Donald Gainsborough brings a seasoned, strategic perspective to the intersection of policy and innovation. As the head of Government Curated, he has spent his career navigating the complex landscape of legislation and public service, focusing on how technology can serve the many rather than the few. Today, he shares his insights on a groundbreaking shift in civic technology: the use of artificial intelligence to bridge language gaps in government services. By analyzing the rigorous framework developed by the Minnesota Department of Human Services, Gainsborough provides a roadmap for how state and local leaders can move beyond the hype of large language models to deliver essential services—from emergency alerts to public assistance—with greater speed, cultural accuracy, and empathy for the diverse populations they serve.
When a state agency is looking at its vast library of public materials, how should leadership determine which specific resources truly warrant the integration of AI-enabled translation rather than sticking with traditional human-only workflows?
The decision-making process must be grounded in a rigorous assessment of both impact and necessity. In the civic tech world, we look at the scope and breadth of the work to see if traditional methods can realistically keep pace with the demand for accessibility. For instance, consider the Provider Hub in Minnesota, which serves as a vital platform for child care, mental health, and substance use disorder service providers to manage their professional licenses. Because the materials for this hub—including maintenance alerts, training modules, and newsletters—require updates and translations anywhere from 4 to 12 times a year, the sheer volume makes an LLM an appropriate tool to explore. We have to remember that supporting these providers with clear, timely information directly translates into better support for the residents seeking those critical services. It is about identifying those high-frequency touchpoints where a delay in translation could lead to a breakdown in essential community care.
Efficiency is a major driver for adopting new technology, but how do we quantify the actual impact that a hybrid human-AI translation model has on the day-to-day operations of a government office?
The numbers tell a compelling story of how we can reclaim time for staff to focus on higher-level tasks. When the Enterprise Translation Office was established in 2024, it was staffed with six multilingual experts who began leveraging ChatGPT to streamline their workflow for executive branch agencies. The results were immediate and tangible, with the time required for certain translations being cut effectively in half. For example, a project in Spanish or Somali that previously took two full hours of human labor can now be completed in just one hour, while Hmong translations have seen a reduction from four hours down to two. This isn’t just about saving minutes; it’s about the emotional relief of being able to respond to a community’s needs in real-time. By utilizing these tools for the initial heavy lifting, the staff can ensure that the “human touch” is applied where it matters most: in the final polish and cultural verification of the text.
Translation involves more than just a literal exchange of words; it requires a deep understanding of cultural nuance. How can agencies technically and linguistically prepare their content to ensure the AI output is actually useful?
Preparation is perhaps the most overlooked stage of the process, yet it is where the quality of the service is truly born. To avoid the robotic and often confusing literalisms that generic AI can produce, it is vital to distill English source material into plain, common language before it ever touches a large language model. We’ve found that by providing the LLM with prompts written in straightforward, jargon-free English, the resulting translations are much more direct and easier for the end user to digest. Additionally, we have to be strategic about what we choose not to translate; for example, proper names like the “Provider Hub” should remain consistent across all languages so that residents can recognize the resource regardless of their native tongue. It’s also a great opportunity to improve accessibility across the board, such as using the tools to generate accurate alt-text for graphics, ensuring that every resident, regardless of their visual or linguistic needs, receives the same high-quality information.
In an environment where a single mistranslation in an emergency or legal document can have serious consequences, why is a dedicated quality assurance plan the most vital component of this framework?
Quality assurance is the absolute heartbeat of this entire operation because it represents our commitment to the public’s trust. You cannot simply “set and forget” an AI tool when people’s livelihoods or health services are on the line. A robust plan involves repeated human reviews to ensure that the content is not only grammatically correct but also culturally appropriate and resonant. While this final review step takes additional time, it is an investment that pays dividends by drastically reducing the need for future revisions and preventing the confusion that leads to a flood of support requests. When a resident reads a translated document and understands it perfectly the first time, it validates their place in the community. That time spent on the back end doing quality assurance makes for a better translation, and ultimately, it ensures the end user feels seen and respected by their government.
What is your forecast for AI in civic technology?
I anticipate that within the next five years, we will see a shift where AI is no longer viewed as a standalone “special project” but becomes a standard, invisible layer of the government infrastructure that focuses heavily on personalization and hyper-localization. We will likely move beyond simple document translation and toward real-time, interactive government interfaces that can pivot between dozens of dialects and reading levels instantaneously. However, this progress will only be successful if it is paired with the kind of interdisciplinary frameworks we are seeing today, where the expertise of instructional designers and multilingual staff remains the guiding force. The future of civic tech isn’t just about faster processors or larger datasets; it’s about using those tools to create a more inclusive, responsive, and human-centric bureaucracy that truly speaks the language of every citizen.
