Subnational Practices in AI Policy: A Working Guide
Summary
The document examines how cities, states, and provinces are navigating AI policy, noting a spectrum of engagement from active experimentation to cautious moratoriums. It highlights four primary goals for subnational AI policy: increasing government efficiency, promoting equity and transparency, influencing industry via procurement, and attracting economic opportunities. The guide details specific practices such as the use of AI registers for accountability in Helsinki and San José, and the implementation of 'responsible experimentation' guidelines in Boston. It also contrasts these local efforts with national strategies in China, the US, and the UK, and supranational regulations like the EU AI Act, arguing that subnational actors currently lack influence over these higher-level frameworks despite their role as testing grounds for policy innovation.
Key insights
- Subnational governments exhibit a wide spectrum of engagement with AI, ranging from active experimentation with tools for traffic management and service delivery to a 'wait-and-see mode' or outright moratoriums, such as the six-month ban on AI for state business implemented by Maine in 2023.
- Subnational AI policy is driven by four primary goals: improving the efficiency and public trust of government services, ensuring equity and transparency to prevent bias, using engagement to influence industry providers, and positioning their regions as attractive hubs for AI economic opportunities.
- To ensure accountability and transparency, some cities have implemented public AI or algorithm registers. Helsinki's AI Register allows citizens to see what data is used and how algorithms affect their lives, while San José's Algorithm Register logs every procured AI system along with its equity standards and oversight mechanisms.
- Subnational jurisdictions are leveraging their procurement power to shape the AI market. California's September 2023 Executive Order directs agencies to reform procurement to consider risks and training, while San José uses a 'Vendor AI FactSheet' to vet systems before implementation.
- There is a disconnect between subnational policy activity and global governance; while local officials use national and international frameworks for guidance, they currently lack influence over the creation of those frameworks, unlike their previous experiences with climate change and sustainability policies.
- Subnational entities are adopting diverse knowledge-gathering practices to bridge the internal expertise gap, including the creation of 'sandboxes' for testing (as seen in Utah and California), the establishment of expert advisory councils (Barcelona), and the use of 'inventories' to track AI use-cases (Connecticut).
Cite the original document
- APA
- Klaus, I., & Polsky, B. (2023). Subnational Practices in AI Policy: A Working Guide. Carnegie Endowment for International Peace. https://carnegieendowment.org/research/2023/12/subnational-practices-in-ai-policy-a-working-guide
- Chicago
- Klaus, Ian, and Ben Polsky. Subnational Practices in AI Policy: A Working Guide. Carnegie Endowment for International Peace, 2023. https://carnegieendowment.org/research/2023/12/subnational-practices-in-ai-policy-a-working-guide.
- Wikipedia
- {{cite report |last1=Klaus |first1=Ian |last2=Polsky |first2=Ben |title=Subnational Practices in AI Policy: A Working Guide |publisher=Carnegie Endowment for International Peace |date=12 December 2023 |url=https://carnegieendowment.org/research/2023/12/subnational-practices-in-ai-policy-a-working-guide |access-date=5 November 2026 |via=Climate Insights Directory (document carnend-000212)}}
- BibTeX
- @techreport{klaus2023subnational, author = {Klaus, Ian and Polsky, Ben}, title = {{Subnational Practices in AI Policy: A Working Guide}}, institution = {Carnegie Endowment for International Peace}, year = {2023}, month = dec, url = {https://carnegieendowment.org/research/2023/12/subnational-practices-in-ai-policy-a-working-guide}, urldate = {2026-11-05}, note = {Indexed by Climate Insights Directory, document carnend-000212} }
Full text
Collected · Record updated