The defense industry has spent the past several years selling artificial intelligence (AI) to navies as a speed play: faster targeting, faster fusion, and faster kill chains. Daniel “Dig” Gallagher, who has worn the uniform as a fleet operator and been responsible for defense industry profit and loss, frames the problem differently, and the distinction matters enormously for anyone allocating capital in this sector. AI does not make a naval force decisive. It makes a naval force’s existing decision-making process faster, for better or worse. “AI gives the warfighter the opportunity to synthesize vast amounts of information quickly and efficiently,” Gallagher says. “This should enable better decision-making but will only be as good as the decision-making process outlined in advance by the warfighting organization.” Read that carefully and it reorders the entire investment thesis. The constraint is not compute, sensors, or model quality. It is whether the organization has thoroughly defined and scripted how combat decision are made.
The Trust Problem That Kills Adoption
Gallagher’s read on where AI programs stall is specific, and it has nothing to do with procurement timelines or budget authority. It is about belief. “AI adoption in naval warfare will stall when the warfighters lack confidence in the completeness and accuracy of the information,” he says. “If they don’t believe, and have to verify the info, then the decision-making cycle will be interrupted, and the advantage of AI will be lost.”
That is a harsher verdict than it first appears. A system that a tactical action officer feels compelled to double-check is not a slow system. It is a negative-value system, because it adds a verification step to the process. The operator now does the original analytical work plus the work of auditing the machine. Every dollar of the promised speed advantage evaporates, and the program still shows up as deployed capability in the briefing slides. Defense executives measuring adoption by installation counts are measuring the wrong factor entirely. The metric that matters is whether the human in the loop stops verify the process, and that is a trust threshold, not a technical one. Vendors who cannot demonstrate completeness of the information their systems ingest, not just accuracy of the output, will find their products quietly bypassed by the people meant to use them.
Mapping The Decision Tree Nobody Has Written Down
The widest gap between what AI can do and what it does at sea is not a capability gap. It is a documentation gap, and Gallagher describes the missing artifact in detail. “Naval commanders have not had time to work out all the factors in the combat decision tree for all factors yet,” he says. “It is an intricate process currently that only highly trained professionals can do. They assimilate vast amounts of data, know what they need to make a decision, and then test it for gaps. They also assess the risks of taking longer to make a decision and examine the alternate courses of action to delay risk and stay safe.”
Break that description apart and the scale of the engineering task becomes clear. A combat decision is not a classification problem. It is data assimilation, gap testing, risk assessment on the cost of delay, and evaluation of alternate courses of action that buy time without buying exposure. Most of that reasoning currently lives in the trained judgment of experienced officers and has never been externalized into anything a machine could operate against. Until it is, AI can only accelerate the fraction of the process that already exists in explicit form. Gallagher expects the payoff to be substantial, but he is candid about the sequencing. “AI is going to help tremendously, but it will take time to map out the decision tree.” Programs that skip the mapping and ship the model are building on ground nobody has surveyed.
Fail-Safes, Not Autonomy
The autonomy debate in naval warfare has become unhelpfully binary: either machines pull the trigger or they do not. Gallagher’s design instinct points toward architecture rather than philosophy. “We are going to have to build automatic fail-safes into the decision tree,” he says, “points where the AI-recommended decisions are tested against a list of intermediate steps to ensure all the courses of action have been accessed for an optimum outcome.”
What he describes is closer to a dialogue than a handoff. “I envision this as a process that is continuous with the AI assembling and synthesizing information, with the decision-maker able to ask questions and get further information as the decision is being developed.” The commander interrogates the system, while the decision forms rather than approving a finished recommendation. On lethal authority, his position is unambiguous: “I don’t think we will get to AI making lethal decisions for a while yet.” For the C-suite, his warning about the next two to three years is about coordination, and the risk he names is fragmentation. “C-suite executives need to work together to create a system of AI decision-making, so that as each organization works on segments or parts of the solution, those segments and parts fit together as a cohesive whole. Otherwise, we will spend a lot of time on competing approaches and non-optimized solutions.” His comparison is DevSecOps, where the community first agreed on the shape of the whole, then divided the work. “The community must get together and decide how to best approach the overall solution, so that individual participants can concentrate on a particular part or segment.” Companies that optimize their own segment in isolation may build excellent components that never combine into a fielded system, and in this market that is a slow way to lose.
Follow Daniel “Dig” Gallagher on LinkedIn for more insights on naval warfare, AI-enabled decision-making, and defense technology adoption.
