Asked and Answered: What is preventing local governments from using GenAI?
by: Kate Garman Burns, MetroLab Network, Executive Director
Last Monday, the Biden Administration issued its most thorough and detailed policy approach on artificial intelligence yet, an Executive Order providing guidance and guardrails on important issues including civil rights, cybersecurity, and research. While it focuses on federal agency policies and congressional efforts, the EO does impact how other levels of government will interface with AI and regulate this transformative technology.
MetroLab Network, a non-profit aiming to equip local governments with science and research, has formed a national task force to produce local government policy guidance on GenAI. With over 45 unique local governments involved, we meet regularly to consider how best to use GenAI, while ensuring it’s used in a just and equitable way. The most common question we are hearing is “how are local governments using it?” And while some use cases can be used right away, there are good reasons to wait. At one of our recent meetings, we asked the question:
What is preventing me from using a GenAI use case? Here’s what we heard and how the Executive Order addresses these concerns.
Local governments are hesitant to use GenAI without proper policies in place first. Community trust is paramount. How will community members know that proper parameters have been put into place without published policy and regulations. [note: this is a step beyond general guidelines for staff use. There is a need for concrete rules and procedures to be put into place.] As Congress, state legislators, and city halls work on AI regulations, public trust will grow. However, without something concrete into place, this will likely be the single most significant barrier for the government to use AI. We expect to see local governments formalizing processes and policies as federal efforts continue.
Privacy concerns. What data sources are being used to teach large language models and generating outputs? Do these sources include personally identifiable information? How can local governments have assurances on this? The EO addresses this pretty well — in fact the word “privacy” is used 38 times. However, most of the actions directed by President Biden pertain to actions by federal agencies. It would be extremely helpful for the National Science Foundation, the Office of Management and Budget, and the Office of Science and Technology Policy to share these new policies for state and local governments to learn and use these protocols and resources.
Lack of diverse vendors. This is an extremely nascent technology. Are multiple options available for vendors and procurement considerations (including women and minority owned businesses)? There may also be biases in the technology provided by private vendors that may impact how it is used for public purposes. The EO addresses this pretty specifically, including supporting small businesses and the enablement of diverse private sector groups to use AI. Bravo. This will increase the marketplace and address this concern head on.
Technology capabilities and lack of common definitions. Local governments have different definitions for “common” words. For example, just ask a procurement officer what a “pilot” means. There are also really important differences in how communities and our residents define terms. Consider definitions on terms such as “equitable,” “community engagement,” or “access.” There must be some assurance that context can be provided for important vocabulary.
There was also concern with respect to accuracy. “Hallucinations” or false outputs from genAI can create outright wrong outcomes or answers. The responsibility to provide accurate information as a government agency is a serious one, and a level of confidence in accurate outcomes when using this technology.
The EO couldn’t possibly “fix” contextual definitions like the examples cited above. However, by providing an entire section of definitions (in Section 3) with terms like “differential-privacy guarantee” and “privacy enhancing technology,” this is a significant contribution to AI policy development. Section 3 definitions are now for the taking with state and local governments, creating more consistency in AI regulatory approaches.
Staff Capacity and expertise. Finally, the task force voiced concerns about staff capacity and expertise with respect to onboarding new tools. While GenAI can ultimately reduce staff capacity, there is a significant uptick in resources required to onboard new tools and policies. “We don’t want to overwhelm IT.” The EO calls for more AI expertise within federal agencies. We urge Congress and federal grant making agencies to consider programs for state and local governments to support this exact goal across all forms of government.
Local Government’s Approach to GenAI Policy
Ultimately, this is a complicated technology to use and to put parameters around. First, GenAI is a how, not necessarily a what. GenAI is a technology that can transform how we do things. It’s not necessarily the end result. For example, GenAI is a tool in which we can improve services and simplify processes. And while it can be an end product (like a chatbot), often it is the means to an end. Local governments are not procuring genAI, they are procuring services or technologies that use it. This elevates the complexity of how local governments can regulate this technology.
Second, there is a level of education and understanding on how the technology works that is required to enable governments to develop policy and consider use cases. This is a huge undertaking. When Uber and Lyft deployed in cities nearly a decade ago, local governments did not need to learn the engineering and mechanics of car engines. This is where MetroLab especially hopes to assist local government leaders: by bringing in expertise and research capacity through partnerships with universities. The education and research components will inform policy development step-by-step.
There is no doubt that this technology can transform how governments everywhere deploy its services. What is quickly becoming a realization is that policy is required first, then use cases. The Executive Order is a significant and meaningful beginning. We must work to ensure all levels of government are supported in developing necessary policy so that the incredible potential of AI can transform government services and safely unlock the potential this technology provides.
— — — — — —
About MetroLab’s National GenAI for Local Governments Task Force
MetroLab Network is a nonprofit in Washington DC that aims to equip local governments with science and research. It serves as a convener of an emerging academic practice focused on integrative, use-inspired, community-focused research, done in partnership with local government and communities.
The “In the Lab” program supports a national effort for practitioners, by practitioners, to produce policy guidance on emerging policy needs. Launched in August 2023, the GenAI for Local Governments Task Force convenes 130+ individuals to develop policy guidance that will include model ordinance language, processes and regulatory recommendations, and a resources library. The task force is currently composed of 45 unique local governments, 16 universities, 25 private sector companies, and 16 other stakeholder groups including four federal agencies, nonprofits, and coalition organizations.
