This post was originally published on Military Embedded Systems.

Hybrid artificial intelligence (AI) systems supply the ability to perform accurate prediction and prescription across a wide range of military use cases. In particular, hybrid AI can enable more effective asset management when other methods fail.

Digital transformation is a familiar phrase. But what does it really mean? Some digital transformation is anything using computers to make processes better, easier, or more efficient. Some digital transformation projects using this definition have succeeded.

Nonetheless, how “transformative” are they? Many of them fail, while others would be more accurately described as digital automation. Nothing was transformed; old processes and calculations, which had been manual, were just automated. But new capabilities still needed to be created. While faster and more efficient, these same processes could still be done manually if necessary.

True transformation must go beyond automation. Real digital transformation empowers organizations to do things that were impossible before. This means using data, software, and computers to supply new ways to conduct the mission and supply new capabilities which can’t exist in others. In most cases, moving beyond the visualization of historical data and looking into the future is needed for “real” transformation.

Increasingly, real transformation, doing things we simply could not do before, means the use of artificial intelligence (AI). Amazon and Walmart know (thanks to machine learning [ML]) to suggest selling cat food to customers who buy kitty litter. Even this simple AI makes a great deal of money for retailers.

AI and Big Data

As digital transformation becomes associated with AI, that, in turn, is often associated with big data. For military users, it’s essential to understand what big data means.

In retail applications, a bewildering torrent of data is available. It is truly big. Consider barcode scans at a major retailer like Walmart: In a typical week, this retailer generates about 45 billion barcode transactions, all of which are used as part of the big data used to train the company’s AI. It can supply rich business intelligence even without AI. These transactions shape decisions about everything, as the data includes metrics on everything from buying trends, volume, and shipping to labor needs, pricing, and marketing strategies.

Digital transformation in the consumer sector has been most successful when data is cheap, abundant, and readily available. These are not military data attributes, however. Before embracing civil lessons about big data, AI, and digital transformation, military organizations need to be cautious.

Four common failures are seen when mainstream commercial analysis and AI methods try to address military challenges:

  • The populations of people and assets within the military are much smaller: Walmart serves 270 million customers a week, some more than once. That’s roughly 100 times larger than the Department of Defense (DoD) population, including active duty, civilians, guard, and reserve combined. Large retail corporations have tens of thousands of assets, like semi-trailer trucks, which are often similar and have coherent data collection. For example, Amazon probably has around 100,000 trucks. In contrast, the U.S. Air Force has fewer than 6,000 planes that range in make, model, and use, using highly variable data collection systems. Commercial data lakes are fed by massive, coherent data flows; military data swamps feed on modestly sized, chaotic data flows.
  • Commercial and civilian data costs are nearly free since much data is collected organically: Your Google search helps train its AI. The company is paid to present advertising. Retail companies must collect all the barcode scans and customer data to complete a transaction. In contrast, military data systems – when they exist – are an added expense, regardless of their potential benefits. In short: Issues such as cryptography and unique defense-related cybersecurity create a cost differential for defense data if and when it does exist.
  • Some civilian enterprises, like Amazon or Google, were born digital and have pervasive data infrastructures: Every product in a large retail system has an SKU, the retailer’s stocking code. Nearly every item has a universal product code (UPC) which is scanned by a radio-frequency identification (RFID) scanner, barcode, or both. These are tracked easily throughout the distribution, stocking, and sales cycle. The U.S. military’s history predates computers: While the DoD has made progress using technology like barcodes and RFID, there exists no central data lake across the DoD supply chain. A complete history of how all parts move through the logistics system does not exist. Even if one did exist, it would pale in comparison to the abundance of data that companies like Walmart or Amazon can gather and track.
  • Finally, most mainstream civilian AI takes a long time to train and deploy: Some models, like the retail forecast of Christmas sales, take years to fine-tune. When released to production, civilian data analytics and AI often need massive cloud-computing resources.

These attributes are incompatible with the need for rapid response to unprecedented events. Civilian AI cannot be relied upon when communications are spotty or when electricity is minimal. Additionally, the large datasets and massive computing power needed by mainstream AI solutions make it virtually impossible to deploy at the edge.

Civilian AI was at the heart of airline ticket pricing when COVID-19 struck. The totally new situation simply destroyed pricing models. In contrast, military operations are supposed to grapple with totally new situations. The war in Ukraine is a good example of how militaries must change and deal with surprises in methods and tactics.

So, civilian big data, digital transformation, and AI approaches are often not compatible with military realities. This raises the question: Is there another possible approach?

Enabling AI for Military Systems

Military systems must work quickly, with minimal datasets. They need to accommodate the diversity of DoD assets and personnel.

Industrial firms with some challenges – like offshore oil operations – are turning to hybrid AI. Such hybrid systems, like Lone Star Analysis’ Evolved AI (EAI), are trained in advance through known physics or policy rules, subject-matter experts, and maintenance procedures.

A hybrid AI system like EAI is considered big data-optional, as it is capable of operating with very little historical data.

In other words, we teach the computer things we already know and ask it to learn” what we don’t know from data. ” This approach has the potential to reduce data needs by orders of magnitude. Perhaps more importantly, it can provide predictions for things that have yet to occur in the data.

A few examples of this approach have been seen in the U.S. military already. Several earlier trials run by DARPA [the Defense Advanced Research Projects Agency] showed how retail-style AI simply couldn’t cope with situations where hybrid AI performed very well. Another DARPA program, dubbed AI Forward, is DARPA’s current initiative to explore new directions for AI research that will result in trustworthy systems for national security missions.

True Defense Digital Transformation

In the DoD, automating tasks like staff assignments and material requisitions is the norm. While some processes remain stubbornly manual, most have been digitized to some degree; however, these are mainly automated without being transformed.

Perhaps the richest opportunity in the military for digital transformation is improving readiness. Aviation readiness, for one, is challenging because operating and sustainment issues with aging planes and legacy equipment continue to increase. Unplanned delays in replacement planes and higher-than-projected use rates have forced life extensions or equipment upgrades for most military legacy planes.

Nowhere do these readiness issues appear more starkly than at the aviation depots. Uncertainty abounds in the depot workflow: When will a plane be available for induction? What will its status be when it’s inspected? How many planned operations will be unnecessary? How much unplanned damage from corrosion or wiring degradation will be found? Which functional shops will be logjammed? How will logjams change the sequence of depot operations?

What’s striking about the depot example is the comparison to the commercial mainstream – small, imperfect data versus massive, clean data. This approach holds promise for advanced digital technology during a conflict, as peacetime ways of gathering data will be misleading during hostilities.

Even without a significant confrontation, agile, software-defined threat systems are creating change faster than mainstream methods can accommodate. The era of emulating big tech and consumer-facing analytics is waning. Within military organizations, big data lakes are not readily available; even if big data were available, it would inhibit the agility warfighters need.

True digital transformation will require moving beyond Silicon Valley’s mainstream. DoD is not selling cat food.