How to (Actually) Recruit Talent for the AI Challenge
Editor’s Note: This article was submitted in response to the call for ideas issued by the co-chairs of the National Security Commission on Artificial Intelligence, Eric Schmidt and Robert Work. It addresses the second question (part b.), which asks how should the government go about hiring and managing qualified people, and the fifth question (part a.), which asks how the federal government can rely on the private sector to develop its AI.
In the global race to dominate artificial intelligence (AI) technologies, talent is everything. Yet the U.S. government has proven unwilling or unable to do what it takes to hire and hold America’s top talent.
Russia and China have long mixed their public and private sector resources when it comes to cyber technologies. Russian organized criminal groups often lend their best hackers to work part-time for the Russian state, while the People’s Liberation Army’s concept of a “modern people’s war” designates experts from major Chinese technology companies and academia as assets for improving national military power.
The United States, as a democracy, has not responded with equally effective ways to leverage vastly superior capabilities present in its own private sector. Instead, the Defense Department directly competes with American technology companies for a limited pool of cyber and AI talent — a competition it all too often loses. Once gone, tech professionals rarely return to government service and engineers with Silicon Valley experience almost never arrive to replace them.
To reverse these trends, the U.S. government should change the way it thinks about careers and embrace a lesson learned by leading tech companies — that software talent comes and goes and that’s a good thing. Skeptics will say that the Pentagon is different from any other institution and cannot succeed with a transient labor force that does not understand its way of doing things and doesn’t have high-level security clearances. Certainly, when it comes to battlefield applications, sensitive intelligence, or operational tactics, the Pentagon has plenty to hide. And its most sensitive software will have to remain off-limits to all but the most carefully vetted employees.
But most of the military’s software powers an ecosystem of backend operations that are no different than those used by private companies. The Defense Department needs to recognize that most of its software is not secret — just proprietary — and modernize its employment practices accordingly. Hospitals, for example, use dozens of specialized software programs that need to be interoperable as the patient moves from the emergency room to the operating theater to the intensive care unit to discharge. Logistics companies, financial firms, and telecom corporations all use custom software for their specific needs. Even so, the private sector manages to build highly customized software applications that enable their businesses to thrive using software talent that comes and goes.
The Pentagon Should Embrace a Non-Careerist Workforce
The Defense Department could take three specific steps to modernize its high-tech labor force, and do so in a way that keeps the nation safe.
First, to tap software engineers with the requisite experience in cutting-edge techniques for creating computer software, the military should consider moving more of its software jobs to the tech capitals where the best software engineers and AI talent already live and mostly want to stay. While top software engineers routinely move between cities with the headquarters of a major software company, like San Francisco or Seattle, or leading computer science universities, like Pittsburgh or Boston, they rarely move to Washington D.C. As a result, the software teams completing federal contracts have often been isolated from emerging best practices in the industry or have limited experience actually implementing technologies that the private sector adopted years ago. For example, while the best software companies implemented distributed computing platforms and cloud technologies a decade ago, most U.S. military departments have just begun this transition. This lack of hands-on experience with technologies routinely used in the private sector means most of the Pentagon is planning to transition its software applications to the Cloud without actually redesigning them to take advantage of the Cloud’s primary features. As a result, each application will continue to work essentially as it did before and most of the potential gains from this transition will be lost.
Winning software companies have long understood that top talent has a force multiplier effect. It’s worth paying more to acquire top talent. Unlike other industries, in which productivity differs only slightly from one employee to another, software firms believe their best engineers contribute ten or even a hundred times more to the success of the company than engineers of average skill. Flawed design choices made by poor coders can ripple through an entire team, lowering the productivity of every other programmer who must interface with their modules. Bad coders occasionally create bottomless pits of bugs and errors that can delay a product from shipping for weeks or months.
The establishment of Army Futures Command in Austin, Texas shows that the Pentagon can choose to set up shop in a city where software professionals want to live. Even better would be to locate acquisitions personnel — who write the specifications for military software and oversee its implementation — and operations researchers — who need to turn data into action just as the private sector must — in cities like Seattle that live and breathe software.
Second, the Defense Department should consider reforming its outdated expectation that tech specialists — military or civilian — will need to be careerists. Managers and leaders in the national security community often justify their aversion to pursuing potential hires from top-ranked computer science and data science programs by arguing that these employees will not stay for more than a few years — if they can be hired at all. Instead, national security hiring managers often prefer employees whom they believe will stay with the military and deeply enmesh themselves in the particulars of the military’s culture and values.
This misunderstands the modern employment landscape. Tech workers turn over rapidly, even ones that are paid as much as corporations like Facebook, Apple, and Amazon are offering. These companies know and accept that their typical employee has worked for their company for less than three years. Recruits who can pass a rigorous technical screening are welcomed — or welcomed back — by any tech giant regardless of their career stage. The best companies always hold positions open at every level to soak up talent whenever and however it appears.
Facebook brags that its newly hired engineers push real code to production within their first week. New hires in the national security world typically wait half a year or more to get a security clearance. Then they suffer more delays as informal internal barriers block their access to data, and rigid and bureaucratic processes prevent the deployment of tools and programming languages that data professionals in the private sector take for granted.
The Pentagon should consider redesigning its personnel policies to accommodate a much greater degree of speed and permeability in its cyber and AI workforce, regardless of what pay and benefits it offers. Its new priority could be to create a working environment that will attract the best employees, allow them to contribute rapidly to the best of their ability, and encourage them to stay only as long as they believe it is in their best interest to do so.
Finally, the federal government could learn from industry models that do not rely on full-time employees to develop world-class software. The open-source community created much of the software that powers the internet, including the Linux operating system and the Apache web server. While the Pentagon might not want to expose the source code behind its software applications to the entire world, it can learn from a community that has managed to create free products that are competitive with the best efforts of better-resourced, for-profit companies —even though many open-source teams pay no salaries, offer no benefits, and provide no career ladder.
The lesson is that most open-source, volunteer software engineers want to contribute to a worthwhile mission and improve the lives of their fellow human beings. Some have all the traditional employment benefits from their day jobs. Most feel empowered by the opportunity to code simply for the joy of tackling interesting and meaningful projects in their free time. Similarly, numerous servicemen, servicewomen, and Defense Department civilians would welcome the opportunity to learn a new skill, improve their day-to-day work environment, or simply explore a topic that interests them. A military version of the open-source model would connect volunteers from America’s best technology companies with the Defense Department’s most talented and innovative personnel. Ideally, it would also create a community of data and software professionals who could share new successes, enshrine emerging best practices, and iterate on exciting ideas.
Adapt or Risk Losing Out to China and Russia
Ultimately, whatever specific mechanism it uses, the Pentagon’s success in deploying the most innovative AI technology will depend on its ability to embrace a culture of creativity, innovation, and self-improvement. If it is to succeed, it will have to find ways to encourage cross-pollination with America’s best-in-class software ecosystem and draw from the country’s immense and impatient talent pool. If not, it could be doomed to lose the race to dominate AI.
James Ryseff is a technical analyst with the nonprofit, nonpartisan RAND Corporation. He previously worked as a software engineer for 13 years at Microsoft, SAP, and Google, among others. He holds a bachelors’ degree in Computer Science from the University of Illinois at Urbana-Champaign and a masters’ in Security Studies from Georgetown University.