Technology, Innovation, and Modern War – Class 7 – Jack Shanahan

We just held our seventh session of our new national security class Technology, Innovation and Modern WarJoe FelterRaj Shah and I designed the class to examine the new military systems, operational concepts and doctrines that will emerge from 21st century technologies – Space, Cyber, AI & Machine Learning and Autonomy.

Today’s topic was Military Applications of Artificial Intelligence.

Catch up with the class by reading our summaries of the previous 6 classes here.

If you can’t see the slides above, click here.

Our guest speaker was General Jack Shanahan LTG (ret), former Director of the Joint Artificial Intelligence Center (JAIC).

Some of the readings for this class session included: Can The Pentagon Win The AI Arms Race?, The Ethical Upside To Artificial Intelligence, The Coming Revolution In Intelligence Affairs, Artificial Intelligence For Medical Evacuation In Great-Power Conflict, Congressional Research Service, Artificial Intelligence And National Security.

AI and The Department of Defense
As a lead up to this class session we’ve been talking about the impact of new technologies on the DOD. AI is constantly mentioned as a potential gamechanger for defense.

General Shanahan founded the Joint Artificial Intelligence Center (the JAIC) to insert AI across the entire Department of Defense. The goal is to use AI to solve large and complex problem sets that span multiple services; then, ensure that all of the DOD has real-time access to libraries of data sets and tools. A key part of the strategy was to work with commercial companies to help build these solutions.

Prior to the JAIC General Shanahan ran Project Maven, an unintentional dry-run for how the DOD and commercial companies could partner (or not) to build AI-enabled apps. Maven partnered with Google to build a computer vision tool for imagery analysts to automatically detect objects/targets. The relationship ended when Google employees forced the company’s withdraw­­­ from the project.

There were lots of lessons on both sides — about transparency, about why companies in the 20th century had “federal systems divisions” that worked exclusively on government projects, while the rest of their company pursued commercial business — as well as a lot of relearning lessons the Valley had forgotten (see The Secret History of Silicon Valley).

This entire class session was a talk by General Shanahan and Q&A with the students. It’s interesting to note how many of his observations echo ones Chris Brose and Will Roper made in the previous sessions.

I’ve extracted and paraphrased a few of his key insights below, but there are many others throughout this substantive discussion. I urge you to read the entire transcript and watch the video.

Tactical Urgency and Strategic Patience
We have a relatively small window to transform our respective Defense Department from Industrial Age hardware-centric organizations, to Information Age software-centric, more risk tolerant ones. That demands the right combination of tactical urgency and strategic patience, but there’s no question we have to move with alacrity on this right now.

War is a very uncertain endeavor run by humans. If we find ourselves on the verge of a conflict with China in 20 years, the idea that we might have a fully AI-enabled force by then will not by itself guarantee victory. If on the other hand, in that same 20-year period, China has a fully AI-enabled force and we do not, I believe we will incur an unacceptably high risk of defeat.

Project Maven – Start With the Problem – Automate Imagery Analysis
We didn’t start Project Maven with an AI solution. We started with a problem: far too much information coming in from the intelligence enterprise; intelligence, surveillance, reconnaissance. It was done manually — intensive, mind-numbing work with analysts staring at video screens for 12 hours at a shift. They couldn’t ever get through that much. It was really hard for analysts to absorb that amount of data and the data was increasing – in volume, speed and in tempo — and it was coming from all sources simultaneously, from unclassified to the highest classification levels.

We weren’t looking to do a 1x solution, a 5x solution or even a 10X solution. We wanted a 100X solution. We needed something that would really change the way we did processing exploitation and dissemination of all this intelligence coming in from all these platforms and centers in the world.

We went out and solicited everything we could find in the Department of Defense. There was tremendous AI work going on in the research labs, but nothing available for doing processing exploitation and dissemination in the near term. The problem was, while the military research labs were doing some of the best research in the world, it wasn’t getting across the classic technology valley of death. (I like to talk about Maven and the JAIC as “AI Now,” and then places like DARPA and the military research labs as “AI Next.”)

Where else did this AI technology exist? DIUx, In-Q-tel and others pointed out to us, “You lament about how much information you’re having to take in and process. Look at how much information YouTube takes in every single day. Your problems are not an overwhelming problem. There are solutions available in commercial industry.”

