Senior Federal agency artificial intelligence (AI) and data experts talked about some of the hurdles that get in the way of moving AI pilots to more extensive production at ATARC’s Federal AI & Data Summit on Nov. 17.
Dr. Heather Martin, who is acting director, Office of Plans, Programs, and Strategy, Directorate for Data and Digital Innovation, at the National Geospatial-Intelligence Agency (NGA), said two impediments to moving AI products from the pilot phase to the production phase are cost and trust.
She said those two factors are among those that “I think really keep us from being able to move well from prototype, to integrating something into our baseline and sustaining it [and] getting it out there.” She added those factors have been persistent hurdles not only in her experience at NGA, but elsewhere as well.
Speaking at the same event Dr. Kenneth Clark, assistant director, Office of Information Governance and Privacy, at U.S. Immigration and Customs Enforcement, (ICE) stated that in order to implement AI to the best extent, organizations need to create system architectures that capture good data that AI tech can best utilize.
“I’m a strong proponent that as you implement AI, you really need to have good acquisition discipline and good program management. ,” stated Clark.
“And I think we know that AI requires so much data, and that data needs to be very close to where it’s being processed,” stated Dr. Clark.
Both panelists offered tips to other practitioners trying to implement AI technologies, which ranged from understanding that AI is “just another tool in your tool belt, it’s not the end-all be-all tool,” to advice that “spending time with your users” is should rank high up on the AI project implementation list.