Government agencies have long made consistent efforts to provide translation services, but Minnesota, like many states, faces challenges. The state’s translation process was highly decentralized, with each department relying on various methods, including engaging external contractors, leading to inconsistent quality, high costs, and delays that could extend up to a month for a single request. Interpretation services were equally challenging, where limited access to approved interpreters made it difficult for people to engage with critical legal and community services effectively.

To address these issues, the ETO began using AI, incorporating ChatGPT as a key pillar of their solution to provide translation services faster, more accurately, and more equitably. The ETO collaborated closely with the Transparent Artificial Intelligence Governance Alliance (TAIGA), Minnesota’s IT Services program, to ensure the program followed safe, ethical, and transparent AI principles. By integrating ChatGPT into the initial stage of translation with meticulous human review and quality assurance, the ETO established a more efficient workflow that sets a new standard for language inclusion in the public sector.

As Adam Taha, Director of the ETO, describes, “The balance between accuracy and speed is the crux of the problem.” By incorporating AI-driven solutions and especially ChatGPT, the ETO has been able to address this challenge, enhancing both the speed and accuracy of its translation services.

Beyond translation accuracy, the ETO’s ChatGPT‑enabled workflow also optimizes for cultural relevance to ensure government messaging is both accessible and inclusive. For example, with Hmong and Somali, the ETO developed spreadsheets with culturally relevant terms and translations and incorporated these into their GPTs to improve the model’s cultural awareness. By incorporating a process where ETO translation experts review and correct ChatGPT translations before uploading them to custom GPTs for future reference, the ETO significantly speeds up the quality assurance stage, with fewer edits required for each translation, ensuring that no mistakes or discrepancies occur twice.