When it comes to artificial intelligence and edge computing across the U.S. Department of Defense, we’re no longer in the talking phase, we’re in the doing phase.

That’s evidenced by the Joint All Domain Command and Control effort and the recent introduction of the DoD’s Chief Digital and Artificial Intelligence Office.

Indeed, we’re seeing the positive impact of AI in virtually all areas of industry, but with the most potential in defense operations, across all operational domains (Air, Land, Sea, Space & Cyber). Air traffic controllers can now use AI-type analytics to keep track of “good aircraft” (i.e., a passenger jets) and “questionable aircraft” (an unexpected aircraft moving in an unexpected manner such as traveling at Mach speed or even a weather balloon).

Thanks to AI-driven image processing and its ability to analyze different characteristics of objects, including size, weight, speed and more, military commanders can gain better situational awareness and accurately determine whether an object seen from afar or on the radar is innocuous or a possible threat.

All this information can be processed and delivered in near real-time, thanks to Field Programmable Gate Arrays, which is a fancy term for CPU accelerators. Commanders and troops operating at the tactical edge can make split-second decisions without having to wait for information to be processed at a centralized cloud location. Using AI in this fashion provides new capabilities to commanders without taking away any of their legacy decision-making tools.

Essentially, AI is enhancing leader’s ability to shorten their OODA Loop cycles and making real-time decision making a reality. OODA stands for “Observe, Orient, Decide, Act” – all within a matter of seconds. AI is making it happen right now, giving fighter pilots flying at supersonic speeds and ground units engaged in military operations the information they need, when they need it, to understand what’s happening, and enabling them to quickly make better decisions on what they need to do next.

And yet, successful use cases such as these are only possible if organizations take the right steps to appropriately prepare their data and AI models.

Here’s how:

Before you start down the AI road you need to be clear about your goals and objectives. The bottom line: What problem do you want AI to solve? What information are you seeking? What data is relevant?

Data must be accurate and clean. The less trustworthy the data is in upon ingestion, the more inaccurate the AI’s recommendations will be. This is particularly challenging, as so much information is being collected from so many different sensors, each with its own proprietary operating system and metadata formats.

Data must be de-duplicated. Collecting millions of data points can result in duplicate data sets. This can make it difficult to understand which data to prioritize or result in conflicting information that could adversely affect missions. A Medivac pilot engaged in a rescue mission could receive multiple recommendations about where to fly, or a troop may end up with the wrong intelligence about the location of an enemy deployment. Also, during HA/DR mission (Humanitarian Assistance or Disaster Recovery) delivering the wrong aid to the wrong place can have ghastly results.

The quantity of data being ingested must be managed. There may come a point where workloads overwhelm the AI’s ability to process data. Thus, it’s important for agencies to regulate data ingestion and only include what’s necessary. This practice, we like to call “Digital Discipline”, is especially critical when dealing with edge devices, which, unlike a centralized cloud, can only manage small chunks of data.

Data must be relevant and timely. For example, a military operation could be relying on multiple radar arrays located in the sky, on a ship, and on the ground. While each radar signal is homed in on the same object, data originating from the largest and fastest transmitters will be delivered first. A good AI engine can timestamp the data and differentiate between timely and stale information, with “stale” meaning data that’s milliseconds older than more current data. Yet those milliseconds could be what saves troops in the event of an incoming missile strike, which makes ensuring the immediacy of the data critical. In this environment seconds matter.

Data must be properly formatted so the AI can interpret it effectively. Consider a person opening the same web page on Chrome and Edge web browsers yet seeing different things; the information being digested is the same, but the way it’s presented is different due to data formatting issues. Data must be formatted consistently and properly for the AI to make sense of it and deliver accurate recommendations. Also building in data converters takes time away from data delivery.

Data integrity must be protected. The appropriate cybersecurity measures must be put in place to ensure that data is secure and data integrity is maintained. Because once data is compromised, it not only becomes a national security issue but also compromises trust in the AI system. Commanders’ and troops’ confidence in the AI will be undermined, and they will begin second-guessing the computer’s recommendations.

AI engine must be trained and kept current on mission objectives so that it can deliver outputs and recommendations to meet those goals. None of the above tactics will be very useful if the AI itself does not understand the mission objectives.

By addressing these points, defense organizations will not only be able to deliver more effective battlefield operations, they’ll also be able to gain better insights to handle events that impact soldiers’ and civilians’ lives. For instance, the military will have greater situational awareness when it responds to domestic crises, such as a hurricane or other natural disasters. Supply chain managers will be able to pinpoint and predict the availability of the beans, bullets, and band aids as well as other critical supplies.

In short, when one talks about “military intelligence” today, they’re likely referring to how hard-fought information is gleaned and massaged by AI. That intelligence is shining through and will continue to do so if agencies take the right steps to ensure the cleanliness, accuracy, and integrity of their data.

Jason Dunn-Potter is Solution Architect for Defense and National Security at Intel and a retired U.S. Army Chief Warrant Officer 5.

Have an Opinion?

This article is an Op-Ed and the opinions expressed are those of the authors. If you would like to respond, or have an editorial of your own you would like to submit, please email C4ISRNET Senior Managing Editor Cary O’Reilly.

More In Thought Leadership