TLDR: City officials across Washington state are extensively utilizing generative AI tools like ChatGPT for a wide array of government functions, including drafting official correspondence, social media content, and policy documents. This widespread adoption, revealed through public records requests, highlights the technology’s potential for efficiency but also raises significant concerns regarding transparency, reliability, authorship, and the need for clear regulatory guidelines.
City officials in Washington state have widely adopted generative artificial intelligence (AI) tools, such as ChatGPT, for various government tasks, according to a recent investigation by Cascade PBS and KNKX. Thousands of pages of ChatGPT conversation logs, obtained through public records requests from nearly a dozen Washington cities, indicate a pervasive integration of AI into local government operations.
The investigation, focusing primarily on Bellingham and Everett due to their comprehensive responses, reveals that AI is being used for tasks ranging from drafting mayoral letters, social media posts, and press releases to generating policy documents, talking points, speeches, and responses to audit recommendations. Officials have also leveraged AI for grant application materials and replies to constituents’ emails.
One notable instance involved Bellingham Mayor Kim Lund’s office, where an assistant used ChatGPT to draft a letter of encouragement to the Washington Department of Commerce for a grant application by the Lummi Nation. While the final letter was not a verbatim copy, approximately half of its sentences mirrored the AI’s output. Mayor Lund acknowledged using both ChatGPT and Claude for ‘standard email communications, proclamations, supporting public speaking,’ and stated that she now ensures she is logged in to save chat histories, which she previously considered ‘transitory.’
Everett Mayor Cassie Franklin emphasized that AI is a valuable tool for efficiency, stating, ‘I think that we all are going to have to learn to use AI. It would be silly not to. It’s a tool that can really benefit us.’ She noted that AI helps compensate for recent staffing reductions in departments like communications, which ‘been the first to want to embrace AI to help generate enough content.’ Franklin also highlighted financial motivations, citing Washington’s 1% cap on property tax increases, which limits budgets and makes AI a way to free up staff time.
Beyond basic communications, city staff have entrusted AI with more complex responsibilities, including researching enterprise software, creating evaluation matrixes for contracts, summarizing court cases and legislation, providing feedback on policy, and synthesizing large volumes of public comments. For example, a Bellingham planner used ChatGPT to seek recommendations for updating sections of the city’s comprehensive plan, including searching the internet for information and suggesting figures.
However, the widespread use of AI has ignited debates over transparency, authorship, and the reliability of the technology in critical government functions. While state guidance suggests labeling AI-generated government documents, the reviewed records showed no such disclosures. Mayor Lund indicated that while labeling has been discussed, she believes it may become unnecessary due to AI’s increasing ubiquity.
Concerns about AI’s accuracy, often referred to as ‘hallucinations,’ are prominent. Records illustrate instances where ChatGPT fabricated data, referenced non-existent state laws, or cited documents that do not exist. An Everett financial official, upon receiving an AI-generated response with incorrect information, retorted, ‘You should not be quoting paragraphs that have totally different information in them.’ Similarly, a staffer working on Everett’s 2044 Comprehensive Plan reprimanded ChatGPT for ‘repeated mistakes’ in analyzing racial disparities in housing data, instructing it to ‘Do not hallucinate.’ Despite these errors, the City Council approved updates to the comprehensive plan that included large paragraphs of ChatGPT’s output, with Mayor Franklin expressing confidence due to multiple layers of oversight.
Anna-Maria Gueorguieva, an information science Ph.D. student at the University of Washington, voiced concerns that unlabeled AI-generated content in public communications could erode public trust, which is already ‘extremely low.’ Jai Jaisimha, co-founder of the Transparency Coalition, echoed this, suggesting that AI’s tendency for ‘sycophancy’ and ‘confirmation bias’ could be problematic when analyzing public comments, potentially distorting public sentiment.
In response to these challenges, Everett’s IT department issued guidance in July 2024, mandating that all AI-generated material ‘released to external audiences for the purpose of public policy decision making should be clearly labeled.’ However, consistent adherence to this guideline has been lacking, with examples like Mayor Franklin’s AI-generated letter to U.S. Rep. Rick Larsen, which made no mention of AI involvement. Going forward, Everett is transitioning staff to Microsoft Copilot for security and compatibility, requiring exemptions for other AI tools.
The investigation also uncovered more unconventional uses of AI, including staff seeking help with personal matters like declining a birthday invitation or managing tricky workplace situations. Some messages even revealed an awareness of AI’s ethical implications, with one staffer asking if a professor could detect AI-written essays, and another inquiring how to tell if a coworker was ‘addicted to AI.’
Also Read:
- UAE Finance Sector Leads Global AI Adoption, Yet ROI Remains a Challenge
- Generative AI Poised to Revolutionize Agriculture, Retail, and Wholesale Sectors by 2025
This report is the first in a two-part series, with the second part expected to delve into the policy questions cities are confronting as AI adoption outpaces regulation.
