San Francisco has adopted a machine learning tool to help streamline administrative work across the city’s bureaucracy, as well as guidelines for AI use, Mayor Daniel Lurie’s office announced Monday.
Microsoft 365 Copilot Chat, described as “an enterprise-grade artificial intelligence (AI) assistant tool,” will be made available across city departments as of July 14, beginning with a five-week training campaign administered by the city’s Department of Technology in cooperation with Microsoft.
“Copilot Chat will be available across city departments, helping staff accelerate more administrative work such as drafting reports, analyzing data, and summarizing documents — allowing them to spend more time delivering services directly to San Franciscans,” according to a statement released by the mayor’s office Monday. “Today’s launch is the first phase of the citywide rollout, with plans for the city to continue to evolve the use of generative AI tools based on employee experience and feedback.”
The move comes on the heels of the recent adoption of guidelines for the use of AI tools, work on which started two years ago under the administration of former Mayor London Breed, and the implementation of some limited pilot programs, such as chatbots for customer service and a tracking system for cars in the parking garages of San Francisco International Airport. San Francisco joins cities such as Boston, Seattle, and San Jose in adopting such guidelines. San Jose first published their policies in 2023.
The San Francisco Civil Grand Jury released a report earlier this year urging faster adoption of the technology. Around 3,000 city employees also participated in a pilot program, running from late 2024 to earlier this year, which gained them up to an extra five hours per week in productivity on tasks such as email management, research, and planning.
San Francisco’s guidelines include principles based on goals of improving response and quality of service for city residents, making the city workforce a leader in the use of AI tools, and robust guardrails to ensure privacy and eliminate bias. But tech watchdogs continue to have concerns over the use of generative AI tools for what should be human-centered decision-making.
“The days of coming in on Sundays to do TPS reports are over.”
“The city’s guidelines state that when AI is used for decisions related to services, enforcement, or eligibility, it can serve as a support tool, but should never make final decisions. They instruct workers to watch for bias. But bias detection requires statistical analysis. It’s not something an individual worker can do on a per-decision basis,” Jacob Hoffman-Andrews, senior staff technologist at the Electronic Frontier Foundation, told The Voice. “People using AI for decision support over-rely on it, favoring the AI’s decision over the one they would make independently. And (large language models) are general-purpose tools, not trained or tested for specific decisions like service eligibility. The city’s policy should be that chatbots may not be used to support such decisions, full stop.”
Members of the Board of Supervisors praised the development, a contrast from two years ago, when then Board President Aaron Peskin indicated that he was unaware of any use of generative AI tools at City Hall.
“City employees work incredibly hard, and they deserve the best tools to deliver the best services to San Franciscans,” commented District 2 Supervisor Stephen Sherrill in the city’s statement. “Whether it’s preparing presentations for public meetings or summarizing complex reports, everyone wants to spend more time communicating clearly and less time moving pixels. Copilot will help with that and more. The days of coming in on Sundays to do TPS reports are over.”