war on the rocks

Who is the Admiral Rickover of Naval Artificial Intelligence?

September 18, 2018

Adm. Hyman Rickover, “the father of the nuclear Navy,” once stated, “All new ideas begin in a non-conforming mind that questions some tenet of the conventional wisdom … Good ideas are not adopted automatically. They must be driven into practice with courageous impatience.”

Rickover and his Naval Nuclear Propulsion Program were able to drive nuclear power into practice in the Navy in fewer than eight years. Between 1947 and 1955, Rickover successfully advocated for a nuclear-powered submarine as head of a new Nuclear Power Branch in the Bureau of Ships. Nuclear power is an “enabling” technology with broad, general applications. Over the years, it has been applied to many types of systems, including tanks, aircraft, space probes, and doomsday torpedoes. In this way, atomic energy is similar to artificial intelligence (AI); both are “more akin to the internal combustion engine or electricity than a weapon.”

But unlike the Naval Nuclear Propulsion Program (also known as the Naval Reactors program) today there is no Navy organization accountable for overseeing a practical application of any AI-enabled combat system that is ready be pushed to the fleet in the near future. Useful military AI is proven in concept. What if the Navy treated operationalizing AI enabled combat systems the way it once treated operationalizing nuclear power?

Scientists in academia and industry have already set the stage for AI-enabled combat systems. Now, that work must be brought to operational fruition and made battle-worthy. That’s the Navy’s job. Technology in and of itself is of limited utility if the organization lacks people who can use it, effective strategies to use it, and training to be good at using it. The Navy should create and fund a major artificial intelligence organization with a mandate to develop, build, and sustain a high-priority AI-enabled weapons project.

Nuclear Power: A Lesson from the Navy’s History

Submarines equipped with nuclear power plants are formidable weapons. They don’t need to re-fuel, allowing them to transit underwater at high speed almost indefinitely. The U.S. Navy first promulgated a formal requirement for a nuclear-powered submarine in 1947 despite having no nuclear-trained submariners, no nuclear industrial base, and no nuclear shipyards. The first atomic submarine, the USS Nautilus, steamed underway on nuclear power fewer than eight years later.

Having operated safely for over 6,200 reactor-years without a single radiological incident, American nuclear vessels barely register on the public’s radar. A naval AI program could benefit from the lessons of the program that produced that record. As Rickover said:

The development of nuclear propulsion plants is a good example of how one goes about getting a job done. It is a good subject to study for methods … It has involved the establishment of procedures and ways of doing government business for which there was no precedent, and which I believe will be necessary in the future for similar large projects.

A naval AI organization equivalent to Rickover’s Naval Reactors would go a long way towards providing the oversight, initiative, and effort necessary to get AI to the fleet.

Creating a Major Navy AI Program

Several recent War on the Rocks articles (including a recent commentary by one of us) have discussed the military tasks AI is suited for and the infrastructure that will be needed for it to truly be militarily useful. These articles recognize that AI is already useful for automating certain military processes, but that much more work needs to be done regarding how to operationalize AI across a military service. Defense Department leadership recognizes the urgency to operationalize on a grand scale, but the Navy’s efforts to accomplish that end have been scientific in nature, small in scale, and lacking in urgency.

One reason for this is that Navy AI efforts are competing for money with legacy programs — and losing. This near-term focus is delaying the next generation of AI-enabled weapon systems. In the 1950s, the Navy’s nuclear power program also faced a difficult competition for limited government resources. Although attention and resources were focused on the destructive power of the atomic bomb, not nuclear power, the Naval Reactors program still managed to get a reactor built.

The “nuclear Navy” would not have achieved its extraordinary success without Adm. Rickover. Today, there is no individual officer answerable to the Chief of Naval Operations who is responsible for operationalizing Naval AI. Like the Naval Reactor program, the Navy’s AI-enabled weapon system program would need a leader with tenacity, work ethic, and vigor. The Chief of Naval Operations should designate this leader and give their program authority over the structure, training programs, and personnel admitted. This person should have an extensive technical background in AI as well as an operational military background. They should be selected for a willingness to aggressively push the interests of their program, as Rickover did so effectively. Perhaps the Chief of Naval Operations will find that combination of qualities only in more junior personnel — Rickover was a captain when he was put in charge of the nuclear power program in 1947. He wasn’t promoted to the rank of rear admiral until six years later, in 1953.

The authority of this new AI organization needs to cut across existing Navy communities. While authorities within the Navy should come directly from the Chief of Naval Operations, the organization will also need authorities from outside the service. Rickover was dual-hatted in his authority to build a naval reactor for the Navy and the Atomic Energy Commission. He assigned people to the Navy Nuclear Propulsion Program using his Energy Department authorities, not his Navy authorities. The Defense Department’s recently created Joint Artificial Intelligence Center  may be able to provide a future naval AI organization with additional authorities analogous to the Atomic Energy Commission in Rickover’s time. Because the center is still being formed, the Navy has an opportunity to shape it. A memorandum of understanding between a major Navy AI program and the Joint Artificial Intelligence Center, signed by both, would go a long way towards providing a basic charter, along with incorporating Navy offices within the center.

