How's Uncle Sam getting on with Biden's AI exec order? Pretty good, we're told

Former Pentagon deputy CIO Rob Carey tells us guardrails should steer Feds away from bad ML

Interview President Biden's October executive order encouraging the safe use of AI included a ton of requirements for federal government agencies that are developing and deploying machine learning technologies.

So far the implementation of that order within the government has been going pretty well, a former Pentagon leader has told The Register.

Rob Carey, former principal deputy CIO at the US Department of Defense and now president of government solutions at Cloudera, told us in an interview you can watch below that federal agencies have mostly begun to get their houses in order per the EO's requirements. Those obligations are supposed to help minimize risk, promote standards and testing, and bake security protections into AI software.

Youtube Video

"Every agency now has a chief data officer, every agency now has a data plan for the most part," Carey said, in the context of the order highlighting of the need to safeguard personal information when training and running models.

You can find a summary of Biden's exec order here from the White House and here from EY, plus a handy tracker detailing what's expected and what's been done here.

So, what about bad implementations of AI, such as faulty systems that deny medical coverage or misidentify police suspects? What's stopping the Feds from deploying that kind of glitchy tech? Keep in mind the executive order was more like a set of guidelines and guardrails, Carey said. 

"I think [the EO] provided guardrails for things that are already going on," Carey told us. "If [agencies] happen to be operating outside of the envelope that is portrayed, they understand they have actions to take." 

AI is coming, as far as the White House is concerned, and so now it's all about ensuring safe and trustworthy systems are deployed, which we discuss above. ®

More about


Send us news

Other stories you might like