant writers compete for money to fund lead-pipe replacement or road repair. Our program is now helping California, Colorado and Rhode Island develop their own versions of the tool.
We’re so busy debating whether A.I. will destroy humanity that we don’t ask how we might use it to deliver services, run elections, make laws and better listen to constituents. The evidence is all around us. What we need now is the commitment to use these tools in ways that really matter to people and to our democracy.
Beth Simone Noveck is the author of “Reboot: AI and the Race to Save Democracy.” She is a professor at Northeastern University where she directs the Burnes Center for Social Change.
The Times is committed to publishing a diversity of letters to the editor. We’d like to hear what you think about this or any of our articles. Here are some tips. And here’s our email: [email protected].
Follow the New York Times Opinion section on Facebook, Instagram, TikTok, Bluesky, WhatsApp and Threads.
The post How A.I. Can Boost Democracy appeared first on New York Times.