So how do you take technology from commercial industry, adapt it for the Department Defense purposes, and do it fast enough and do it at scale across the entire Department of Defense? The only way we could do that was to start bringing in those commercial technologies faster and faster.

So, Project Maven started us on the path of bringing an AI-enabled solution for the purpose of intelligence. Largely computer vision in going after the intake from drones. At first tactical drones, then mid-altitude drones, eventually to high altitude manned airplanes and even commercial imagery, working with the National Geospatial Intelligence Agency.

The JAIC – Go at Scale and Speed
After a couple of years of doing Project Maven, Bob Work (then-deputy Secretary of Defense) was itching to move just beyond intelligence. He said Maven was never designed to be only about the intelligence enterprise, we needed to bring in every single mission in the Department of Defense, and we needed to go much faster. There was a revolution happening in commercial industry and the DOD was not was not getting it.

I was chartered to stand up the JAIC in a form that Joe and Raj, and I’m sure Steve would appreciate. Somebody signed a memo saying, “Go stand up a joint AI Center. You don’t have any people, you don’t have any money, you don’t have anywhere to live, but you’ll figure it out.”

Lean, MVPs and the DOD
Project Maven became the basis for what we did in the JAIC. We started with a cross-functional team. In commercial tech you call that an integrated product team, It’s acquisition with software engineering with UI/UX. This idea of U/UX user interface/user experience, a ruthless focus on user experience, the DOD has never been very good at that.

It’s also learning how to put things on contract fast and get things done and field them quickly. It was whatever you can do to get away from the classic DOD stovepipe way of doing business to a much more agile software approach.

Josh Marcuse and the Defense Innovation Board were instrumental in helping us see what those agile principles were and how we could get moving and go fast. And we took the classic commercial tech approach of building minimal viable products: Get something that will work good enough. Let the user know what they’re going to get as opposed to just forcing something down their throat claiming that it was much better than it really was.

Minimal Viable Products
The idea of minimal viable product is not what the department of defense has lived with, it’s out there in pockets. But we need to be much better at that.

What’s “good enough” when we field these capabilities? It was never for us to say. It was for the users of the capabilities to tell us.  We would give some parameters like, “This is a 95% test and evaluation solution. Do you accept it?” “I don’t know. Let me try it out.”

The other reason that we wanted to push AI-enabled capabilities out to the field as quickly as possible is because until users get to play around with it, it’s just science fiction. You hear all these grandiose stories about what AI is. But few people are saying what AI is not. And in the Department Defense it’s more about what it’s not than what it is. There’s so much more we must do to get people to use those capabilities. As soon as they touch it, they say, “Well, if it can do that, what about this?” Every time they’ve said that, we said yes, it can be it can be done. So we’ll make this product better and better in a rapid sort of agile approach of continuous integration, continuous delivery.

That is the future for the Department of Defense. If we don’t get that right, we’re doomed because AI capabilities left to themselves six months down the road will become useless. Just like any commercial software, it’s got to be a continuous development cycle. So that idea of MVP, putting capabilities in the user’s hand, letting them tear it apart, tell us what worked, what didn’t work, what they’d like to see better. That was really the core concept about Maven, and then the JAIC.

The idea of Ops and Intel working closer together, those two worlds merging, is the holy grail idea of ops/Intel fusion. These AI-enabled capabilities are getting us closer and closer to that environment. And this idea of user-defined operating picture is with us today. The performance of that will increase exponentially over the next couple of years.

Continuous Integration, Continuous Development, Continuous Delivery
The other thing about AI is that you train it against one set of data that will reflect, to some extent, the real world. But once you put it in the real world you learn it never works as advertised. The first time you use it in Afghanistan you realize you never trained it against data that had women wearing full-length black burkas, it didn’t know what those were. Interesting little problem. So you get real-world data, feed it back into the algorithm and it performs better and better. A second, third time, so on, and so on. It was all about the user defining what success look like.

You need a cycle of continuous integration, continuous development and continuous delivery.

We were proud at Project Maven that the first updates of our initial algorithms were out the door in four to five months and then got better and better and better — in some cases about every two weeks. In our personal lives, we get updates pushed to us hourly in some of our apps. So that’s the goal. To be able to do that, you need this backend infrastructure and architecture, so that a soldier, sailor, airman, Marine, Space Force wherever they are, can design an app on the spot, relying on this backend infrastructure, that Joint Common Foundation.

