Written by: Laura Dempsey
As nations across the globe increase and advance their space capabilities, the space domain is becoming increasingly congested – and contested.
Artificial Intelligence (AI) is poised to become a game-changer for United States Missile Command and the United States Space Force (USSF), both of which understand the need for the integration of autonomy and AI solutions.
Dr. Kerianne Hobbs, Run Time Assurance Research Lead at the Air Force Research Laboratory’s Autonomy Capability Team (ACT3), understands how AI – specifically, reinforcement learning – has recently demonstrated abilities that surpass humans in high dimensional decision spaces such as Go as well as in complex, real-time strategy games such as StarCraft, and sees practical applications in USSF and Air Force systems.
She also acknowledges that, “today, there is no certification criteria for placing AI in control of USAF or USSF systems. However, there is significant investment in developing AI control technology. There will soon be a set of capabilities that could give the USAF and USSF an edge, but no way of ensuring their safety.”
Dr. Hobbs’ experience with Wright Brothers Institute’s use of Sprints (weeklong, intensive analysis of a stated challenge or issue) spurred her to reach out to Bob Lee, WBI’s Open Innovation Project Manager, seeking to set up a Sprint to develop a prototype argument of safety for a hypothetical run time assured neural network controller.
“Currently our learning and artificial intelligence teams have a hard time collaborating,” she wrote. “We know of the environment that was set up for AFRL’s Materials and Manufacturing Directorate (with WBI’s guidance) and think we could set up an AI gym that would allow us to collaborate with a variety of uses both in academia and industry quickly.”
WBI has the connections in all senses of the word: The pandemic has made WBI experts in collaborating over the cloud, contacting other expert parties, and facilitating the Sprint, which took place in February.
“We’re bringing in experts to see if they can come up with legal arguments for the safety of a hypothetical runtime autonomous system – how do you prove that it’s going to be safe?” asked Mr. Lee. “We’re bringing in a unique team for this Sprint to see if we can build a prototype argument.”
The Sprint was successful .