5 July 2021

WINNING THE FUTURE WITH ARTIFICIAL INTELLIGENCE

Dr. Mark T. Esper

Advances in artificial intelligence will change the character of warfare for generations to come, affording decisive and enduring advantages to whoever harnesses it first. Recently, the National Security Commission on Artificial Intelligence’s Final Report released on March 1, 2021 emphasized the importance of AI to our nation’s security when it set an aggressive target for America’s armed forces: become “AI-ready by 2025” or risk ceding our overmatch—a target I believe is crucial to meet. As Congress considers the president’s budget request, it should keep this goal in mind.

In today’s era of great power competition, the United States must win this future against China and Russia. Doing so requires a 3–5 percent real increase in annual defense spending, coupled with the retirement of legacy systems and other reforms that will enable efficient investment into the digital workforce, systems, and infrastructure needed to ensure our AI readiness.

As secretary of defense, I felt so strongly about AI’s importance that we sped up the fielding of AI capabilities at scale to meet warfighter needs through the Joint Artificial Intelligence Center. We also published DoD’s first-ever set of AI ethical principles to ensure the United States is the global leader in the responsible development and use of AI.

But to win the AI future, the Pentagon must also successfully advance and integrate four main ingredients: standardized and shared data; relevant AI algorithms and models; strong cybersecurity to protect the data and integrity of the models; and a robust transport mechanism (e.g., 5G) to push AI out to the warfighter.

In 2018 as secretary of the Army, I initiated a comprehensive review of Army capabilities and programs called Night Court. The Army’s leadership looked at over five hundred programs and asked, Does this program support the Army’s modernization priorities? If not, then we canceled or reduced it, allowing us to realign over $30 billion to higher-priority programs. A second round of reviews in 2019 harvested another $13 billion.

When I became secretary of defense, I launched a similar review for the entire department. We made good progress despite a late start, but more can be done. To be “AI-ready by 2025,” the Pentagon must continue reforming while gaining congressional support to trim low-priority, underperforming, and legacy systems ill-suited for tomorrow’s conflicts. Savings can then be reinvested in game-changing technologies, such as digital infrastructure and the innovations needed to win AI-enabled warfare.

To help make these difficult decisions, the department should also continue taking the following steps to drive more rigorous analysis and evaluation of its programs and systems: first, build the capacity to support data-driven decision-making; second, identify the AI-dependent capabilities the US military needs to defeat a near-peer competitor; and third, establish more substantive criteria for AI readiness.

Data-Driven Decision-Making

AI will vastly accelerate the pace and scope of conflict, and will place an even greater premium on smart, timely, and accurate decisions in war. Data-driven decision-making enabled by AI can also help the Pentagon make better investment choices during peacetime. To achieve these outcomes, leaders must have broad, on-demand access to authoritative data.

During my tenure as secretary of defense, we established the first official chief data officer under the CIO and issued the first Department of Defense Data Strategy to guide improvements on the availability and reliability of department information. We also took additional steps to support management operations by developing platforms for senior leaders to interact with live data. To this end, an important first step was the launch of the DoD comptroller’s ADVANA data platform, integrating hundreds of department datasets to support the Pentagon audit and other core business processes.

Much work remains, however. Data is the fuel that AI needs to be relevant. On its own, algorithms are worthless without the right inputs. The hard work that lies ahead is for the Pentagon to rationalize and share its data across the services and its many agencies, with the most important step being to cultivate a culture of data sharing.

DoD had the largest research and development budget in its history in 2020, and recently unveiled an even larger one that we programmed last year. Our investment focus was on the top eleven modernization initiatives—beginning with AI—that would ensure our continued overmatch well into the future. To effectively optimize our portfolio going forward, DoD must assess and weigh its core investments in AI, quantum computing, directed energy, 5G networks, and other technologies based on the weapons and systems they will bring to bear, as well as new capabilities they could create or enable. For example, a review of DoD’s modernization portfolio today should consider foundational investments in AI and emerging technologies necessary to support joint capabilities, such as Joint All-Domain Command and Control (JADC2) and long-range precision fires, while simultaneously determining how and where to place bets to advance next-generation capabilities. To accomplish this, the Pentagon must first develop curated datasets of its technology investments, labeled with the operational and technological attributes of each, along with the digital platforms and applications that will allow automated tracking and broad trade-space analysis.

We should also reimagine DoD’s analysis of alternatives (AoA) process, which requires comparing systems under development to alternative solutions before procuring them, using evaluation criteria unique to each program. With better data and AI-enabled analysis, the AoA process could more quickly and effectively generate estimated returns on investment for both current and future systems, allowing apples-to-apples comparisons that inform department-wide investment and divestment decisions. This will also help DoD communicate modernization priorities more effectively with Congress, allowing for better-informed oversight of budget decisions and their necessary trade-offs.

Capabilities to Defeat a Near-Peer Competitor

My top priority as secretary of defense was implementing the National Defense Strategy (NDS), with a clear focus on China. We had a very good understanding of the capabilities needed to ensure the comparative advantage of US forces in a near-peer conflict. These requirements were based on a clear set of facts and assumptions used to guide our thinking. A few stood out.