A Need for Training and Oversight

AI appears frightening to many. Such fears are rational — it has potential to do enormous damage. Similar fears surround nuclear power, especially after Three Mile Island, Chernobyl, and Fukushima. Military AI will be complex relative to consumer-oriented AI because AI combat systems will be more complex, complex ethical issues will be involved, and the consequences of failure will be higher. Thoughtful oversight of AI-enabled weapons should be provided by accountable military personnel in positions of public trust, not by the companies hired to build AI enabled combat systems. Along with the establishment of a governing AI organization, to safely operate future AI-enabled combat systems, the Navy needs personnel capable of operating such systems.

Sailors don’t need to be able to design AI to understand and operate it. However, because weaponized AI has the potential to cause so much damage, military AI operators must have a more detailed understanding of its capabilities and limitations than their commercial counterparts. Military programs will need to educate sailors on basic theories of AI and train them to operate their AI-equipped systems. In the same way, nuclear-trained sailors receive a general understanding of nuclear power through Nuclear Power School followed by more specific training on the operation of specialized nuclear systems.

The application of AI must become part of the Navy’s warfare and engineering communities. Maintaining and operating AI-enabled weapon systems will require training, operating manuals, and maintenance procedures. The Navy’s civilian engineers and engineering officers will be responsible for designing and building combinations of AI hardware and software to include communication paths, databases, training algorithms, and computing power. Existing military organizations can help facilitate this training. Warfighting development centers, the Defense Innovation Unit, the Joint Artificial Intelligence Center, and the Naval Postgraduate School can all contribute. Furthermore, there is no reason this system wouldn’t also be available for other services and government organizations looking to implement AI.

The training program should be rigorous and primarily focused on theoretical and practical aspects of AI, with an emphasis on rejecting blind trust of algorithms. Due to the complex and dynamic nature of war, AI-enabled weapon systems cannot be inflexibly bound by rigid procedures. For example, while shutting down an AI system in a confused environment may be the safest way to reduce both fratricide and civilian casualties, during a fight it may not be the safest path overall. Operators must have sufficient knowledge to intelligently assess alternate procedures if an emergency occurs. A rigorous training program won’t turn operators into AI experts, but it will give them an adequate understanding of the basics so that they can operate future systems safely and intelligently.

As with the Navy’s nuclear program, the AI organization we envision will need exceptional people with technical education. Keeping AI technical talent in the military, especially when there is enormous commercial demand for such talent, will not be easy. There are only two ways the Navy can compete: train its own and offer the most interesting problems to solve. If the Navy’s AI program gives those with technical expertise the ability to work on complicated and rewarding projects, the service will be able to recruit and retain these talented experts. Starting now will help the Navy find young operators and train them, even during this early stage. The Navy already has many special financial incentive programs to retain operators with special skill sets: For instance, “nuke pay” is the special incentive designed to help retain personnel in the Naval Nuclear Propulsion Program. AI-trained operators should receive similar financial incentives in the form of special pay and bonuses. They should also have clear paths to promotion within existing Navy communities and should be promoted at a greater rate than operators with less relevant skill sets. If the Navy doesn’t adequately incentivize its AI talent, that talent will leave for industry.

At the same time, the strength of the connections between a Navy AI organization and industry will be especially important. Private-sector companies, not government organizations, will remain the leaders in AI innovation for the foreseeable future. It is worthwhile to demonstrate to the people in those companies that the military has a plan to use AI ethically and responsibly. More Navy partnerships with technology firms would create more overlap between the service and Silicon Valley. However, the value of military tours with industry can be overstated since industry often doesn’t have many of the bureaucratic impediments that are actually limiting the application of AI inside the Pentagon.

Conclusion

The Navy will require a major AI program and tremendous engineering efforts to successfully operationalize its AI combat systems. However, the organization has successfully operationalized other complicated and potentially dangerous enabling technologies in the past, including nuclear power. Battle-worthy AI is not a pipe dream; it’s a serious endeavor that is within reach. Adm. Rickover’s program shows us the way forward. His Naval Nuclear Propulsion Program overcame organizational and bureaucratic challenges inside and outside the Navy to get the job done.

Naval officers are responsible for ensuring operationalized AI meets its military potential; they can succeed through strong leadership and training programs designed to produce an organized cadre of technically trained AI professionals. But to make all of this happen, the service needs to designate a tenacious, empowered individual — one who can say, “I am accountable for operationalizing Naval AI” — to push the Navy into the future.

 

Lieutenant Commander Connor McLemore is an E-2C naval flight officer with numerous operational deployments during 18 years of service in the U.S. Navy. He is a graduate of the United States Navy Fighter Weapons School (Topgun) and an operations analyst with Master’s degrees from the Naval Postgraduate School in Monterey, California and the Naval War College in Newport, Rhode Island. He is currently with the Office of the Chief of Naval Operations Assessment Division (OPNAV N81) in Washington D.C.

Lieutenant Eric Jimenez is a nuclear submarine officer with 7 years of service in the U.S. Navy. He is currently serving as a campaign analyst within the Office of the Chief of Naval Operations Assessment Division (OPNAV N81) in Washington D.C., where he interprets scientific studies and conducts campaign analysis to guide strategic investments. He graduated with a Bachelor’s degree in electrical engineering from Oklahoma State University. The views expressed here are theirs alone and do not reflect those of the U.S. Navy.

Image: Nuclear Regulatory Commission/Flickr