The future of AI in military operations and strategy

The next digital arms race has begun, with artificial intelligence (AI) emerging as a pivotal asset in military strategy and national defense. This is no longer a theoretical exercise—AI drives real-time decision-making, enhances operational resilience, and secures competitive advantage in an increasingly volatile global landscape. As governments and defense sectors invest heavily in its development, AI’s integration into military frameworks has become a critical determinant of security and strategic success. 

The decisive advantage of today’s modern military force lies in its employment of edge AI: artificial intelligence technology deployed not only on the cloud but also as part of a warfighter’s tactical kit on platforms like drones, underwater vehicles, and wearables. When tactical devices can package AI models, processing power, and sensors together, these platforms can help the warfighter make precise, split-second decisions faster than the adversary.

Military forces that employ edge AI can lean on technology to help with split-second decision-making and save lives. To that end, research is currently underway on use cases that support military operations and armed forces to develop edge AI capabilities as a decisive advantage over adversaries. Adversaries also perceive the urgency of building AI into their military operations and are working on their own solutions. 

AI in the digital arms race

The current conflict in Ukraine offers a glimpse into the future of electronic warfare, one defined by a non-stop cycle of capability deployments and resultant countermeasures. Today, most of these engagements involve warfighters modifying physical assets (drones and tanks) to achieve mission goals. In the next fight, when AI-enabled devices are the norm instead of the exception, the same engagements will be centered around the rapid pace of software and AI adjustment at the operator level. The way to empower warfighters to win tomorrow’s battles is to provide interoperable, quickly adaptable AI solutions suited to a variety of mission types. The battlefield advantage lies with the side that not only wields the best technology but the one that achieves rapid adaptability at scale.

Consider some advanced use cases for AI in military operations:

• Underwater threat detection: With the help of computer vision models deployed on unmanned sensor platforms, sailors can safely scan and clear commercial shipping zones or contested waters.
• Real-time threat detection and target recognition: In real time, machine learning (ML) models are employed to autonomously recognize and classify various threats, from vehicles to aircraft and personnel. These capabilities enable human operators to make quick, informed decisions.
• Autonomous reconnaissance and surveillance: Edge AI enables real-time data processing on drones, unmanned ground vehicles (UGVs), and naval assets, allowing them to detect and track enemy movements without relying on cloud connectivity or consistent human intelligence. This ensures continuous intelligence, surveillance, and reconnaissance (ISR) even in contested environments, providing uninterrupted situational awareness. 

“It’s going to be important to use AI as a deliberate set of tasks and functions enabled over time,” says Lieutenant General (Ret.) Dave Bassett, “where you can see what the system is doing on behalf of the soldier. This is not about having the system decide which targets get prosecuted, but the system could identify targets that are difficult for the soldier to see. It could help nominate targets that would make a gunner’s task significantly more effective.”

AI, human oversight, and ethics

The use of AI in military operations raises new ethical ambiguities and risks around human control and the safe, ethical development of AI that require both domestic and international guidance. As a domestic guideline, DoD has issued its Ethical Principles for the use of Artificial Intelligence to ensure capabilities are to be responsible AI, equitable, traceable, reliable, and governable, balancing ethical standards with “the US military’s strong history of applying rigorous testing and fielding standards for technology innovations.” To identify, track, and improve the alignment of AI projects to those Ethical Principles, DoD’s Chief Digital and Artificial Intelligence Office (CDAO) has published the Responsible Artificial Intelligence (RAI) Toolkit. The process outlined in the RAI toolkit enables traceability and assurance of responsible AI practice, development, and use.

The decisive advantage of edge AI

While responsible AI ensures ethical and reliable systems, the shift toward edge AI unlocks the real-time processing power and autonomy critical for military decisiveness in dynamic battlefield environments. Edge AI involves deploying AI capabilities directly on the devices and sensors that sense, monitor, and record data. Devices in an edge AI deployment have the processing power to play an active role, running AI models locally and performing tasks autonomously.

Instead of shuttling data from in-theater sensors to a cloud or server for processing, edge AI focuses on bringing computing power to the sensor. Processing data at its source eliminates the network roundtrip of sending input to a remote computer for analysis and then sending output back to the source device.

Edge AI has the advantage of lower latency (i.e., less time lost) than traditional client-server architecture. More importantly, edge AI in military applications offers the precious advantage of independence from cloud connectivity. Now that the battlefield extends to data networks, the side relying on a functioning, trustworthy network is at a disadvantage.

Warfighters who don’t have to worry about remaining connected can address rapidly changing threats more effectively. As Rear Admiral (Ret.) Doug Small observes, “That necessitates the pursuit of edge AI, and the capability to run models in disconnected environments and trust that they are going to do their job exactly as intended.”