To begin, China should continue to be the pacing threat as it works toward its stated goal of becoming the world leader in AI by 2030, part of its plan to complete its military modernization by 2035. The Chinese Communist Party is operationalizing this plan by driving massive investments into applied AI research and leveraging its strategy of military-civil fusion to harness private sector innovation for military advantage.

Second, US forces will need to be able to fight and win across all domains, especially the comparatively new ones of outer space and cyberspace. Operating as a joint force across these domains will require an unprecedented level of connectivity and interoperability. The concept of JADC2, for example, will provide the connectivity required to coordinate all-domain operations against near-peer competitors with capabilities to disrupt or deny US battle networks.

Finally, AI will change warfare across the ground, sea, air, space, and cyberspace domains. As such, the department must have existing platforms and systems that are ready to operate within these radically changed environments, but must also position itself to quickly adopt and operationalize emerging and disruptive technologies.

One of my “Top Ten” NDS implementation objectives was the development of a modern Joint Warfighting Concept (JWC) that enabled our transition to all-domain operations. DoD should continue to ruthlessly validate the JWC through rigorous experimentation, wargaming, and exercises while leveraging modern disciplines and tradecraft such as digital engineering and digital twins that enable virtual representations of warfighting systems and platforms. This will require foundational investments in the technical infrastructure to develop and scale modeling and simulation sandboxes where cleared warfighters, researchers, and industry partners can iterate and validate operational concepts against countless adversarial scenarios and capabilities.

DARPA provided a glimpse of what is possible during last year’s AlphaDogfight Trials when a simulated F-16 operated by an AI agent defeated not only other AI agents in a simulated digital air-to-air combat, but also a human F-16 pilot. Experimentation and wargames like these will not only provide great insight into operational concepts, but also enhance training for warfighters by tailoring learning to maximize performance.

Defining and Achieving AI-Readiness

Finally, the department should set clear schedule and performance objectives that drive measurable improvements in the digital infrastructure and workforce necessary to achieve AI-readiness. These objectives will help define investments in a digitally enabled force not by five-year milestones or set quantities of air, ground, or sea platforms, but rather in terms of the ability of our military forces to adopt emerging and disruptive technologies at a pace and scale that creates uncertainty in the minds of our adversaries. Additionally, these objectives will help us clarify and refine the key technical and operational attributes required by AI-enabled technologies.

DoD must define and invest in the underlying digital ecosystem that enables digital innovation. To use an analogy from home renovation shows on TV, the process of modernizing a house can’t start with quartz countertops, wide-plank hardwood floors, and recessed lighting. It begins with the heavy-lift investments to upgrade the plumbing, electrical wiring, and insulation behind the walls. Moreover, an open-concept home that replaces individual rooms requires an expensive overhead beam to support the load. In a similar fashion, an AI-ready force requires a commitment to resource the data, trained AI models, cloud storage and computing, software development, and secure networking and processing—the foundational digital infrastructure to support AI-enabled applications and technology.

While these initiatives are a major undertaking for the department, the Pentagon has a head start with the recent blueprint for a digital ecosystem published by the National Security Commission on Artificial Intelligence. The blueprint envisions a technical foundation that: 1) provides access to leading cloud technologies and services for scalable computing; 2) enables the sharing of data, software, and capabilities through well-documented and hardened application programming interfaces based on industry standard protocols along with proper access controls; and 3) gives all DoD developers and scientists access to the tools and resources they need to drive new AI capabilities.

The armed services have made great strides building out this digital environment with programs like Platform One, the Joint Common Foundation, and other development platforms. Each of these efforts is a step in the right direction, but ultimately no one service or entity can own and support the full ecosystem required. Rather, the Pentagon should follow a federated approach, building on and connecting existing resources and pathfinder efforts, and linking individual agency initiatives through interoperability and adherence to a community-driven, open architecture.

Perhaps the most difficult change DoD must make in the coming years is the transformation of its warfighters and workforce to ensure digital relevance. Our military must embrace new and disruptive emerging technologies, including AI. To effect this, personnel at all levels—from our most senior leaders to our junior servicemembers—must understand how these technologies unlock new capabilities. To do so, we must provide them with the knowledge and skills required to employ AI as an opportunity to augment, or even displace, current methods.

Moreover, we must integrate civilian and uniformed technologists into our warfighting teams, and provide them with appropriate and rewarding career paths and opportunities. Talent management was a major focus of mine as secretary of the Army, when we began overhauling the ground service’s personnel management system. There is a great deal to be done here across all of DoD, however, and it will take considerable time and effort. But “people” are our most valuable resource, so top-down leadership is required if we are going to drive the change needed to recruit and retain the best our nation has to offer in the years to come.

Much has been accomplished the last couple years to make the US military “AI-ready” and able to win the future, but much work remains. I am acutely aware of how difficult it is to prioritize, reform, and redistribute money within the Department of Defense. But at the end of the day, we must do what’s right for our warfighters and our nation’s security—and at this time, the right thing to do is dedicate more resources to building our AI capabilities and the other ingredients that will ensure DoD’s success. Making the tough prioritization decisions and defending them before Congress will help set the conditions for the Pentagon to out-innovate and out-deploy our adversaries in this crucial space, thus ensuring US and allied overmatch against China and Russia in the decades ahead.

No comments: