Maintaining a competitive edge over global competition in artificial intelligence (AI) has risen swiftly over the past several years as a bipartisan national security priority. It’s enshrined in strategic guidance documents from the National Security Strategy to the National Defense Strategy.
That high-level strategic guidance is now transitioning to implementation. Two important steps were taken recently. First, the release of the Executive Order on Maintaining American Leadership in Artificial Intelligence and second, the release of the 2018 Department of Defense (DoD) Artificial Intelligence Strategy: Harnessing AI to Advance Our Security and Prosperity.
These documents made clear that delivering results for the nation on AI is the responsibility of individual federal departments and agencies doing their part. These documents also rightly advance what my CSIS colleagues and I presented in a previous report on AI and National Security termed an “AI ecosystem.”
The AI ecosystem includes a skilled and knowledgeable workforce; a digital infrastructure for capturing, handling and exploiting data; a technical foundation of trust, security and reliability; and an investment environment and strategic policy framework that provides the top-cover and fuel for growth.
This complex, supportive structure around the algorithms is often underdeveloped. Agencies have seen early successes by applying AI in data-rich fields where aspects of the ecosystem already exist.
By leveraging an “incubator” approach within a larger organization structure, these early success stories have applied AI to missions with rigorous analytic needs or repetitive tasking. For example, DoD’s Project Maven succeeded by leveraging an existing trove of data at a time when machine learning and image recognition capabilities were sufficient.
Careful attention to growing the AI ecosystem may require an investment of resources in time, money and support before returning dividends. Like a well-tended garden, the Al ecosystem must be cultivated; it will not spring up overnight.
The realities of federal planning and budgetary cycles mean that momentum will be built in the coming months as agencies looking to deploy AI move through the appropriate channels. Difficult decisions on priorities and financial appropriation will have to be made.
While defense research and development has been dropping as a share of the DoD budget for the previous eight years, the share of the budget devoted to AI has increased over last four or five, as well as increasing in terms of absolute dollars spent.
Federal departments and agencies will demonstrate they are serious about growing the ecosystem to deliver on the vision set forth by the president and secretary of Defense in three key ways.
Invest in the AI workforce. Often, the technical skills for success in AI vary drastically from the skills for success in a mission domain. Agencies must invest in their workforce by providing opportunities and funding for formal training, internships and fellowships.
Equally important, leadership and management must value talent once the workforce is in-house. Agencies must provide the pathways for the federal workforce to use their newfound skills in a structured environment conducive to software development.
This year, the Air Force is beginning to treat programming language competency as it does foreign language, providing incentive pay and duty assignments to skilled coders.
Invest in data management. Government must bring about a strategic shift in the way it thinks about data in addition to procuring the right computer and networking infrastructure. Simply capturing the growing volumes, varieties and sources of data is not enough.
Numerous examples of spoofing image-processing, machine-learning algorithms by slight distortions of the inputs demonstrate the importance of data quality. Turning information into an advantage means balancing security and access, ensuring the data is trusted and clean and managing the introduction of new data.
Pursue a strategic approach in software acquisition. AI is implemented through software. Current processes that treat software acquisition as interchangeable with hardware present significant barriers to innovation — if not stifle it altogether.
The development cycle time and iterative improvement nature of software products requires a faster, more agile acquisition system. While traditional hardware platforms may measure development in years, software progresses on a scale of weeks and months. Departments and agencies must learn to innovate at the speed of software.
While the executive order identifies the AI ingredients, it is up to departments and agencies to gather the ingredients and deliver success for their missions. The whole of government has common needs in finance, human resources, personnel management, logistics and maintenance and cyber defense.
More important than ever, each organization brings its own existing standards of ethical and safe use that will inform how AI is applied to solve problems. AI will flourish most in cases where the federal government takes a problem-centric, people-centric approach while maintaining standards of safe use.
Lindsey Sheppard is an associate fellow with the International Security Program at the Center for Strategic & International Studies, where she supports various projects in emerging technology, including AI and machine learning, and in security applications, ranging from strategic to tactical.