Artificial Intelligence Is Not the Right Tool for Everything, Top Army Scientist Says

Discussions about the future of artificial intelligence with some technology novices—and experts—are often rife with references to "The Matrix" and "Terminator," but such romanticizing can stall agencies from adopting the technology, according to a top government scientist.

“Do not be seduced into thinking [artificial intelligence] is different than any other IT application—it is not,” said Alexander Kott, chief scientist at the Army Research Lab. “AI is not a magic dust.”

Kott, alongside technologists from government and industry, worked to demystify artificial intelligence and outline strategies for expanding its use in government at a workshop hosted Tuesday by FCW. Panelists stressed the need for agencies to take a deliberate, baby steps approach when developing artificial intelligence tools and make the technology as accessible as possible when rolling it out.

Before diving headfirst into a major artificial intelligence project, it’s necessary feds step back and take stock of the issue they’re actually hoping to solve, panelists said. Despite the current hype, artificial intelligence isn’t always the right fix for every problem, they said, and there can often be a cheaper, less-technologically advanced solution that works even better.

And if artificial intelligence seems like a good fit, make sure to not bite off more than you can chew, they said.

“You need to tightly bound your first foray into this area,” said Patrick Arvidson, special assistant to the deputy national manager for national security systems at the Pentagon. Bringing AI to government requires getting top leadership on board, he said, and when big, costly projects fail early on, it can hamper support for future initiatives.

And Kott said those straightforward applications can make a huge positive impact, even if they don’t seem that innovative.

“There is a very thin, almost unnoticeable line between AI and non-AI,” he said, and just as the GPS apps that use AI to devise routes once seemed unthinkable, many of the tools people today take for granted could be the most effective solution to a problem.

Similarly, building AI into systems employees are familiar with can help speed up adoption, said Stephen Dennis, director of the data analytics engine at the Homeland Security Department’s Science and Technology Directorate. People are often resistant to changing their workplace habits, so making the transition as easy as possible could help overcome initial reluctance, he said.

But integrating artificial intelligence into government processes will ultimately require more tech-savvy federal employees, and Dennis told Nextgov he’s “very concerned” about the government’s lack of fresh talent.

“I’m not sure we can take the current workforce where it needs to go,” he said. Not every federal employee needs to understand all the details of artificial intelligence, but somebody must be responsible for maintaining it, and if feds can’t keep up with new technology, “then you’re going to have problems,” he said.