The U.S. military’s approach to AI is equal parts offense and defense, acknowledging that primary adversary China could also weaponize the technology as a form of asymmetrical warfare in which U.S. military superiority is blunted by dual-used AI technologies.
Hence, the service said it would seek to organize and leverage its vast trove of operational data to develop new algorithms for future sensor as well as command and control systems.
“Depending on the strategic choices we make now, our ability to operate around the globe may be blunted or bolstered by the adoption of—or hardening against—artificial intelligence,” notes the U.S. Air Force’s AI strategy released in September.
The Air Force plan is part of a broader Defense Department effort to ramp up military AI capabilities under a Joint Artificial Intelligence Center created last year to coordinate AI R&D. The DoD strategy relies heavily on partnering with technology companies, a stance that has generated push-back at leading AI developers like Google.
Among the priorities listed in the Air Force AI strategy is recognizing the primacy of data “as a strategic asset.” Hence, the blueprint calls for “consistently generating training-quality data for algorithmic development” that “provide the shortest path between development and operational events.”
Among the service’s objective under the data initiative are creating searchable databases containing “massive training data,” the strategy notes. That training data would in turn be used to develop diverse sets of algorithms. Open-source algorithms also would be leveraged.
Beginning in 2017, DoD launched its Algorithmic Warfare Cross-Functional Team, also known as Project Maven, to accelerate algorithm development for scanning hours of reconnaissance video.
“Technology has always underpinned the changing character of war,” the Air Force blueprint concludes. “For those of us in the military sphere, AI is akin to the development of stealth aircraft and precision guided munitions.”
The plan also acknowledges the ethical dilemmas poised by AI and growing tech industry opposition to developing military applications for AI. The Pentagon’s conundrum in formulating its AI strategy is the reality that it must work closely with industry and academia to keep pace with China.
Hence, the Air Force strategy reflects DoD-wide efforts to find a middle ground, pledging to “engage in dialogue on the ethical, moral, and legal implications of employing AI in military operations in concert with the Joint AI Center.”
Meanwhile, the service is also integrating its AI efforts within its own research labs as well as department-wide efforts as DoD hustles to counter China’s aggressive push into machine learning and other dual-use AI applications.
For example, DoD’s top research agency is pouring billions of dollars into cutting-edge research efforts such as “common sense” AI.
The Defense Advanced Research Projects Agency also is using the aerial dogfight scenario to leverage AI as a way of improving the interface between pilot and extremely fast-moving machines. In May, DARPA rolled out another AI effort dubbed ACE, as in Air Combat Evolution, to develop trusted AI to assist fighter pilots in making split-second decisions while flying at supersonic speeds.
Program officials noted that the highly “nonlinear” characteristics of aerial combat make it a “good test case for advanced tactical automation.”
The DoD recently deployed its largest AI supercomputing system, TX-GAIA (Green AI Accelerator), at the MIT Lincoln Laboratory Supercomputing Center in Holyoke, Mass. The 4.7 petaflops (Linpack) system will serve the HPC and AI needs of the DoD and will support the recently announced MIT–Air Force AI Accelerator with 100 mixed-precision tensor petaflops.