EXCLUSIVE: Q&A with Joint AI Center Chief Technology Officer

Nand Mulchandani

Defense Dept. photo by Lisa Ferdinando

Nand Mulchandani serves as the Joint Artificial Intelligence Center’s chief technology officer. He previously led the JAIC as acting director where he oversaw the burgeoning organization meant to better unite the Defense Department’s various AI projects. On Oct. 1 — the first day back in his CTO position following the confirmation of Marine Corps Lt. Gen. Michael Groen as the new head of the center — Mulchandani spoke with National Defense magazine Senior Editor Yasmin Tadjdeh about the trajectory of the JAIC, what the future has in store and how the center is managing through both a pandemic and budget uncertainty. The following Q&A has been lightly edited for length and clarity.

How have you seen the JAIC change and accelerate in the past two years? Would you say it's on the right trajectory?

Here's the way to think about it. Every startup organization — and I wouldn't say technology startup in general — but every organization when it starts up brand new ... it takes time to find your market, your customers, your products, and how they all fit together. It's very complex ... It's hard.

Congress in their obviously great wisdom ... decided that AI was a high priority, it had been festering for a while and they just effectively created the JAIC and pulled in Lt. Gen. [John N.T. “Jack”] Shanahan and a small little team and said, “You're the JAIC, go figure it out.” And literally that was as much guidance as they got at the time.

So, when I joined last year the team was still evolving. And the business model for the JAIC is also very unique. We are an organization that builds technology products, but at the same time we have acquisition folks, we have AI ethics folks, we have AI policy folks. We run the [Executive Steering Group], which is a steering group of three-star and above general officers across the DoD. We have a missions team that is formed out of military and civilians and contractors. ... Our workforce is incredibly diverse with lots of different backgrounds. We integrate folks from across the services.

It is such a unique organism, even within the DoD. ... So it's been very difficult, very hard, but very exciting, too, because you're doing something very unique. I think all of our JAIC employees walk in the door and know that we're unique and one of a kind.

Now, let me address your second part of your question, which is how do you measure success and how do you know you're getting things done?

On our technology and product side, we operate 30-plus different products. We're building 30-plus different products across six different verticals. The point is that the JAIC is very much [built] around getting a spark going or getting a prototype or making a market in some way, and then handing it off for transition and scaling right to a customer.

We're now starting to demonstrate great and exciting success across those products.

[Our products] are now being used to help fight the California wildfires. We've got ... detection products that are in the field. [We’re working with] Special Operations Command, we had an engine health model that used AI to predict engine failures. Well, it's now fully operational and in the field and lo and behold, we are detecting engine failures that basically allow commanders to pull helicopters out of operations should an engine ... forecast a failure. And they’ve found instances where we detected a failure or forecasted one, and indeed the engine had a part or something that was going to fail.

We have other products that are in prototype mode with hundreds of users in pilot and will probably go to production over the next couple of months.

So we gauge success in the sense of customer adoption on the technology and product side. We just held the AI symposium for the DoD, the first ever. Two-thousand-plus people showed up to help build a community of users across that. We held the first AI partnership for defense that came together about two weeks ago — 13-plus different countries attending to start talking and working on AI policy and ethics amongst ourselves and building out sort of a core nucleus of countries that's going to grow.

And then acquisition authority — we're coming up with some unique products. ... [The JAIC is working with] innovative, smaller companies who are doing great work, who would find it very hard to work with the DoD. …

How we measure it is by seeing the deliverables and output because the JAIC is not a research organization. We are out in the field as a delivery organization. And in my 26-plus years in the [Silicon] Valley one of the things that you learn ... is the pat on the back goes to the organizations that know how to deliver products and deliver things to customers.

Research, other things are all great and important, but it's the hard part of taking those ideas and concepts and getting them into actual software and hardware that people are using, or in the case of safe policies or acquisition or community building.

You have previously mentioned that the joint warfighting mission initiative is a flagship product for the JAIC and one of its top priorities. Can you share any details about where it is in development right now?

The [joint warfighting] mission initiative is the broader concept of how do we take AI and help not only improve existing processes or work that we do in joint warfighting, but completely transform the way we go to war. … Our mission is warfighting. It's either defense or offensive warfighting. … Our early products from the JAIC were really focused on kind of starter AI projects when it came to things like predictive maintenance and humanitarian assistance and disaster relief. We were dealing with text, numerical data, string data, etc. The algorithms were not that hard. ... It gave us the ability to do end-to-end work on how to adopt AI for certain products or missions.

