Approximately one in six children in New Jersey lives in a household facing food insecurity. Since 2024, the federal government has provided eligible families with $120 per child to assist with summer grocery expenses. Eligibility is determined through various safety net programs, each managed by different agencies, which has made it difficult to identify all qualifying children. To address this, New Jersey’s Office of Innovation developed an artificial intelligence (A.I.) system to cross-reference records across agencies and identify children who had not yet received benefits. As a result, the state extended summer grocery assistance to an additional 106,000 children during the program’s first year and 96,000 in the second, without requiring families to complete any new paperwork.

Beth Simone Noveck, New Jersey’s inaugural chief A.I. strategist, views this initiative as an example of how A.I. can improve government services and rebuild public trust in democratic institutions. While acknowledging the risks of unchecked A.I. deployment, she argues that the technology’s potential to enhance civic operations should not be overlooked.

Beyond New Jersey, other governments have embraced A.I. to improve public service delivery. For instance, after California’s 2025 wildfires, officials utilized A.I. to analyze over 1,500 community responses on recovery needs, shaping plans to bury power lines and enhance emergency communication systems. Governor Gavin Newsom recently signed legislation institutionalizing this A.I.-driven approach. Similarly, the British government reported that its Consult platform quickly processed more than 50,000 public comments within two hours, a task that would typically require weeks.

To deepen A.I.’s integration into government functions, experts recommend three primary strategies. First, comprehensive training programs are vital to equip public employees with the skills to effectively and safely use A.I. tools. Pennsylvania’s training of 175 state workers in using ChatGPT reportedly saved an average of 95 minutes per day in drafting and research tasks. Lack of proper training, however, can lead to serious errors, as illustrated by an incident in Victoria, Australia, where a child protection worker's misuse of A.I. resulted in a flawed report.

Second, providing government workers with access to multiple A.I. models can reduce dependency on a single provider and allow for comparison and selection of the best tools for varied tasks. Third, governments have the leverage to demand more transparent and accountable A.I. solutions by partnering with technology companies and supporting the development of open-source models. South Korea, for example, has secured agreements with multiple firms to provide A.I. services to its entire population, while Spain has invested in open-source A.I. deployments. U.S. states and municipalities are encouraged to pool resources to build and acquire similar models.

Even when using commercial A.I. services, governments retain ownership of the end products their employees create. In South Australia, an English teacher developed an A.I.-powered tool to assess student proficiency in under a minute—a process that traditionally took half an hour. This tool has since been expanded statewide with support from a dedicated public sector innovation fund.

Collaborative efforts are also taking shape across the U.S. At Northeastern University, students involved in the A.I. for Impact fellowship partner with state agencies to build tools that simplify access to federal funding for smaller municipalities. These projects are currently benefiting states including Massachusetts, California, Colorado, and Rhode Island.

While much public discourse emphasizes existential risks posed by A.I., proponents like Noveck stress the immediate, tangible benefits the technology offers to government efficiency and democratic responsiveness. They advocate for a balanced approach that combines caution with proactive adoption to address pressing societal challenges.