For Immediate Release
Contact: Fiona Gettinger, Communications Associate
Advances in Artificial Intelligence, Robots and Swarms Demand Attention
New CNA Report Outlines Specific Actions Necessary to Move US Military Into Robotic Age
Arlington, VA — The rapid pace of the development of artificial intelligence (AI) technology breakthroughs in the private sector presents tremendous challenges for the U.S. Department of Defense (DOD) in the race to successfully integrate autonomous weapons into the nation’s military arsenal and to counter rising threats according to a new CNA report issued this week: "AI, Robots, and Swarms: Issues, Questions, and Recommended Studies" by CNA Principal Research Scientist Andrew Ilachinski.
The report cites recent advances in AI research by Google, Microsoft, Facebook and other technology companies and identifies the challenges that DOD will face as it embraces AI, robot and swarm technologies to "enhance and weaponize its fleet of unmanned systems with higher levels of autonomy."
"We believe this research takes the first critical step towards dealing with these challenges and trends," said Katherine McGrady, CNA President and CEO.
Ilachinski says that unlike in most other weapons technologies, there is a level playing field in artificial intelligence between the U.S. and its adversaries. China and Russia are actively investing in the field. "We're very much in a race whether we realize it or not," he adds."
While the arrival of artificially intelligent robots that are capable of acting independently on the battlefield will "offer a variety of obvious advantages to the warfighter," the report identifies conceptual and technical challenges in the design and development of these autonomous systems:
- Researchers have struggled for decades to solve key problems that are critical to utilizing AI systems in warfare: They must be able to sense, perceive, detect, identify, classify, plan for, decide on, and respond to a diverse set of threats.
- There is a "scaling problem." Systems must be able to assimilate, respond and adapt to dynamic conditions that were not considered during their design.
- Systems must have a built-in capacity to learn and do so without human supervision.
- The effectiveness of the systems will depend on the interplay between the human operator and the machine in a given environment, in real time, in concert with the human’s adaptation to dynamic contexts.
- There is a challenge translating human goals into computer instructions as well as depicting the machine’s decisions to the human operator in a way that is understandable.
- As autonomous systems increase in complexity, we can expect a decrease in human ability to predict and control the systems.
"There is an aspect of this which is potentially very frightening because we’re developing technologies that we frankly are incapable of understanding," Ilachinski says.
Ilachinski also says that DOD faces a problem that boils down to simple math: Its current acquisition process takes 91 months (7.6 years) on average, roughly the same amount of time it took for the commercial sector to achieve breakthroughs in AI that went from experimental prototypes to the highly publicized defeat of the 18-time reigning world champion of the board game Go by an AI computer.
"There's something there that needs to be not just fixed, but overhauled, and it’s an enormous undertaking," says Ilachinski. "It behooves the DOD to think very seriously about how it procures weapon systems, how it develops and designs of them in the context of AI."
The report identifies four key technical gaps that must be addressed to overcome the challenges as DOD prepares to cross-over into the Robotic Age:
Gap 1: There is a fundamental mismatch between the accelerating pace of technology and research in the commercial and academic research communities and the timescale of DOD’s existing acquisition process.
Gap 2: There is an underappreciation of the unpredictable nature of autonomous systems, particularly when operating in a dynamic environment such as a battlefield.
Gap 3: Currently there is no universal conceptual framework for autonomy that can be used to anchor theoretical discussions and serve as a frame-of-reference for understanding how theory, design, implementation, testing and operations are all interrelated.
Gap 4: DOD’s current acquisition process does not allow for timely introduction of "mission ready" AI/autonomy and there is a disconnect between system design and development of concepts of operations. (CONOPS)
CNA is a nonprofit research and analysis organization dedicated to developing actionable solutions to complex problems of national importance. With nearly 700 scientists, analysts and professional staff, CNA takes a real-world approach to gathering data. Its one-of-a-kind field program places analysts on carriers and military bases, in squad rooms and classrooms, and working side-by-side with a wide array of government decision-makers around the world. In addition to defense-related matters for the U.S. Department of the Navy, CNA’s research portfolio includes criminal justice, homeland security, energy security, water resources, enterprise systems and data analysis, and education.
Note to writers and editors: CNA is not an acronym and is correctly referenced as "CNA, a research organization in Arlington, VA."