Good Will Hunting: The Strategic Threat of Poor Talent Management
Pentagon leaders frequently say, “People are our greatest resource.” They are certainly the largest resource — the Defense Department employs 2.87 million people and spends nearly $140 billion per year on military personnel. Overall pay and benefits for both military and civilian employees of the department cost $273 billion, 42 percent of the Fiscal Year 2018 budget.
Yet the department is dealing with a Good Will Hunting scenario: It underutilizes its talent. The Pentagon has the luxury of being able to operate ineffectively so long as jobs are filled and tasks are completed. The demand to maintain traditional operational consistency creates an environment with no margin for risk — but which in fact creates the greatest strategic risk of all.
It is important to ensure that quotas for pilots, submariners, infantrymen, and other more traditional warfighting occupations are filled, in order to support the incessant mission directives each ship and aircraft receives. To improve job placement across the department, artificial intelligence and machine learning-enabled systems can identify traits that best correlate with operational success and guide service members with those traits to the billets where they can be most useful.
This approach, known as dynamic manning, would use intelligent systems to help automatically place service members into the jobs that best suit their skill sets while also optimizing senior leadership’s objectives. We propose that the Department of Defense take steps to turn the personnel job placement system into a market that seeks an equilibrium, allowing applicants and hiring managers to review job opportunities alongside applicants’ predicted performance score. This digitized, automated system would greatly improve Defense Department job placement, but it should be accompanied by increased focus on the types of skillsets and jobs the military should be valuing in the first place. For instance, the Defense Department undervalues critical opportunities outside historical navy roles, such as the Navy’s Cryptologic Warfare and Cyber Warfare Engineering communities. We believe history will show how important these nontraditional domains are if the United States is to outpace near-peer competitors. Talent management must guide — and open up more spots for — smart coders into the digital domains that America’s competitors have already embraced headlong. For instance, Chinese leadership has seized the talent management initiative with cyber security schools to train the best coders in a government-oriented curriculum, providing a pipeline to China’s cyber warfare community.
As Elsa Kania, an expert on military-technical competition between the United States and China has said, “the real ‘arms race’ in artificial intelligence (AI) is not military competition but the battle for talent.” The Pentagon’s existing human resources process places bodies in seats rather than skill sets in jobs. In an era of great power competition centered on emerging technologies and how militaries adapt to them, human capital inefficiency is a strategic risk.
How the Solution Could Work
A sailor logs into a secure website. The intuitive interface shows them all available jobs in the military. They can filter by location, organization, hard requirements (e.g. ranks or schools), or any other category. Next to each job title, they see a single number: a predicted performance score. This is a machine learning model output prediction based on past job holders’ performance, anticipating how someone with their experience and education would do in that role. Now the sailor can decide, based on their own desires and how desirable they appear to the system, how to rank their prospective jobs.
Next, the job owner, perhaps the commanding officer of that billet, logs into the same system. Their interface is a little different. They see each applicant ranked based on their predicted performance score. The owner does not see the applicant’s personal preferences; the system might even be set to scrub names for an added layer of anonymity. Now the job owner must decide whether to change these rankings, most likely for a reason the predicted performance score could not anticipate — a strategic change based on a shift in the mission or an attempt to infuse the command with more technical skills. At any rate, the job owner’s infusion of the human element offers a chance to shift the rankings from what the predicted performance score would suggest.
At this point, both job seekers and job owners have provided their inputs. The system executes a stable marriage optimization (exactly the same process that the National Residency Match Program uses to place applicants into medical residency and fellowship positions) to match the rankings for both parties. This optimization could also incorporate the preferences of service chiefs. One goal could be to minimize the distances and costs associated with personnel relocation. Another goal could involve preferential hiring for critical or new commands. Currently, commanding officers cannot account for such strategic-level considerations due to the sheer amount of information. Machine learning can assist in mitigating that problem. Dynamic manning systems would weigh and factor in both the individual needs of a service member and a prospective command, as well as the national objectives of the service chief. Consider a hypothetical service member, Navy Lt. W.T. Door. A model could create a predicted performance score for Lt. Door for each open job in the Navy. This score would be calculated by examining Lt. Door’s skills, the skill demands of each job he could possibly be assigned to, his past performance, and the skills and past performance of other individuals who previously held each potential job or had similar backgrounds.
An algorithmic model could be trained to develop this prediction with surprising accuracy. The necessary data is already inside the military, albeit in disparate databases and evaluations that natural language processing can assist in digitizing and categorizing. If a job is new or if there is little data available, the military can “bootstrap” data with similarity scores. Fascinatingly enough, this could be done using the same mathematical construct with which Netflix suggests movies based on what one has already watched. We leave to future articles discussions of how the military can more accurately and thoroughly capture records of the undocumented, underappreciated skills of its service members. But, as with all machine learning models, the better the data, the better the solution. Perhaps the very adoption of this system would provide incentive for better record-keeping.
In the dynamic manning process, jobs are then placed, and all parties are optimized with the least amount of loss. This cycle can be re-run instantaneously, allowing the next round of jobs and service members to be optimally allocated.
A Human Capital Problem
There is currently no way to meaningfully capture skills that a service member built outside the military. This lack of talent transparency in personnel records is a long-term risk to modernizing the armed forces.
Recognizing the problem, cutting-edge organizations within the Defense Department, such as the Air Force’s software unit, Kessel Run, the Defense Innovation Unit (DIU), and the Defense Digital Service, use products like LinkedIn and Google Hire, rather than typical military channels, to find talented military candidates. Other offices complement these outside practices with a strong current of by-name references and hand-selected resumes of trusted and well-educated associates. But these shorter-term solutions are an unsuitable model to roll out across the Defense Department. Curated resumes and self-reported skill sets allow these systems to run, yet a more sustainable set of principles may apply when considering the future of the most innovative units and conventional commands alike.
