The state’s new AI translation services for applicants seeking unemployment assistance is showing good results, according to Gillian Gutierrez, senior adviser and director of New Jersey’s unemployment insurance modernization.
Before officials deployed that tool, one of every 120 applications were made in Spanish. That figure has increased to an estimated one of every 44, she told Government Technology.
The work is part of a broader effort to bring more “human-centered design” into the state’s unemployment insurance system — a larger trend in government technology that focuses on platform improvements centered around how people really use websites.
More than one-third of New Jersey residents speak Spanish at home, according to state figures, so focusing on that language was a no-brainer. But there are older ways of achieving that goal and newer ways.
“We decided that the experience should be naturally bilingual, and not dependent on automated translation services,” Gutierrez said, adding that better translation for state services advances the goal of digital equity.
The state, which has a relatively active AI task force, and a chief AI strategist, used support from Google and the nonprofit U.S. Digital Response to produce materials and tools to create cutting-edge AI “translation assistants,” according to a recent statement describing the effort.
The materials rely on large language models for training and “off-the-shelf generative AI models” to help widen access to services for people more comfortable with Spanish than English.
New Jersey has now made those training materials available to officials in other states.
“With these training materials, government agencies across our state and the nation can begin leveraging the transformative potential of generative AI to make services more efficient, accessible, and responsive to the needs of all residents — regardless of what language they speak,” said New Jersey Chief AI Strategist Beth Simone Noveck in the statement.
This way of combining artificial intelligence and translation can “teach” websites to grasp the difference between, say, “laid off” and “fired,” according to Gutierrez — vital distinctions for those seeking unemployment pay and associated benefits.
The tool also can help the state — and, presumably, other agencies — deal with a dearth of human translators, who are often called upon for more tasks than they can reasonably complete during a normal work day.
“There are only so many people who can speak Spanish in our call centers,” she said, and that goes even more for the other 128 languages reportedly spoken in that state. “There are only so many people who can speak Haitian Creole.”
The training materials from New Jersey can reduce costs when compared to similar services offered by software vendors, state officials boast in the statement. Those materials also can save other public agencies from signing “long-term contracts that lack flexibility and adaptability over time,” according to the statement.
Backers of this effort have big hopes for what it could lead to — hopes that come as other states are leaning on AI to improve their own unemployment services.
“These training materials will enable any government to easily create their own AI translation assistant for unemployment insurance and other benefit programs,” said Marcie Chin, product delivery manager for U.S. Digital Response, in the statement. “It’s a prime example of how we can democratize generative AI for the public good by inviting people with lived experience navigating public benefits to participate in the design process."
