Nand Mulchandani serves as the chief technology officer of the Joint Artificial Intelligence Center (JAIC).He is responsible for transforming the DOD into an agency prepared to adopt and leverage next-generation analytical techniques supported by modernized software technologies. A veteran of several successful Silicon Valley startups, Mulchandani has introduced an agile, venture capital-influenced approach to capability development within the Pentagon to more rapidly develop, test, evaluate, and field emerging artificial intelligence (AI)-based capabilities across the DOD. He joined the JAIC in 2019 and previously served as the agency’s acting director for a short stint in 2020. Army Sustainment sat down with Mulchandani to discuss the JAIC’s efforts writ large and those specific to driving logistic efficiency and readiness across echelons.
The JAIC was established in 2018 to integrate and scale AI efforts within and across the DOD – how have you and your team worked to best define the center’s mission so it’s able to operate and deliver like an agile startup?
At the highest level, defining that mission upfront is extremely important. In our earliest days, Lt. Gen. John Shanahan, the inaugural director of the JAIC, and I put together a slide that sought to hone and portray our business model and operating plan effectively. We each came to a few key conclusions, some of which were inspired by my time spent in Silicon Valley. The one thing that tech startups always must do well is to identify their operational constraints so that anything they build or develop will generate leverage. So, we, the tiny little JAIC which at the time, were probably 150 people or so sitting in the middle of a two-million-person organization, worked to find the key control points and develop a business plan in a model that would allow us to do just that. From the get-go, we knew we had to put ourselves in our customers’ shoes and understand at a very deep level what their needs are. Then, translating those features and functions into what a product team needs to deliver in order to satisfy those needs can be executed. We knew we needed to make the JAIC a fully customer-centric organization. We’ve gotten at that issue by structuring it very much in line with a tech company to make sure those products are delivered rapidly with customer needs at the forefront—from ideation to deployment.
It sounds like the JAIC had to be very tactical about how it picked its battles from its onset—how did your team approach hiccups or roadblocks in that product development space as you approached full operations?
I’ll start by saying I’ve been at the JAIC for about two years, and that makes me an old-timer in our office—which is a cool dynamic. Congress appropriated money to stand up the JAIC, and we were given a mission to go execute. Our biggest challenge up-front was forging an entire team out of raw material, so to speak. Some of the first folks detailed to the JAIC didn’t have tons of software experience, so that made us be intentional about how we structured the organization up-front by taking the best of both military and classic tech startup structures. There’s a common misconception that tech companies are inherently flat by organization and those in the military are inherently hierarchical, and that’s true to an extent, but there is nuance. In general, tech companies tend toward being very specialized, hierarchical, and functionally oriented. Each employee, down to the most junior engineer, would know what was going on in the company and the challenges it’s confronted with. When I got to the JAIC, everybody thought that the whole organization was flat, meaning everybody was reporting to a three-star general. With that, poor Lt. Gen. Shanahan was just completely overloaded with about 150 people trying to get his attention. So, we had reconfigured our structure, and we did this by creating divisions in terms of who did what and for what purpose. Naturally, we were able to create product manager and owner roles in the likes of private industry with respect to the government context we were operating in. Once we were able to strike that operational balance, then things were able to take off—and all of that was borne by a focus on our people, which I know the Army places a large emphasis on.
People tend to use AI and machine learning (AI/ML) as a buzzword that will hopefully be a catch-all problem solver, but there’s foundational legwork to be done between now and that end-state, it seems. What does that end-state need to look like when AI/ML is effectively scaled and utilized within the Army and the other services, and how will we know when we’re at that point?
I don’t necessarily think we’re ever going to be done, and I don’t believe anybody will ever be done in the space unless we simply stop iterating on tech completely. Here’s a somewhat rudimentary but still functional analogy that I like throwing out there: think of yourself as living in a land before fire, and somebody walks up to you and tells you about their latest hot discovery. They tell how fire will transform everything we do and how we do it. In this case, AI/ML are fire, and JAIC is the Department of Fire trying to integrate everything. With fire, there is no “end-state,” you simply keep finding new ways to better make it work for you. The same holds for AI/ML—we will continue to innovate and adapt it to whatever space we’re in. At some point, it will absolutely become commonplace. AI has already revolutionized certain parts of our lives, and logistics is certainly one of those parts. With logistics, we’re currently at a point where we can leverage what’s called narrow AI, where that’s applied to specific use cases—such as with predictive maintenance. That’s already given us great results, but there’s lots more juice to be squeezed from that orange.