Joint warfighting is the hardest problems at the DoD for us to take on because we can't go out in the wide world and just buy warfighting products. They don't exist, right? Our larger vendors don't just sell targeting solutions or … sensor-to-shooter, or tactical edge AI deployment on hardware like a UAV or an airplane.

So that's the work that the JAIC is now starting to focus on. … We are focused on the national defense priorities as they're changing. Great power competition is back. We have near-peer competitors that are also investing heavily in AI and other new technologies. And the JAIC is effectively helping support the national priorities of the Defense Department and the government. ...

Now how we're approaching it — you start with things like human-machine teaming, decision support and a number of other things, because … there's an area of improving the current processes. And so when you bring the machine in or a system into a current process there is a time ... it takes for folks to absorb that technology, to get used to it. There are different ways of displaying information, about communicating information, about absorbing information. So we're spending time with our commanders, with training and education, etc., on how to absorb AI-enabled systems. And we want to do that in a very systematic, deliberate way where we start out with human-machine teaming, decision support, etc., and then work our way towards things like autonomy and others.

Fiscal year 2021 has begun and we're starting it with a continuing resolution. What impact is that going to have on the JAIC and how have you prepared for that?

We went through a CR last year as well. And we have very experienced DoD folks here in the finance side of things who understand that this is not an abnormal occurrence. So believe it or not, we've actually been planning for this scenario literally since the last one ended. So the JAIC has been preparing for this for a long time. ... The good news is many of our projects and products, etc, have actually been pre-funded through much of the money that we got in FY '20. So they don't literally end on Sept. 30.

We have contracts and vendors and other things working months and months out into the new fiscal year, which is fantastic. So we're not in a crisis mode at all.

What this really impacts … is new-starts, our ability to start a bigger new project that we have been potentially forecasting for starting with new FY '21 money. We'll just delay those or move those out a little bit. But also we actually will be receiving some money as part of the CR that will allow us to kick-start some of these new projects and things along the way, and then scale them ... when we get out of the CR mode. So we've been doing careful planning in terms of making sure that the products with long lead times that are well-defined, those got funded in FY '20, and as money comes in, we'll be able to kick-start new products and things. But the great news is we've been preparing for this. So literally we're not in any crisis mode or surprised that this actually happened.

Can you tell us about any of the new-starts that might be affected?

Joint warfighting is really the bigger area that we are focused on. … When you see the two big areas that we're investing in, sort of thematically in JW, first and foremost is what we call autonomy — think of AI for small unit maneuvers. So things like [Special Operations Command and] …how does autonomy impact hyper-local warfighting, whether it be close-combat, whether it be operations that our teams do with SOCOM, or ground-based units that are doing work at a tactical level. So we're doing a lot of work on that front there. And then there's a broader-based work that we do that is more around global things like [joint all-domain command and control] and [Advanced Battle Management System], ... targeting, sensor-to-shooter which tends to be more about end-to-end conductivity and systems that are global in nature. So we've got projects along there now. Most of the work we're starting to do in joint warfighting is at higher classification levels. So broadly speaking, I'd say the new-starts are going to be in those two big areas.

With the JAIC now two years old, how have you seen its relationship with industry change and evolve? Has it become a more close-knit relationship?

Oh my God, yes. ... It was never a problem to begin with in all honesty. ... I think the media always fixated on the older stories about Silicon Valley not wanting to work with the DoD.  ... First and foremost, straight out, the relationship is fantastic. We are working with every single large tech company here at the DoD broadly, but the JAIC specifically. Name an AI vendor and we either have work going on with them, or they're involved in some way in some of the newer projects that we're doing.

Over on the smaller side, there are emergent areas ... in AI that are still at the cusp of research versus development, whether it be things like ethics or testing or training, data labeling, data curation, fundamental AI platforms. ... [There are] lots and lots of vibrant startups. We are super deeply engaged with them.

A lot of people ask us [about] the whole thing with Google and Project Maven and whether that's still ... [a strained relationship]. Well, guess what, I mean, we're working with Google on a number of projects directly ... whether it be health or other types … of products there. We have contracts with Google that we're working on, but all the other bigger vendors as well.

So, it's been great. We have a ... small but very strategic industry relations team headed by Col. Doug Drakeley who is positioned, believe it or not in Mountain View, [California], so he's not in Washington, D.C. He's parked out in Mountain View right in the heart of Silicon Valley. And I spend a ton of my time with industry, with vendors because AI, as you know, is an incredibly vibrant area with stuff getting funded all over the place. And just staying on top of it is a big part of our work here at the JAIC.

