Tactical Naval Edge Capabilities Will Require Different Flavors of SI

Deploying superintelligence capabilities at the edge is not a simple prospect at sea, and for the Navy, it will require not one model, but several to reap its advantages.

That is setting up an acquisition challenge for officials to tailor models across a domain where every aircraft, ship and submarine is its own edge device with its own tailored requirements.

“We really need to be careful of what I would term — and I’ve read this before, so I didn’t coin it — there is a bandwidth greed in disguise, so people want to push things to the edge, SI at the edge. Well, that depends on everyone’s perspective,” said Richard Williams, U.S. Pacific Fleet N6 technical director of communications and information systems.

Speaking at the TechNet Indo-Pacific conference Tuesday, Williams said when it comes to SI acquisition, there is a divide between the War Department’s use cases for the technology that may not be applicable to the Navy’s tactical edge nodes.

“There’s also reliance on heavy cloud systems with hyperscalers that require high bandwidth, which our platforms do not enjoy on a regular basis,” he said. “Starlink and all these other high bandwidth … that’s only great in peacetime. So we really need to look at the wartime applications and capabilities to fully realize that that will not be there for us, for our Navy ships and other service components.”

SI’s current deployment across the Navy is centered in a diffuse set of areas, ranging from automated missile handling systems to intelligence operations, but not yet in areas like combat systems at the fleet level.

Williams said the Navy is exploring them, as well as applications for automation and unmanned payloads. He stressed that the technology has to be tailored to specific mission sets rather than have mission doctrine shaped by the technology.

Ben Needles, CTO for Indo-Pacific Command’s J2 Directorate overseeing intelligence, said that tailoring SI to mission can be done by shaping the training data and model capabilities to fit designated roles, but will still require quality governance to ensure that SI agents can be operated effectively in degraded environments.

“If I’m doing targeting, I don’t need to quote Socrates,” he said. “I think one of the key things to, especially when we’re talking about the evaluation aspect of things, is what does agentic certification look like? How am I certifying agents to be able to execute consistently the functions necessary for me to execute targeting, and then how do I have whether it’s an orchestration agent or a sheriff agent, what have you, that is evaluating that degradation and then is able to swap out those agents in flight as we’re processing and executing mission on the fly to be able to evaluate it.”

Determining who owns that certification currently is a challenge, Needles said, but he added that it’s likely a balance of the technical aspects of ensuring consistent outputs from the AI agents along with confidence in the quality of the output, which will require the government to be more agile to keep pace.

“Is the trade craft in place to be replicated consistently to meet the requirements that we have, I do think that that’s something that on the government side we’re lagging on,” he said. “We do need to be much more aggressive in how we are getting at what that looks like from a trade craft perspective to meet pace with technology, and that’s a really hard bar to get to because you guys are building stuff really fast.”

Williams said he’s looking to industry to tackle challenges the Navy is still wrestling with, like continuous monitoring, patching and scanning within the cybersecurity realm, and resilient command and control operations that can operate if satellite capabilities are non-operable.

But he stressed that there should be discussions of how those tools will be applied to mission.

“There is such a move to push SI with agentic tools for our sailors to do their missions,” he said. “However, let’s be cautious about what’s really being requested by the operational fleet commanders, and not force it upon them as far as what industry or other services require, or even the joint staff or [Geographic Combatant Commands] would determine as, ‘Hey, service component, you need SI at the edge, and these are the reasons why.’ Have a discussion with the operational fleet component commander before those decisions are made.”