For Artificial Intelligence to Really Work, The Military Will Need Better Connectivity

While talking with CENTCOM’s chief technology officer, Schuyler Moore, I opined that in a peer fight the U.S. military’s data and communications networks may not hold up for even 24 hours. She waited a beat before saying, ‘I won’t comment on that but… heard and received.”

In the event of a conflict with China or Russia, what U.S. forces may be able to hear and receive will be critical. That includes the data that they will seek to mine for tactical and strategic insights using artificial intelligence (AI). At a recent conference in Austin, Texas, Ms. Moore said that simple connectivity is “at the core of anything related to” AI’s real-world adoption.

If that connectivity and a working, layered infrastructure for passing data from forward platforms and operating locations back to places where AI can be applied is impaired or disrupted then even the best algorithms are of little use.

Aiding Connectivity and Computing

Moore breaks down the means by which CENTCOM and the broader military implement AI into two chunks – connectivity and computing.

“Connectivity means, can I receive the data I need in order to apply artificial intelligence and can I send the outputs of that where they need to go?” Computing, she says, is a question of whether users can run algorithms where needed.

When discussing AI infrastructure, starting on the computing side is useful Moore suggests. To bring AI to bear, she explains, we have to ask where we are trying to do computing and processing of raw data? Is it done on a physical system whether on a Navy ship or with a soldier out in the field holding a device? Is it [done] at a headquarters?

Depending on where AI analytics are run there are associated size, weight and power (SWaP) issues. Having right-sized hardware becomes critical. And while having the ability to apply AI on-the-spot would be welcome, it’s not currently practical.

But there is work being done at CENTCOM and elsewhere to disperse processing power so that data can initially be run through limited AI at the tactical edge. “Then you can push that data back to another point that has a bit more computing power where they can run advanced analytics,” Moore says.

The commercial sector is leading the charge on improving computing capacity in small packages, Moore adds, which may allow CENTCOM and others to do denser AI processing near the front lines or at them. In fact, connectivity, or the lack thereof, may force a reliance on edge processing.

“Our ability to access bulk data determines where the processing has to happen,” Moore observes. “For example, imagine an unmanned aerial vehicle collecting video feed in a certain area. Depending on the bandwidth available, it may only be able to push so much data back to its headquarters if it’s beyond line of sight. That means you need to be very specific about what data you’re sending back. That creates demand for processing at the edge.”

Processing more data at the edge could relieve bottlenecks but so, obviously, could creating more bandwidth, more connectivity to ensure that more data can be pushed to a centralized processing location. “We’ve experimented with that at CENTCOM [using] radio-mesh networks, low earth orbit satellite networks and a number of other solutions that allow us to attack that problem,” Moore says.

The results have been encouraging. Moore explained that radio-mesh network experiments have largely been done with the Navy.

“We’ve been seeing how far we can extend unmanned surface vessels at sea and have them still be able to push back meaningful data to headquarters. Those experiments have been promising. We’ve had the ability to have UAVs aloft serving as network nodes, other surface vessels serving as network nodes. The more stuff you get out above, on and below the water, the more robust you’re network and the higher the odds of being able to push back data.”

Moore acknowledges that such a “radio-repeater” type of arrangement is stronger if some processing is done at the edge, reducing the volume of bulk data that must be transported back to a headquarters for AI processing. The combined edge-mesh approach dovetails with the agile combat employment (ACE) concept the Air Force is pushing and similar dispersed operational concepts the other services are evaluating to adapt to the challenging distances and tactical obstacles of the Indo-Pacific.

Finding a Place for AI Data

While these experiments and others CENTCOM has undertaken offer promising potential to expand connectivity in support of AI, they still rely on the physics of the RF spectrum. Adversaries and allies alike understand the high and low frequency options for such data transport intimately and understand that in any conflict bandwidth demand will be at a premium, limiting “space available” for data transport for AI.

Disrupting connectivity is at the top of the to-do list for any U.S. adversary and for American forces themselves so figuring out how much bandwidth will practically be available in a serious fight is a matter of speculation. Recent experience in Ukraine has demonstrated that cyber conflict may be more limited and attritive than originally thought but the manner in which the war is Ukraine being fought likely differs substantially from how a conflict in the Pacific may be waged.

Moore won’t comment on what chunks of spectrum AI infrastructure may most heavily rely on. I asked if CENTCOM had experimented with photonic communication. Moore replied that CENTCOM is testing “a range of spectrum options and are encouraged by the results we’ve seen”.

The Army sees line of sight, beyond line of sight, fiber, host nation, 5G, and SATCOM with LEO
LEO
, MEO, GEO satellites as communications/data pathways for its Integrated Tactical Network and the Integrated Enterprise Network which endeavor to create a “Unified Transport Network” combining command and control with a ubiquitous data layer.

Units from the company up through the battalion and brigade to division, with other maneuver type elements sprinkled in, will seamlessly access the UTN and AI data/processing would reside therein. It’s a construct similar to the Pentagon’s broader Joint All Domain Command and Control (JADC2) effort, popularly referred to as an internet-of-military-things. But an awful lot has to go right for such seamless connectivity to remain seamless in time of war.

While Moore won’t be pinned down on how much space-based assets will be relied upon for enabling the connectivity that increasing AI use may require, she acknowledges that space will be important. Even if CENTCOM and others rigorously seek to exploit alternate pathways for data and processing, the position, navigation and timing (PNT) data necessary to accompany meaningful application of AI is at risk and it is largely space based.

Increasing edge processing- effectively localizing AI compute functions combined with human analysis in edge environments – may take some of the data transport burden away. But it too can be compromised if the space layer cannot yield the accurate PNT awareness needed to for AI inputs.

Moore adds an appreciated dose of reality to the infrastructure discussion when explaining CENTCOM’s outlook on the demand for AI output.

“Our view is that sometimes people incorrectly regard AI as giving you the answer to your homework. The reality is that it can point you in the right direction or cut down on some of the research time but at the end of the day, decision-making has to sit with a human operator. We’re learning where AI is most effectively and efficiently deployed.”

That includes considering the risk that bad data – regardless of how effectively it is transported and processed – may lead to faulty AI insights. For now, it’s not a large risk Moore says because “We’re applying it in such limited circumstances right now that you’d be hard pressed to say that decision-makers are overly influenced by it.”

“We are experiencing live,” she adds, “what the benefits and risks associated with algorithmic applications are and I think that because we’re taking an incremental approach it’s allowing us to set expectations for our leaders accordingly.”

As CENTCOM and others try to strike a balance between the discrimination needed to make the most effective use of AI with the means of processing it, they more clearly see the importance of the connectivity it requires and the challenge of maintaining it.