Commerce looking to publish AI-ready data guidance in coming months
Guidance from the Department of Commerce aimed at establishing a first-of-its-kind framework for using the agency’s public federal data with artificial intelligence tools could come in the next several months.
Victoria Houed, director of AI policy and strategy to the undersecretary for economic affairs at the Department of Commerce, said the agency is aiming to publish that guidance by the end of the year or, if not, January 2025. It will address topics such as documentation, improving metadata, licensing, usage, quality and integrity, Houed told reporters following a panel discussion about AI and data policy at the Data Foundation’s govDATAx event.
While focused on the Department of Commerce, the document could end up being useful to other agencies as they also work to get their own information ready to be used with AI tools.
“No one has created guidance like this before,” Houed said during the panel discussion. “And our hope is that we’re going to publish it publicly and that other agencies and any other open data publisher will be able to benefit from our framework and help us iterate on that.”
Efforts to create such guidance and best practices began around April with a request for information by the AI and Open Government Data Assets Working Group within the department’s Data Governance Board. That comment period closed in July.
That initial search was an effort to see what was already out there in terms of AI-ready data guidance, according to Houed. “We pretty much were searching high and low for any sort of AI-ready framework, which, at the time, didn’t exist,” she said.
Since that response period concluded, the Data Foundation has released its own guide to support use of data for AI, which Houed praised. She also noted that Patrick McGarry, general manager for U.S. federal at data.world and a member of the Data Foundation’s AI Working Group, has been helping on that effort as well. McGarry moderated the discussion.
One area that the department has been thinking about a lot with respect to public data being integrated into AI systems is copyright, licensing and usage rights, and that is a portion of the coming framework, Houed said later in the discussion.
That “has not really been effectively figured out just yet — about who can use the data and for what,” Houed said. She said the department is trying to figure out how to signal usage information within the metadata and documentation, which are generally sources of information that provide more information about a set of data itself.
Under the Evidence Act, the current standard for federal data is machine readability, but the new guidance will aim to go beyond that to improve the AI-readiness of that information. Speaking on a June panel, for example, Commerce Chief Data Officer Oliver Wise said “machine readability is a necessary but not sufficient condition to really meet user expectations in the AI era.”
The guidance will also likely carry weight given that the Department of Commerce, which has the moniker of “America’s Data Agency,” houses several of the government’s most prominent agencies for federal data. Those agencies include the U.S. Census Bureau, the National Oceanic and Atmospheric Administration, and the Bureau of Economic Analysis.
The panel discussion largely underscored barriers that agencies listed in their recent compliance plans for a White House AI memo (M-24-10). A FedScoop analysis of a sampling of those plans found agencies frequently cited data and data readiness as a barrier to responsible AI implementation.
In that same vein, Paula Osborn, deputy chief data and AI officer at the Department of State, who was also on the panel with Houed, pointed to legacy systems as a challenge to apply AI frameworks, noting that it’s “hard to get the data to be the quality that we needed it to be.”
Other challenges for the department include data sensitivity and transparency, Osborn said. Some State Department data have security risks, “but frequently, a lot of our data will get sort of over-classified,” she said, adding that in other cases people don’t want any of their data released.
Ultimately, when asked about what is standing in the way of data and AI, Osborn pointed to funding needed to overhaul or replace legacy systems. “It’s such a resource-heavy initiative that we don’t have the budget to do,” she said.
Meanwhile, Houed said it would be helpful to have a high-level federal data strategy that includes AI readiness, defines what that means, and provides a technical toolkit to agencies for approaching those tasks.
“A lot of the answers should be answered up front, instead of us all having to do our own digging and investigating of how to be … producing high-quality data,” Houed said.
