AI translation tools require ‘due diligence’ from state and local leaders, expert says

A new Washington law that went into effect this month calls for state leaders to create standardized language access guidelines in a bid to enhance the government’s service delivery. 

The Washington bill, signed into law by Gov. Bob Ferguson in March, directs the state’s Office of Equity to develop guidelines for agencies to ensure the consistent and effective delivery of state-administered content, such as written, verbal, virtual or recorded communications. 

A similar effort is underway in California after lawmakers passed a bill late last year that aims to enhance the public’s access to and participation in state and local legislative meetings. Starting July 1, for example, agencies must offer multilingual translation services for public meeting agendas and other communications.   

The bills reflect a longstanding priority among state and local leaders across the U.S. to expand language accessibility, including translating real-time events or written content, for increasingly diverse communities. But the cost and difficulty of hiring human interpreters presents major challenges to many jurisdictions hoping to help residents better understand critical announcements, public meetings and other resources. 

Artificial intelligence has been touted as a solution to optimize translations for government settings, said Suresh Venkatasubramanian, director of Brown University’s Center for Tech Responsibility. But the technology should not be treated as a blanket solution to language accessibility. 

Wisconsin policymakers, for instance, introduced legislation last year that would have allowed state and local organizations to permit the use of AI translation tools instead of human interpreters in courtrooms. The bill, which received pushback from observers concerned that AI interpretations were not suitable for replacing human knowledge and judgement, died earlier this year after it failed to pass both chambers. 

But recent survey findings suggest growing confidence in AI-powered translation tools among professionals who manage meetings and events. The data, based on more than 200 respondents, found that nearly 95% of professionals agreed that AI translation solutions are easier and more affordable than human interpreters. At the same time, nearly 80% of respondents said the share of non-English speakers is increasing at events. 

The report, commissioned by software company Wordly and conducted by Dimensional Research, also found that 66% prefer AI-powered translations and only 25% said they still prefer human interpreters. 

Those findings are particularly valuable for state and local leaders who may have been hesitant to use AI for translation services thus far, said Dave Deasy, chief marketing officer of Wordly.

As more jurisdictions deploy AI tools to translate communication during public meetings, public safety broadcasts and other events, “we have found with a lot of communities, as they have started to offer more language support, they’re actually getting more [community] participation,” Deasy said. 

That means more individuals can, for instance, better understand and comply with policy changes to benefit programs they’re enrolled in, developments in immigration enforcement laws or participate in feedback opportunities like public comment periods, Deasy said. 

While language accessibility and inclusivity is a vital function of government, AI-based solutions to language gaps should not come at the cost of responsible adoption, testing and implementation of the technology, Venkatasubramanian said. 

“Serving a population is not just about the cheapest way to do it, but the most efficacious way to do it,” he said. For instance, the common argument for AI translation tools is their potential to reduce expenses for hiring and retaining human interpreters, which could be tempting for governments already facing limited funding and staff capacity, he explained. 

Realistically, state and local leaders should prepare to spend the appropriate money and time to vet AI translation services from vendors and ensure outputs are accurate and relevant to local community needs, Venkatasubramanian said. 

He suggested governments and agencies looking to adopt AI translation tools to start with temporary contracts that enable them to pilot such services. This approach can also help government leaders exercise more leverage to customize and evaluate AI tools based on their unique needs, he said. 

Jurisdictions often have specific terms for their geographic locations, department titles, documents and other government-affiliated language, and officials must ensure a proposed AI model can address and maintain such specificities across languages, he explained. That is where an AI product that worked for one city may be less effective for another, making it crucial for state and local leaders to conduct their own evaluations of such solutions. 

Venkatasubramanian also underscored the value of human interpreters as part of the process of adopting and deploying translation tools. While AI-powered translations are relatively accurate and successful for one-on-one conversations, like between locals and tourists, government-based use cases require heightened scrutiny when a resident’s safety and well-being could be on the line. 

Indeed, human interpreters can be especially valuable when it comes to assessing the accuracy and validity of AI-powered translations, as they can better understand nuances and sentimental differences between languages, Minnesota officials said at Code for America’s 2026 Summit last month in Chicago. 

During a breakout session at the conference, officials for the Minnesota Department of Human Services also shared a framework to help government leaders assess whether or not AI is the right solution to language translation services. 

Venkatasubramanian also pointed to Boston as an example of a municipality taking a “very thoughtful and careful” approach to AI’s use in government, including AI-enabled translations. In 2023, city officials released generative AI guidance that directs municipal staff to limit their reliance on the technology for conducting translations. 

For instance, users should keep in mind the limitations of the technology because “it is not well documented the extent to which ChatGPT and other models can use other languages,” according to the guidance. The document also encourages users to avoid using content generated in a language they are not familiar with without consulting someone proficient in the language and to ensure AI models can consider different regional dialects within the same language. 

Ultimately, an AI translation tool “is going to be much better for you in the long run, if you have a solid [AI] system that is actually tuned to your context,” Venkatasubramanian explained, adding that “you have to do your own due diligence.”