How is the JAIC approaching rapid acquisition? What authorities and contracting vehicles are you trying to take advantage of?

We currently have partnerships with a couple of contracting vehicle organizations, so [General Services Administration], [Defense Information Systems Agency] ... and then [the Defense Innovation Unit] out in Mountain View as well. So we've primarily been using them. There is legislation right in front of us that Congress is working on to grant, or thinking of granting, the JAIC direct acquisition authority. So we obviously are very excited about that. It's not done yet. So when the final vote happens and we do get it, we'll be very pleased and happy. And if we don't get it, well, we'll still be obviously continuing business with the partners that we have. ...

We have an initiative that we talked about at the AI symposium called Project Tradewinds which is a very exciting sort of new way of approaching acquisition that allows smaller companies … to come and work with the JAIC and also the broader DoD. So these are vehicles that we will create that anyone across the DoD should be able to access and get access to … teensy weensy, little companies that normally would hate to work with, or wouldn't know how to work with, the DoD. They can use Project Tradewinds’ acquisition frameworks to be able to interact with us in a very low overhead way. So we're very excited.

We're almost seven months into the COVID-19 pandemic. How has the JAIC adapted in that time? What lessons learned have you gleaned from this period that you can apply in the future?

Organizationally, it was very funny because the Pentagon I think in general … [has a] a meetings-in-person at the Pentagon face-to-face [culture]. I think it's because of the background of the folks plus the nature of the work that we do lends itself very well to confidential and classified work, etc. But obviously when it hit, we couldn't do that anymore. But one thing that really surprised me, I was shocked — and I think everybody else was shocked — how quickly everybody moved to being incredibly productive working virtually. I think our productivity probably actually increased as an organization because I think the digital tools and products that we started using that we really hadn't prioritized as much probably came into use.

And I think that gave everyone a productivity boost. What we did as an organization, you may have heard of Project Salus, which was a product that we built in an incredibly short amount of time to support Northern Command and the National Guard and others on logistics around food shortages and resource shortages. The lesson we learned there was that when it comes to a crisis, the ability for a technology organization to move very quickly to get product in prototype mode ... out the door, in the hands of our services and combatant commands is insanely important. ...

When you look at most of the large successful companies in different verticals out there, whether it be financial services, companies like Goldman Sachs, or you look at transportation, logistics, Federal Express, or UPS, when you look at manufacturing, whether it be people like 3M or others, when you look at great companies and great organizations, at the heart of it, you realize these are software companies that happened to be in their different verticals. They're just great at technology adoption and technology deployment and technology use.

One point that I make is that if we want to be successful as an organization — we the DoD — in the next five, 10, 15, 20 years and stay relevant and be leaders, we need to get great at software. Not only artificial intelligence, but even the more basic level. We need to be great at software and our ability to crank out code. We call this concept-to-code, like [what we did with] Project Salus. That to us was not only exciting because of the functionality we built, but the fact that we realized that as an organization, we could actually get code out the door so quickly and get it into customers' hands. That was a huge win. So my sort of bigger picture thing is that if we the DoD are able to do this in a consistent basis, at scale and open this up for our DoD units to be able to build software functionality that quickly and use it, we will be incredibly successful. That will be the true mark of success for us.

You've often quoted former JAIC director, Retired Lt. Gen. Shanahan, and noted that there is no Department of Electricity, and eventually there will no longer be a dedicated AI organization at the Defense Department because the technology will be a part of everything. How far away is that future?

I am a big believer in the technology commoditization curve as ... I've seen every single piece of technology in my career go from incredible high unit costs down to cents or pennies on the dollar to deliver. And AI I think is susceptible to that. And AI, by the way, in my mind and everybody will agree, is not a single technology. It is a collection of techniques, technologies, algorithms ... etc., to deliver functions called AI, but you take it on a case-by-case basis. So when you think of things like predictive maintenance or image recognition, even natural language processing to a large extent, these are highly commoditized and highly mature technologies, which have been absorbed and are being absorbed into all kinds of form factors. You have little Alexas running around, you can do this stuff on your phone. It's a matter of form factoring this stuff now, as opposed to big breakthroughs in technology. So that's great, that's all scaled. And that is going to go the way of the Department of Electricity. But there's still very big technical problems from a vertical perspective, things like full motion video, image recognition [which is] not a solved problem, especially with different sensor technologies, etc. Or for instance, other types of unique sort of warfighting things or for AI security or reliability. We still have very brittle algorithms in certain areas. So there's tons of work to be done.

Topics: Defense Department