Commentary

How upskilling analysts with AI is meeting 4 key DoD data modernization challenges

Within the DoD’s massive data repositories lie solutions to problems ranging from how best to counter cyber threats to optimizing talent management to avertin...

Kyle Rice

December 7, 2023 3:01 pm

6 min read

As the largest enterprise in the world, the Defense Department collects dozens of terabytes of data every day from external sources, internal systems, connected platforms and personnel around the globe. Making sense of it all is a complex mission, encompassing everything from evaluating sensor signals on a ship patrolling a distant ocean to detecting electromagnetic interference patterns and anomalies in space.

To complicate matters, there’s a shortage of data experts. There aren’t enough data scientists to fill a fraction of the need, and talent pipelines can’t keep up with the expanding demand.

Fortunately, the answer to this thorny problem lies in two things the department is already really good at: bootstrapping existing talent and technology adoption.

Here’s a rundown on how defense analysts can be upskilled by using AI-driven data exploration, helping DoD meet four big data modernization challenges:

Getting faster answers from data

Filling the talent gap for data experts

Keeping pace with rapidly changing threat scenarios

Keeping humans at the center of decision-making

Challenge 1: Getting faster answers from huge volumes of data

Within the DoD’s massive data repositories lie solutions to problems ranging from how best to counter cyber threats to optimizing talent management to averting equipment downtime to ensuring operational readiness. AI-guided data exploration leverages AI to find what really matters within these enormous datasets, at speed.

Using such intelligent exploration, an analyst can hone 20 million records into 20 groups that are important, then be guided to areas within the data that are statistically significant. Analysis can be done at scale, with less risk of bias or guesswork, before a predictive model ever gets built.

The Air Force Global Strike Command (AFGSC) is using AI-enabled data exploration to improve mission-capable rates, quickly figuring out what’s causing aircraft to be unavailable then pinpointing the optimal ways to get them back online. Such data analysis starts by breaking the problem down into its essence: “What’s causing aircraft to be grounded?” All relevant datasets — maintenance records, inventory, supply chain data, work unit codes and personnel resources — are holistically evaluated. The machine learning algorithm rapidly identifies significant patterns and insights within this data.

Analysts can locate the source of the problem with granular visibility. What parts are needed for repairs? Are inventory problems from a slow parts pipeline creating delays? Are there maintenance personnel gaps triggering installation delays? Commanders can fix issues before they happen, like seeing a personnel shortage from a forthcoming engine retrofit with enough time to implement a training plan.

Challenge 2: Upskilling data analysts to fill the demand for data expertise

According to the FBI, even if all its cyber agents and analysts focused on the China threat alone, Chinese hackers would still outnumber FBI cyber personnel at least 50 to 1. This illustrates why treating DoD’s dearth of data expertise as a hiring problem won’t be enough. Data science grads and cyber experts can’t be churned out fast enough to catch up with the exponentially growing need. Especially considering the attractive private sector opportunities for this talent.

The solution is to equip the department’s huge cadre of existing analysts with easy-to-apply, no-code AI techniques. This cohort of hundreds of thousands has the drive and aptitude to become more data-science savvy. They can perform sophisticated analyses that used to be reserved for data scientists, using out-of-the box algorithms to wrangle even the most knotted datasets into manageable data clusters or segments.

Plus, since these analysts are already subject matter experts in their domains, they can be both sophisticated data users and healthy skeptics, ensuring that the AI models are used efficiently and accurately.

Challenge 3: Keeping pace with rapidly changing threats and threat actors

In the AI era, the types of threats and volume of adversaries to ward off are only going to accelerate. Generative AI allows information operations actors with limited resources and capabilities to up their game, at scale. Fortunately, AI can also help tell the threats that need to be addressed apart from those that aren’t of immediate concern.

Take the example of an uptick in social media activity somewhere in the world in the days before a cyber incursion took place. Analysts can explore what happened, generating visualizations that illustrate the relationships, anomalies and trends hiding in the signals. They can leverage the AI system to help determine where to focus their attention.

Once this AI-enabled analyst understands the environment, it becomes possible to see what online activity likely foreshadows future cyber incursions. This knowledge can be used to harden particular facilities or utilities — and also to train predictive models built to react when similar activity is observed. First understand which hidden signals are important indicators, then build a model to predict those behaviors in the future.

Challenge 4: Keeping humans centered in data-driven decisions

The purpose of AI systems is to help humans keep up and stay ahead, not to replace people with machines. The tragic October 2023 terrorist attacks against Israel from Gaza illustrated the fallibility of overly automated defensive systems. Israeli border surveillance was heavily reliant on cameras, sensors and machine guns operated remotely, and these were easily shut down by the attackers. The imagination of the attackers exceeded the training of these systems. Involving more humans in direct monitoring may well have made a difference.

Another area of risk in automation is that a computer model only knows what it’s been trained on. If an adversary knows how an AI system has been trained, they can forcibly trick it.

These are far from the only reasons it’s so crucial to keep an analyst in the loop in AI systems. The insights generated by AI must be transparent in order to inspire confidence among all stakeholders. DoD’s data strategy for becoming more data-centric calls for data to be accessible, understandable, trustworthy and understood by both those who produce it and those who consume it. Commanders don’t want to put their trust in black-box AI algorithms for mission-critical scenarios. They want to know the “why” behind findings.

AI-guided data exploration keeps humans at the center of analysis. It makes advanced analysis accessible across platforms, pay grades and roles, fostering collaboration among all those who need to interpret data. People still make the decisions; decisions are just far better informed thanks to supercharged analytics.

Making sense of massive data streams and shifting adversaries, at speed

It wasn’t long ago that the tools of an analyst were printed maps, printed photographs and grease pencils. AI data analysis is bringing changes as profound as those that came when analysts moved from paper to digital. The good news is that the American workforce has no equal when it comes to pivoting to meet the moment.

AI-driven data exploration is upleveling the tradecraft of rank-and-file analysts. Its DoD use cases are endless: optimizing supply chains to improve operational readiness, evaluating biometric data of pilots in flight simulators to improve training, sorting out real cyber threats from all the noise. Integrating this AI technology into analysts’ workflows keeps human judgment an organic part of data-driven decision-making. It’s the way forward to get accurate insights from data, delivered at the speed our national security requires.

Kyle Rice is federal chief technology officer for Virtualitics.