Despite its lethargy, the Defense Department has indeed made some important strides in improving its human capital. The 2018 National Defense Authorization Act supports merit-based promotions within the officer corps, allowing lateral entry by giving credit for civilian job experience, and other methods for enhancing the quality of officers while avoiding brain drain. The 2018 National Defense Strategy explicitly seeks to improve “workforce talent” by improving professional military education, civilian workforce expertise, and talent management. Additionally, the Chief of Naval Personnel Adm. Robert Burke has begun testifying before Congress to change laws concerning the required number of billets in each military occupation. He suggested the need to entice outside specialists in cyber, robotics, and artificial intelligence to join the military as officers. Burke said he intends to create a mechanism to allow these types of recruits to join at higher ranks as well as to stay in service longer than currently allowed. And yet, all of these talent acquisition efforts are massively hampered by the lack of accompanying talent management solutions.
The services have made initial efforts to create talent marketplaces, but bureaucratic sluggishness is impeding relatively easy innovation. A year ago, the Navy announced that 2018 would be the year of new age evaluations, to replace an antiquated performance evaluation system. It did not happen. The current efforts are going down the same path as the Defense Integrated Military Human Resources System, a case where the military tried to build a digital solution in a bureaucratic way, costing taxpayers nearly a billion dollars for a failed product. The Army’s pilot project, “Green Pages,” ran from 2010-2012 and sought to set up a web-based market place just as we have described in this article. While it yielded initial successful results, it moreover “reinforced the need for total system reform,” as Jim Perkins wrote in War on the Rocks. The lessons learned from the Green Pages system highlight the need for any future solution to have senior officer support and a strong technical team to oversee implementation.
The private sector is already placing a strong emphasis on learning-enabled talent management. Deloitte found that more than 40 percent of companies surveyed consider their application of artificial intelligence and machine learning critical to the success of their business operations, including hiring and talent management. Numerous startups already work on machine learning-based hiring and retention: Beamery manages the career of employees in a way similar to how a customer relationship management platform, such as SalesForce, manages customers. Pymetics utilizes a series of games designed by neuroscientists to understand the abilities and desires of current and potential employees and map those skills to jobs within the organization. These companies offer great promise for potential dual-use applications within the Defense Department, but must also consider the unique nature of the military workforce.
The solution we propose would make the job placement process significantly more transparent and objective. Service members will understand exactly how their placement occurs, while knowing their wishes were objectively heard and their talents systematically considered. This is a trust that does not exist in the current system. The automation would immediately lighten the load for the overworked personnel placement staff, often called “detailers.” These people are in charge of guiding the careers of hundreds of service members, yet do not have the superhuman bandwidth to absorb the input of niche skills, the hopes of service members, or the nuanced desires of job owners. Furthermore, they are hamstrung by ancient information technology systems.
The first step to implementing our proposal would be an information technology solution, creating the architecture to host the marketplace. Plenty of precedents for an online job marketplace lie in Google Hire, LinkedIn, and Glassdoor with respect to database management, business logic, and user interface. The next step would be integrating the ranking and optimization system, a process the aforementioned medical residency governing board could aid the department in piloting.
The machine learning-based predicted performance score could be explored in parallel to these efforts. This would require a coordination of data gathering and cleaning, only after which appropriate models could be explored. In a previous article in these pages, one of us detailed the necessary foundations the Defense Department needs to efficiently implement artificial intelligence applications. Dynamic manning is no exception — it must be enabled through labeled data, cloud environments, agile software development, integrated teams, and culture transformation.
The Defense Department has a gargantuan human resources mission: ensure that millions of service members and civilian counterparts are technically and tactically proficient enough to fight and win wars while at the same time not fully knowing the character of the next war. Training millions of people is not an easy task, nor is making sure every job is filled with someone capable of performing the duties required.
This article has explored a few of the most pressing issues with talent management and proposes a solution powered by artificial intelligence. Utilizing machine learning, an existing stable-marriage problem solution, and plain old human guidance from command leadership, the Defense Department of the near future can better utilize the people it has as well as attract even more talent. We encourage senior leadership to see this concept’s viability in the private sector and amongst America’s competitors and work to solve the “Good Will Hunting” problem before the nation finds itself mired.
Lt. j.g. Richard Kuzma is a Navy surface warfare officer passionate about how the Defense Department adapts to emerging technologies, particularly artificial intelligence. He is an alum of the Defense Innovation Unit and the Harvard Kennedy School, where he wrote a thesis on how the Defense Department should structurally change to implement artificial intelligence. He is an associate at Harvard’s Technology and Public Purpose Project. He is on Twitter, @rskuzma.
Lt. j.g. Zac Dannelly is a cryptologic warfare officer serving at Fort Meade. He has researched the effects of emerging technologies on organizations through degrees from the Judge Business School at the University of Cambridge, on the Gates Scholarship, as well as from the Management Science and Engineering Department at Stanford University.
Lt. j.g. Ian Shaw is a cyber warfare engineer and is on a gradient descent to optimize the application of his skills in support of the National Mission. He became passionate about the power of machine learning while studying computational math at Stanford University.
Lt. j.g. Drew Calcagno writes on artificial intelligence and machine learning policy for the Office of the Secretary of Defense as well as on sexual assault prevention and response for the Navy. He is an alum of the University of Oxford as a Rotary Scholar and the University of London as a Fulbright Scholar where he wrote theses on how the Department of Defense and the intelligence community could better coordinate counter-terrorism strategy, particularly on the African continent.
The views expressed here are the authors’ own and do not reflect those of the Department of Defense, the Navy, or the Defense Innovation Unit.
Image: Lance Cpl Leynard Kyle Plazo