Google assisting the Pentagon in developing AI for its drones
TensorFlow APIs are being used for object detection
Google is working with the US Department of Defense to develop AI algorithms to identify objects in videos taken by drones.
So far, it’s unclear exactly what technology is being used and how involved Google are. But it’s all part of Project Maven, a programme launched by the Pentagon last year in April.
The goal is to start an Algorithmic Warfare Cross-Functional Team to advance computer vision techniques using machine learning to autonomously extract useful information from footage. Hopefully the systems will be smarter than human operators, who can't seem to tell the difference between an Afghan wedding and a terrorist encampment.
Google employees started discussing the project on an internal mailing list last week and the chats leaked to Gizmodo. Some Chocolate Factory staffers were, apparently, “outraged” and others used it to bring up ethical dilemmas surrounding the use of machine learning.
In a statement, a spokesperson told The Register that the search giant were providing “open source TensorFlow APIs that can assist in object recognition on unclassified data”. TensorFlow is a framework used by Google for machine learning and neural networks. Its APIs make it simple to use image recognition models with its cloud service.
Google did not answer questions on the exact API that was being used. And it’s unknown if the model is public or if developers are working to tailor algorithms for the DoD’s drone videos.
The fear of military use of AI is rising as its capabilities advance. The idea of autonomous weapon systems and surveillance systems has been widely discussed in AI safety and ethics reports, especially since neural networks have been described as “black boxes”.
It’s very difficult to untangle how a model utilizes its training data to arrive at an answer, making it risky to trust in the machine’s decisions, whether its for diagnosing diseases or warfare.
The Google spokesperson added: “The technology flags images for human review, and is for non-offensive uses only.” And admitted that “military use of machine learning naturally raises valid concerns. We’re actively discussing this important topic internally and with others as we continue to develop policies and safeguards around the development and use of our machine learning technologies.”
The DOD has made public its intentions to ramp up research efforts in AI and machine learning and the need to partner up with industry.
Bob Work, former Deputy Secretary of Defense, previously said that “As numerous studies have made clear, the department of defense must integrate artificial intelligence and machine learning more effectively across operations to maintain advantages over increasingly capable adversaries and competitors.
“There is no ‘black box’ that delivers the AI system the government needs, at least not now. Key elements have to be put together … and the only way to do that is with commercial partners alongside us." ®