The US military has not been shy about its interest in using AI to bolster its power, and now it wants to use it to train human soldiers.
The use of artificial intelligence (AI) for military purposes is somewhat controversial, with Google recently getting into some hot water over its work with the US military on Project Maven, which is using AI to analyse drone footage.
While a number of Google employees signed a letter to CEO Sundar Pichai calling for an end to the collaboration, there is nothing stopping the US Army Research Laboratory from conducting its own research.
Now, the military research group has revealed it is using AI not to produce a new weapon, but as a means of getting human soldiers to learn quicker, making them more lethal on the battlefield.
When implemented, it could help soldiers decipher hints of information and respond faster; for example, recognising threats such as a vehicle-borne improvised explosive device, or potential danger zones from aerial warzone images.
Using low-cost, lightweight hardware, a research team was able to implement something known as collaborative filtering, a machine learning (ML) technique capable of achieving a 13.3-times speed-up of training compared to a state-of-the-art optimised multicore system and 12.7-times speed-up for optimised GPU systems.
The system also consumes considerably less power than other devices at just 13.8W, rather than as much as 235W in GPU platforms. This makes it a potentially useful component of adaptive, lightweight, tactical computing systems.
AI on the battlefield
Its lead developer, Dr Rajgopal Kannan, said the technique could eventually become part of a suite of tools on board one of the US military’s next-generation combat vehicles.
More broadly, the research he and his other colleagues are working on is part of the military’s recent focus on AI and ML, with the aim of staying one step ahead of its global rivals, such as Russia and China.
This includes working on developing several techniques to speed up AI and ML algorithms through new designs on state-of-the-art, inexpensive hardware.
A recent US military white paper authored by the chief of the network science division at the Army Research Laboratory, Dr Alexander Kott, outlined a vision for how AI and the internet of things will be seen in future battlefields.
“In addition to physical intelligent things, the battlefield – or at least the cyber domain of the battlefield – will be populated with disembodied, cyber robots,” it said. “These will reside within various computers and networks, and will move and act in the cyberspace.”