There are several misconceptions surrounding what AI/ML can and will do, which leads to mismanaged expectations. What’s the best and most realistic sales pitch for AI/ML when it comes to enabling sustainment?
I don’t think you need to have heaps of direct military experience to quickly learn how pivotal or foundational logistics and sustainment are to making the Army run. Where AI/ML and other aspects of software engineering, which may not be super germane for this conversation, really come into play are breaking down those silos which enable warfighting operations on a scale that’s incredibly massive—large companies face similar barriers, but it’s the DOD’s scale that really makes the issue unique. What we’re trying to tackle at the JAIC is changing how we think about integrating internet-scale platforms and other systems so we can really break down those silos and integrate sustainment and warfighting functions so that we can really unleash AI/ML to deliver insight at the speed and scale we desire. The Army and the other services are being intentional with how they are building and integrating these systems to most effectively consolidate their critical data in a place that’s easily accessible, which is key to this whole movement. The next step is taking that data and making it ready for AI/ML—the irony here is that the presence of AI algorithms is not limiting; it’s the readiness or cleanliness of the data feeding those algorithms. This all comes down to delivering optimal functionality and fully integrated logistic awareness to a commander in the field, so they are connected to work done on a piece of equipment by a maintenance operator in a depot or elsewhere.
There seems to be a necessary balance between functional expertise and digitizing old processes that limit our ability to best leverage that expertise, and data appears to be at the heart of that issue. How will data be commoditized so it can be used as an enabler of AI/ML?
As I mentioned earlier, the data aspect of this is critical, which is having the data in a clean, AI-ready format. As an example, some of our earliest days at the JAIC spent on various predictive maintenance projects really revolved around data cleaning. We built an algorithm to make sense of handwritten and highly variable maintenance descriptions used as data. From there, building an optimization algorithm is the less difficult part, honestly. Accessing usable data is the biggest hurdle. For instance, we worked with the 160th Special Operations Aviation Regiment on algorithms to predict engine fatigue and failure. At the start, all that data had to be sourced from hard drives and downloaded. In other cases, the Army may not even own this data, which is another challenge that we, as the DOD, are working to address early in the system acquisition process. The real next question will answer how we need to continue to leverage the experience and intelligence of the logisticians that are doing that really challenging work to ensure what we develop on the AI/ML side of the house supports those efforts as opposed to just running parallel or even serving as a barrier to that work.
How well are we approaching these developments from a joint perspective?
There are two ways to look at this since everything we do at the JAIC is, of course, joint in nature. When I first started with the JAIC, I was warned that service-specific constraints, like budgets and priorities, will make integration a huge challenge. While we’re still evolving as a joint force in terms of AI/ML, we’ve made huge strides in some foundational efforts which will support that integration at scale in the future. Namely, we’ve worked to establish the Joint Common Foundation (JCF), to build a common development environment for AI/ML that anyone across the department can use. We now have all the services developing within the JCF, which is amazing to see and a huge accomplishment in advancing that joint concept. We still have work to do to really break down those data silos we talked about earlier, but this is a big step in the right direction for the Army and DOD writ large.
What excites you most about what the JAIC is doing to support the sustainment space in the near and long term?
I cannot think of a more important system or suite of operations, such as those for sustainment, which will connect to the broader electronic fabric that we’re working to build. There’s a wealth of optimization capability that’s already built around logistics and sustainment that we can apply now and in the future at scale. The next frontier is ensuring we have the right digital systems in place with the right architecture and platforms to go execute this. Our military’s logistics operations are world-class, and these functions are imperative in perpetuity. We now have the opportunity to accelerate these capabilities and do them even better, which really excites me as we work to keep moving ahead in such a competitive environment. I’m confident we’ll build on the wins we’ve already achieved as a joint force to keep advancing in this space on behalf of logisticians and sustainers in the Army and across the services.
Lt. Col. Altwan Whitfield is currently serving as the deputy director of the Army G-4’s Logistics Initiatives Group. Previously, she was the commander of the 841st Transportation Battalion at Surface Deployment and Distribution Command. She holds a bachelor’s degree in Special Education from Converse College in Spartanburg, South Carolina, and a master’s degree in Public Administration with a concentration in Education from Troy University in Montgomery, Alabama.
Mike Crozier is a strategic analyst in the Army G-4’s Logistics Initiatives Group. He holds bachelor’s and master's degrees from Georgetown University.
This article was published in the Oct-Dec 2021 issue of Army Sustainment.