The State Department is beginning to see positive results from having taken on artificial intelligence and machine learning (ML) tech work in the earlier stages of their development, an agency official said this week.
Matthew Graviss, chief data and artificial intelligence officer for the Office of Management Strategy and Solutions at the State Department, discussed some of the department’s success during a webinar hosted by Federal News Network on April.10.
“We’re seeing the gains and the benefits of AI and machine learning on the mission side and on the management side,” said Graviss.
“Using image analytics that supported the International Criminal Court arrest warrant for Vladimir Putin is just a quick example [and] supporting climate negotiations at the climate summit in November is another example,” he added.
While the State Department has been quick to move to adopt AI and ML technologies, Graviss made it clear that putting in place well-trained personnel and human-in-the-loop strategies are essential to gaining full potential from AI technologies.
He cited one use case in which the agency applied the technology to a requirement that it declassify information after a 25-year waiting period.
“We were able to use machine learning in that case to an accuracy level of over 97 percent and reduced over 60 percent of the workload,” he said.
“And that is through human-in-the-loop training on data from previous decisions, testing and evaluating the models against a blind set of data, deploying it, sampling, and going through it,” said Graviss.
“That took guardrails, it took a sampling approach, that’s really important to the training of having an executive that knows where business value can be unlocked with this kind of technology,” added Graviss.
He said that while the agency has been a fast mover on AI, the department is still looking to expand the use of generative AI tech from governance and cultural standpoints.
“Generative AI at scale in a secure environment for all of our users is probably the top technology priority, but it’s also the top priority from a governance and access and security standpoint, as well as a training culture change management standpoint,” said Graviss, who added “there a lot of basic use cases” in the agency that can benefit from generative AI use.