Lieutenant General (Ret.) Jack Shanahan concurs. “It was obvious to us that the fight of the future will be in an environment where you’re going to be disconnected,” he says. “You’ll have adversaries trying to jam, cyberattack, and physically attack. You may not have access to the cloud for days or months at a time. You’ll need capabilities at the edge, from the theater level all the way down to an individual somewhere.”

AI development for military application

There are three main obstacles to building AI applications and updating existing applications with AI:

• Long development cycles
• Security in an environment of constantly changing conditions
• Flexibility in updating and re-deploying models at the edge

Those obstacles are especially prominent in the context of military applications.

“First, I think there’s a dramatic business case around providing the kind of tools that let developers easily pick the right existing AI model for their application and their deployed hardware environment,” says Lt. Gen. (Ret.) Bassett. “If you pick wrong, it can be time-consuming and expensive to rework all that software, and it can delay the capability delivery.

“The second piece is that the Department of Defense is such a dynamic operational environment that, no matter which model you pick, you’re going to have to retrain it or adjust and change it over time, as the environment around that model changes.

“Finally, you’ve got to have the flexibility to update those models where they’re deployed. The best model in the world doesn’t do you any good sitting in the lab. It needs to be in the hands of the troops in the field.”

Therefore, successfully developing and deploying military AI applications means addressing the model life cycle from the first deployment to the next update. The iteration must be rapid, and if these models are distributed into the field, they must be secured.

Standardized development for rapid deployment

Shrinking the AI development cycle to get AI systems in the hands of the operators requires standardization on a scalable, trusted, and modular machine learning operations (MLOps) pipeline. Military organizations are developing capabilities with a modular open system architecture (MOSA), to allow for plug-and-play integration and consumption of AI capabilities, both from government and industry partners. Today, the military is creating public-private partnerships through DoD organizations like the Defense Innovation Unit and Chief Digital and Artificial Intelligence Office to enable: 

• Developers with varying degrees of experience in ML design to create production models quickly
• Accelerated test and evaluation with standardized tools
• Streamlined, trusted MLOps pipelines that facilitate adjustment to data, models, or hardware

Enabling human control in the field

At the most tactical level, AI needs to meet diverse mission needs. It is more than just bringing powerful AI to the edge; it means providing for human control by empowering the operators to analyze results and make adjustments during the mission. While in the field, operators need to take filtered data from edge devices and enhance it based on mission requirements. They need to be able to integrate additional data (such as new training data acquired while in the field), update existing models, redeploy, and forward insights to their command center. To do so, they need durable tools with user-friendly interfaces and low-power, low-bandwidth capabilities.  Tools like Latent AI’s Ruggedized Toolkit are designed to enable operators to adjust and redeploy models, integrate with their local tactical network, and operate offline. 

Latent AI proof points in defense applications

Latent AI’s Latent AI Efficient Inference Platform (LEIP) has been shown to enhance military readiness and adaptability. It combines an ML development pipeline with optimization and security in a single framework that is ready for impact level 5/6 (IL5/IL6) integration.

In Project Linchpin, the Latent AI enables the US Army with LEIP to optimize and secure AI models for various hardware targets within a unified workflow. LEIP covers all stages of the machine learning lifecycle, including design, testing, integration, deployment, and monitoring, ensuring that the Army remains agile and prepared in diverse operational environments. The platform meets Army requirements for security and provenance tracking to protect sensitive information and for rapid deployment through quick updates to models in the field.

LEIP has been selected for the US Navy’s Project Automatic Target Recognition using MLOps for Maritime Operations (Project AMMO) for model optimization and secured edge AI deployment. Model updates that previously took six months occur in just a few days (a 97% reduction), improving the Navy’s ability to respond to real-world situations. And, in human terms, using reliable, adaptable AI models has improved sailors’ confidence in the system’s effectiveness.

Paving the way for the future of AI in military applications

Latent AI envisions a future where collaborative, human-AI teams and advanced AI-powered systems redefine military strategies and activate edge AI as a battlefield capability for military forces. As Latent AI Strategic Advisor General (Ret.) James E. Cartwright summarizes that future, “We have to work together to get our ROIs away from just model accuracy and put it much more on ‘Are we improving the speed [and quality] of decision?’ If we had the opportunity to collect the data at the edge, have [analysis] at the edge, that would change the game.”

LEIP can play a pivotal role in that future, enabling the secure, efficient use of edge AI in military applications. Lt. Gen (Ret.) Shanahan adds, “So you’ve got an architecture piece [LEIP Design], you’ve got an optimization [LEIP Optimize] piece, and you’ve got the edge [LEIP Deploy]. Those three are where we always wanted to go. We couldn’t solve for size, weight, and power, and people didn’t have the techniques that you have at Latent AI to now do this on the edge.”

For more information about Latent AI and edge AI, see “The new frontier of AI: How Latent AI is leading the shift from cloud to edge.”