Advertisement

Energy planning first AI training module for the fall

The Artificial Intelligence and Technology Office is working with the department's 17 National Labs on additional modules.

The Department of Energy’s artificial intelligence office plans to address agencies’ struggles developing talent in computer science by releasing its first training module in the fall.

Future modules will be developed by the Artificial Intelligence and Technology Office in coordination with DOE‘s 17 National Laboratories and see experts explain the benefits AI can bring to their organizations.

Beyond training, DOE components have struggled to recruit, retain and develop a workforce structure for AI professionals, Cheryl Ingstad, director of AITO, said during an AI in Government virtual event Thursday.

“One of the things we’ve heard is that the scientists really want to work in the area of their specialty, but they don’t want to become AI experts to utilize these tools or to manage the data,” Ingstad said. “So how do we get data managers and how do we get AI experts to the operational levels here and the research levels to support our scientists that way?”

Advertisement

Rather than upskilling, AITO is planning a “conversion” of technically savvy employees with an interest in AI into specialists, she said.

DOE has drafted an AI strategy that applies to virtually every aspect of the department, but it remains under review, Ingstad said. She plans to chair an AI working group for sharing best practices and supplying dashboards that inform transformations across agencies.

A big part of DOE’s AI strategy are cross-cutting partnerships between internal offices and external agencies, academia, first responders, and companies, Ingstad said. That includes partnerships addressing adversarial AI, cyberattacks that cause machine learning models to misinterpret system inputs and behave the way the adversary wants them to act.

“We’re looking here at the datasets that we have and: Are these datasets secure? And what’s the provenance of these data?” Ingstad said. “How can we ensure that they’re clean, that there aren’t any built-in attacks to it?”

Autonomous decision-making will eventually be tied to datasets, but defending against adversarial AI extends beyond data to securing AI tools during every stage of development. Developers must think like attackers and look for lines of attack when creating AI tools, Ingstad said.

Advertisement

AITO adversarial AI partnerships are in the beginning stages, but DOE sees itself as entering a new “space race” with AI for the next several decades most likely, she said.

“It matters who wins,” Ingstad said. “We are in this race against others, who do not have the same values that we have.”

Latest Podcasts