The Department of Defense’s (DoD) Chief Digital and Artificial Intelligence Office (CDAO) is continuing to make progress on its responsible AI efforts, including by issuing its Responsible Artificial Intelligence (RAI) Toolkit that is now available to the public.
The CDAO first published the RAI Toolkit in document form in September, but it was only available to DoD employees. Today, the toolkit is finally available in an interactive version for the public to use.
At the Intel Public Sector Summit in Washington, D.C., today, Deputy CDAO Margie Palmieri said the toolkit is just one way the CDAO is continuing to build trust in AI. The toolkit is a key deliverable of the RAI Strategy and Implementation Pathway (S&I), which the DoD issued in June 2022.
“[It] really gets into the mechanics of how do you think about responsibility in the context of artificial intelligence,” Palmieri said. “There’s a lot of pages around that. I like to distill it into three things … does the AI do what it’s supposed to do? Does it not do what it’s not supposed to do? And do our users, are they trained on how to use it appropriately?”
The CDAO said in a blog post today that it built the toolkit based upon other proven RAI and ethical AI frameworks and toolkits. The online tool provides industry partners with a voluntary process that improves the alignment of AI projects to RAI best practices and the DoD AI Ethical Principles.
The office also plans to continually improve the toolkit, a process that Palmieri said is crucial when it comes to AI technologies.
“We have a series of experiments throughout DoD, that are just normal and routine and we incorporate different aspects of AI into those, and then we do specific experiments and sprints on how to develop capability, put it out there, use it, test it, and get some feedback. And then once it’s out there – this is the key – it’s not done,” Palmieri said.
“So much of our traditional capability development and acquisition processes are around hardware systems, things you can touch,” she continued. “AI is much different. Even software, you can even go back to and you can refresh and you can update the code, but with AI, users have to constantly be involved. And our most successful AI projects are where there is baked into the user interface the opportunity to say, ‘That’s the right answer. That’s the wrong answer.’”
The CDAO’s RAI team said it plans to add other features to the toolkit “in the near future,” including RAI project management tools, an acquisition toolkit with standardized contract language, and a human-machine red-teaming guidebook.
The CDAO is welcoming feedback on the toolkit to best support the needs of the DoD community. The office said the best way to submit feedback is by emailing osd.pentagon.cdao.mbx.dod-rai-toolkit@mail.mil.
