How does AI improve combat readiness?
Soldiers prepared for combat are not only well trained, but also well supplied. Combat readiness depends heavily on an efficient chain of supplies, including mechanical widgets for all types of complex warfighting machinery. But often getting the right part to the right place at the right time is made difficult by requisition software fraught with bad data, which can prohibit the accurate prediction of supply needs.
A Leidos program with the U.S. Army used AI and machine learning (ML) to help solve this problem, and was recently featured on Open Data Science. To learn more we welcome Dr. Donald Horner, who was Principal Investigator on the program. Dr. Horner is a graduate of West Point (B.S.), MIT (M.S.) and Stanford (Ph.D.). He is a senior analyst at Leidos and a former engineer at Lockheed Martin’s Skunkworks.
Q: To illustrate the importance of an efficient military supply chain, can you remember a time from your service when you weren’t well supplied?
Dr. Horner: During the Bosnia conflict in 1995, I was the forward commander of a heavy transportation truck battalion. The battalion was particularly large given that we had several American companies augmented by Dutch and German companies. Our 'heavy trucks' moved everything from ammo to fuel to M-1 Abrams tanks to M-2 Bradley fighting vehicles. For whatever reason, our heavy trucks were burning up alternators at an unusually high rate. We had no idea why. And, because these alternators were typically not in the Army supply system in the quantities needed, we had several inoperable heavy trucks—which meant that cargo did not move.
Q: How did you solve the problem, and how might AI-ML have helped you back then?
Dr. Horner: My battalion supply technician found a truck dealer in Sarajevo that carried the alternators we needed. We probably bought 25 of these alternators, which were anything but cheap. And, I paid for all 25 with my U.S. Government American Express card. Big invoice. True story. AI-ML is particularly helpful in these non-standard requisitions because they can predict and prescribe items for resupply. Often times these predictions and prescriptions are items that humans might easily miss. AI-ML and deep learning in particular uncover esoteric but important clusters and relationships between variables in the data that humans don't see.
, Senior Analyst, Defense
You don’t want bad data, but bad data is a reality.
Q: The program you write about uses AI-ML to solve a major problem in military supplies requisition. How would you summarize the problem?
Dr. Horner: The Army has this wonderful logistics data set, but it had a lot of errors. An example of this can be something as simple as a person had entered a “one” instead of a lowercase “L” for a stock number, but you can imagine how those errors propagate through the Army system worldwide. These errors can lead to major delays in the supply chain. In the meantime you’ve got a piece of equipment, like an Army Black Hawk helicopter, that is not operational because it needs the part. It can take weeks. And all the while you’ve got an important weapon system that’s not operational.
Low and behold, this is normal with data sets across academia and across industry. You don’t want bad data, but data is a reality. More often than not, we start with bad data sets. So the first thing you have to do is cleanse the data. We started using a cleansing algorithm to see if we could teach it to identify the flaws. We found that we could use AI and ML protocols to accurately identify the flaws. And in the process, we found other flaws no one knew were there. Then we wrote some code that allowed us to fix the flaws almost instantly
Q: What was the key to success for the AI-ML tools you used?
Dr. Horner: The key is you’ve got to teach the algorithms the data by repetition, repetition, and repetition. With gargantuan numbers of repetition, the algorithm detects very hard-to-find patterns. And by this pattern analysis, AI-ML can then cleanse the data instantly when it would have taken humans months. So the work of data cleansing and data correction is highly dependent on this deep learning.
Q: What are you most proud of about the program?
Dr. Horner: An efficient requisition process means combat readiness is improved, and combat readiness means fewer soldier will be put at risk. That’s what this is all about. The whole military supply system is based on supplying equipment to enhance soldier survival. This AI-ML program enhances readiness and increases soldier survivability by getting our warfighters the right materiel at the right time.
Q: What do you envision next for AI-ML in the military supply chain?
Dr. Horner: We’re going beyond predictability and now creating AI-ML systems that are very prescriptive. You get on Amazon today and order a pair of sneakers. You say yes I want that pair of sneakers. What does Amazon do? They have algorithms that immediately tell you that if you bought that pair of sneakers, you probably should have this pair of shoe laces. It’s gone beyond predictability to prescribing for you, the consumer, what you should have. That’s exactly the transformation I see happening in military supply logistics.
Another example is the Smart Cities initiative—well known in the private sector with success in cities such as Columbus, Ohio and San Francisco. ‘Smart Hub’ is the military analog to Smart Cities. Think of 'Smart Hub' as a hyper-IT interconnected base which yields more efficient operations while delivering more visible, highly rationalized, just-in-time, last-mile distribution and logistics services to troops in contact at reduced costs. Smart Hub hyper-IT connectivity refers to multiple connections with real-time information through an array of remote sensors and a robust suite of data analysis tools, including AI-ML. The predictive, prescriptive modeling strengths of AI-ML are a perfect fit for Smart Hub.