What Do State Governments’ Public AI Inventories Reveal?

From wildfire detection to unemployment benefits, AI is shaping decisions in state government. Now, at least 13 states are creating inventories or repositories to track which agencies are using the tools, and how.

A couple states are taking a big step forward in transparency by publicly posting the inventories that detail which tools are used by who and for what.

Government Technology analyzed the public AI inventories posted by Vermont and Connecticut to uncover what exactly the inventories reveal about the use of AI in state government.

Vermont’s AI inventory is included in the Agency of Digital Services Annual Report. The report, published in January of 2024, included an inventory with 16 different AI tools identified.
The inventory reveals that the state's Agency of Digital Services uses the most AI tools, primarily for purposes such as strengthening cybersecurity and threat detection.

The state's Agency of Transportation also uses several AI tools to classify roadway pavement quality and support decisions for project prioritization and selection as well as funding requests. The same department also uses AI to inventory roadway signals and support decisions for sign replacement.

Vermont's Agency of Administration uses an AI tool called Gen TAX to process tax information and identify fraud risk in an effort to reduce tax fraud in the state.

Across the state, the most common benefits from using AI noted in the inventory were improving efficiency (six), lowering costs (two) and improving security posture (two).  

Vermont's inventory also tracks how artificial intelligence tools impact citizens. At the time the inventory was published, only two sets of tools were identified as having a direct impact on citizens.
The first is unspecified security tools used by the Agency of Digital Services to detect and stop intrusions, attacks and malware in which AI makes final decisions for identification, prevention and resolution of security issues. The second tool, under the use of the same agency, is multifactor authentication (MFA) tools that use AI to determine final decisions for the MFA prompt, a process intended to make “robust authentication mechanisms less cumbersome for some users.”

Vermont’s inventory also requires agencies to report if the AI tools have been bias tested. At the time the report was published, most had not.
The tools the agency reported as being bias tested included Bing Copilot and Azure OpenAI.

Meanwhile, Connecticut’s AI inventory lives in the state’s open data portal, where according to its metadata, it's been downloaded more than 600 times. It will be updated annually, and the latest information was added in December 2023 when there were 10 AI tools listed in the inventory.

Connecticut's AI inventory is detailed, documenting any variation of automated tools, including services such as Zoom, WordPress, Microsoft Teams and Microsoft Office. However, the inventory also includes AI tools the general public would be less familiar with, such as Abnormal Security, a product used by the state to provide real-time detection and response capabilities for email-based threats.

Connecticut’s inventory also requires the state to disclose if an AI impact assessment has been completed for any of the tools in use. As of December 2023, none of the state's 10 AI tools in use had been assessed.

It also requires a determination of whether the AI tool has the power to make decisions in the state. While most of the entries in the current inventory do not, a few do, such as providing real-time detection and response to a wide range of cyber and email threats.
As several states are still building their AI inventories, it remains to be seen how many will post their lists publicly, and whether AI inventories will be considered public record and a deliverable in the case of a public data request.