Someday soon, the email response you get from a county official, the new report going to the Board of Supervisors and the county job description that catches your eye could all be generated at least in part using artificial intelligence.
Sonoma County officials have outlined those particular uses of artificial intelligence and others in a newly adopted policy meant to guide use of an expanding array of AI technology for certain types of government work.
The policy is designed to open up approved AI tools for departments and agencies while navigating the many pitfalls of using the technology — the largest concern being how to safeguard confidential information.
It is a discussion unfolding at local governments across the state and the county as well as within organizations like the California State Association of Counties and National Association of Counties.
“Generally, the focus has been about balancing engaging in these technologies with ethical concerns,” said Rachael Serrao, public affairs manager for the California State Association of Counties, a research, policy and lobbying group that represents the state’s 58 counties in Sacramento.
With the passage of its policy on Tuesday, Sonoma County has joined a handful of counties that have adopted a guiding blueprint. Others include Santa Cruz, San Benito and San Francisco, according to Serrao.
But other California counties are not far behind in developing AI policies of their own.
Sonoma County has been monitoring generative artificial intelligence tools since ChatGPT — a chatbot and virtual assistant created by OpenAI — launched in 2022, said Dan Fruchey, Sonoma County’s information systems director.
Generative AI can create content, including text, images, audio or video, when prompted by a user, according to the county’s definition.
“We knew this was something really big out there that we had to pay attention to,” said Fruchey, who also is a member of CSAC’s Artificial Intelligence Committee and has worked with the National Association of Counties’ AI exploratory committee.
Under the county’s policy, departments and agencies may use approved tools for tasks including writing emails, reports, policies and job descriptions, completing spreadsheet calculations and data analysis and developing or debugging code.
But those permitted uses come with big caveats — users must not input any confidential information or information that is too specific to the county, such as the details of a legal case, and users must review and fact-check any AI-generated information.
“There can be risks in using this technology, including uncertainty about ownership of the AI-created content and security and privacy concerns with inputting proprietary or confidential information about an employee, client, operations, etc. when interacting with the AI technology,” the newly adopted document reads. “Additionally, the accuracy of the content created by these technologies cannot be relied upon, as the information may be outdated, protected, misleading or fabricated.”
The policy also restricts departments from using approved AI to make decisions such as reviewing or selecting new hires or employees for promotion, reviewing proposals from potential vendors and determining health care and benefits.
The main driver behind local governments’ interest in AI is efficiency.
“It’s a huge timesaving tool to go ahead and use AI in the appropriate context,” Fruchey said. “So there are things that would normally take you minutes or even hours to go ahead and produce and (AI) can go ahead and produce in seconds.”
Offering an example, Fruchey pointed to a study he is working on about county employees’ use of physical phones. After stripping out information specific to Sonoma County, he used ChatGPT to analyze the “anonymized” data for any trends.
“It did it all in just a few minutes. It would have taken me hours to go through all that,” Fruchey said.
Beyond protecting sensitive information, another challenge in harnessing AI technology for government work is writing policy that will keep pace with AI’s evolution and with any legislation being debated at the state level.
“The primary challenge is that the technology’s landscape and implications are constantly evolving, requiring policies that are flexible and adaptable,” Serrao said in an email. “Additionally, the surge of state legislation this year has introduced an extra layer of complexity.”
Sonoma County’s answer to that challenge was to avoid baking specific AI tools into the policy itself. Instead, the policy refers to a separate list of approved tools and how they can be used. That technology suite includes ChatGPT, OnePlan, Microsoft Teams and Microsoft Azure.
Departments interested in using a tool not on the list can submit the technology for a review both by the information systems department and the County Counsel’s Office. The county is currently reviewing Microsoft CoPilot, another generative AI tool.
The county has ruled out certain tools, at least for now, including virtual agents, transcription services and a set of tools called Meeting Bots, which is a software that allows you to attend events virtually, Fruchey said.
Those tools were not compatible with the county’s need to safeguard confidential information, Fruchey said. But the county may revisit them as the technology evolves.
“We’re trying to go ahead and dip our toes into the pool and make sure that what we are doing is effective and could lead to other things that are valuable and supportable for the future,” Fruchey said.
Editor’s note: this article has been updated to correct the name of Meeting Bots.
You can reach Staff Writer Emma Murphy at 707-521-5228 or emma.murphy@pressdemocrat.com. On Twitter @MurphReports.
