Collaboration in Federal AI Enable Better Auditing, Climate Change Efforts

Federal agencies are all looking to ramp up artificial intelligence capabilities, and while it takes the right workforce, data, ethics, governance and development to successfully deploy them, federal officials also highlight the importance of collaboration.

These partnerships are cross-cutting, and federal officials from the Government Accountability Office, General Services Administration and State Department highlighted at GovernmentCIO Media & Research’s AI Gov event Thursday how they rely not only on collaborative relationships across agencies, industry partners and even international organizations and the public at large.

Cross-Agency Collaboration

GSA’s AI center of excellence is one hub within the federal government fostering a web of AI collaboration projects across agencies. About a year and a half after standing up center, AI Lead Bryan Lane said he realized AI is a “team sport” and is better achieved through partnerships rather than in silos.

The center has collaborated with civilian agencies, looking at their needs to kickstart AI projects and share best practices. The agency also launching communities of practice around robotic process automation and AI to create a cross-pollination of ideas. Most recently, GSA has partnered with the Defense Department’s Joint AI Center to help the military create its own AI development hub.

“One of our priorities right now is partnership in the DOD, and specifically, cultivating partnerships to provide capabilities and services to help the armed services in their AI adoption,” Lane said. “It’s collaboration around delivering AI capabilities — so that could be providing software or analytic tools, helping with data, supporting tests and evaluation activities, and even in some cases, prototype development delivery for scaling infrastructure.”

While GSA is working with both DOD and civilian agencies to develop and acquire AI capabilities, GAO is working across the government and with the federal oversight community to audit algorithms and AI tools to ensure that government-deployed AI is trustworthy, ethical and effective. Taka Ariga, GAO’s chief data scientist and director of its Innovation Lab, highlighted work with the National Institute of Standards and Technology in best practices for AI evaluation, from the implementation to the domain levels.

Throughout GAO’s partnerships, the Innovation Lab’s experimentation and numerous reports on AI, the agency is looking to publish its first AI Oversight Framework in June.

“It really is meant as a foundational document for oversight agencies, such as GAO, to start asking questions around performance, around societal impact that we are all very aware of and do so across the lifecycle of AI development, so that we can collectively build that trust across the implementations that are happening across the landscape,” Ariga said.

The State Department has also realized the value of collaborative efforts such as this in preparing for ethical and effective AI, the agencies’ Office of Management, Strategy & Solutions CTO Landon Van Dyke added.

“We’re part of the interagency group that’s focused on trying to implement some of the guiding principles for ethics and AI,” Van Dyke said. “We’re looking at how to make things auditable. How do we make it transparent, making sure that our AI implementations are purpose-driven, obviously respectful for citizens and the law? But we’re also doing a lot of preparation on the logistics, the infrastructure, cybersecurity, data storage strategies, sharing agreements between interagencies, just standardization of datasets and catalog and all of that.”

These partnerships involve the Office of Science & Technology Policy at the White House, as well as the Office of Management and Budget and the National Security Council, Van Dyke said. As part of State’s climate change-driven efforts, the agency is collaborating with NASA, the Environmental Protection Agency and National Oceanic and Atmospheric Administration to share data and AI tools to monitor air quality and build equipment for the Energy Department.

Citizen Science Data

The State Department is also taking some of the lessons learned it has gathered from federal partnerships and is applying those best practices on the international stage. Working with the Wilson Center and the United Nations, the State Department gathers
internet of things” sensor data from all over the world to applying AI models and gather insights about climate change and climate issues, Van Dyke said.

State involves the public through his team’s work in standardizing citizen science data. Although the science community creates a good amount of data, Van Dyke said incorporating the citizen science community can improve datasets to further heighten the work in climate change science.

“We’re working with lots of different environmental groups that are collecting this [citizen science] information, including NOAA and NASA and seeing what we can do to improve the datasets and make them more usable to the science community and adding the extra metadata, adding the extra origin story of where this data came from and coming up with different ways where we can cross-validate this information,” Van Dyke said. “Once that information is brought into the science community with all this information, it makes it more AI ready. It makes it more interoperable and easier to digest, and it’s easier for them to align this type of data with other datasets that they already have from satellite or from land-based or satellite-based sources.”

Van Dyke is working with people on the ground in different countries to consolidate datasets and bringing that information back to labs to apply to machine learning to recognize, for example, differences in air quality by cross-referencing photos of the sky taken by the public and cross-referencing information from those photos from sensor data.

The Role of Industry

Throughout these federal and international partnerships, industry also plays a role in improving the AI landscape for agencies.

Lane, Ariga and Van Dyke all highlighted ways in which targeted capabilities are one of the greatest areas that industry partners can help with, ensuring commercial AI tools are aligned with cybersecurity and architecture at their respective agencies.

For an agency like GAO, for instance, understanding the agency’s requirements around the appropriateness of models and algorithms they use are important because of the agency’s AI oversight role across other agencies, Ariga said. Avoiding replicability with the AI that GAO audits is critical for industry partners to understand.

Lane further highlighted the innovative work in new acquisition strategies that the acquisition workforce has developed has helped smooth over barriers to collaboration that agencies have often encountered in the past.

“Oftentimes, having the right capabilities and the right vendors at the right time is a challenge, and I’ve seen the federal acquisition workforce really rise to that challenge and figure out how do we get the timing and capabilities and industry partners all in sync,” Lane said. “I think we’ve done a tremendous amount of work there in terms of both traditional and non-traditional acquisition.”