10 June 2020

US Air Force seeks hardened artificial intelligence algorithms to assist aircraft piloting

The US Air Force (USAF) is in the early stages of developing artificial intelligence (AI) algorithms hardened against enemy attacks to command aircraft in partnership with human pilots.

Will Roper, USAF assistant secretary for acquisition, technology, and logistics (AT&L), said on 9 June that these algorithms, coined R2D2, would likely be first tested in the USAF’s Skyborg autonomous unmanned combat air vehicle prototype. Roper wants to better leverage convolutional neural networks in deep learning that tech companies use to analyse images posted on social media without human involvement.

An illustrative concept for the Skyborg UAV released by the US Air Force. The service envisions using the vessel to prove out hardened artificial intelligence that could assist pilots in combat.

Modern AI is fragile and vulnerable, Roper said, making it acceptable for entertainment-related functions. However, in combat, an adversary will attempt to thwart and confound that AI, spurring the USAF to engage in basic research to accelerate a hardened AI technology that will be suitable for warfare.

The USAF is not simply developing algorithms to replace pilots, but to make them more effective.

“We want to design a pilot that can deal with an adversary that is … trying to thwart that convolutional neural network in a type of algorithmic warfare that has never existed, but will on the future battlefield,” Roper said during a Mitchell Institute for Aerospace Studies event.

The US Air Force (USAF) is in the early stages of developing artificial intelligence (AI) algorithms...

Jane's Find unrivaled intelligence, consultancy and advertising solutions to the defence and national security sectors.