We’re not close to doing that yet with some very limited exceptions like Kessel Run, which is not AI, although they’re starting to go down that path, too. But they’re getting into that model of, “How quickly can I get this in an agile sort of software pathway of doing things and push it out to the field as quickly as possible?”

Warfighters Need to Be Demanding Customers
The biggest role that the warfighters have in this is to be very demanding customers and to say what needs to be fixed and not just accept what we’ve done to operators too many times, which is, “Look, here’s the product, take it or leave it. We’re going to get you another version five years from now, and probably $200 million over budget.”

That world is gone. We have got to get to that point where, “Here it is. What’s wrong with it? We’ll fix it as quickly as we can and we’ll update it with real-world data, it will make it better and better.” Demanding customers provide ruthlessly candid feedback, which was never a problem with the Special Operations community.

This concept of leverage and UI/UX, and all of that is so essential to everything we’re doing. And the backend requires things in addition to a common foundation, a data management platform, open API, is all the T&E tools, everything should be available to everybody in the Department of Defense.

You Need a Disrupter
And then of course you need a classic disrupter. Somebody who does not take no for an answer, who breaks a little glass and makes some people a little bit upset in terms of overturning their apple carts. Because there are so many obstacles in the Department of Defense it would be easy to become disillusioned and just give up on the whole enterprise. The role I needed to play was top cover for the disrupter. It’s the combination of this top-down advocacy and pressure, and this bottom up innovation. And then you bring in a disrupter that gets this thing going.

This is what Raj Shah did at DIUX. It’s what AFWERX has to deal with. It’s what Hondo Geurts had to deal with when he was running SOFWERX. It’s what Enrique Oti has done up at Kessel Run, Chris Lynch at the Defense Digital Service. These were classic disruptors — strong personalities running innovation organizations in innovative ways.

Scaling The Organizations Across the DOD
But now you had to take all those models and begin to learn, how do you scale them across the Department of Defense? That is incredibly hard. And that’s what the JAIC was designed to do, you spark the movement for the next decade, 15, 20 years of movement.

But unless you take those organizational models and begin to scale them across the Department of Defense, they’re sitting there as one-off organizations. They’re critical but they’re insufficient. You need to inculcate a startup culture in the institutional bureaucracy of the Department of Defense.

One of the most important things we were working on in the JAIC is this thing called the Joint Common Foundation.

Joint Common Foundation
It is an architecture and infrastructure — for lack of a cleaner term, call it a platform as a service, a DevSecOps or AI/Ops environment. On top of an Enterprise Cloud environment that’s supposed to be called JEDI. To build this dev SEC ops, AI ops environment, which is to give everybody in the Department of Defense equal access to everything from data, to AI tools, to all the security environment, to Jupyter notebooks, you name it, everything you would need to develop AI. The JAIC is building that as a common foundation.

There are 200 people in the JAIC, and that includes contractors, it’s probably not going change the entire Department of Defense and their AI way of doing business.  But what it has to do is leverage more. The idea of everything that JAIC does is a product available to everybody else in the Department of Defense, they come, pull it off the shelf, so to speak, and use that to leverage the entire Department of Defense. Otherwise JAIC will be an interesting organization and last for a few years and go away.

The Joint Common Foundation is designed to make that available to everybody across the Department of Defense. And I think over the next few years that Joint Common Foundation will be what the JAIC becomes known for as much or more than anything else it’s doing. The services will figure out the technology piece, it’s just they need a little bit of a push, that flywheel has to get turning a little bit faster.

New Operating Concepts Versus Doctrine
Even more important than this, is the idea of developing new operating concepts. New operating concepts do not come out of the Pentagon. Pentagon writes great doctrine. Doctrine is all sorts of instructions and directives and great PowerPoint slides; the operating concepts will come from those on the tactical edge or the operational environment.

And what does the world of AI, Enterprise Cloud, 5G, and someday quantum look like? I don’t know, but it’s not for us to say, it’s for the users to be able to figure it out, by letting them try it out in operational settings.  I believe there is no mission in the Department of Defense that will not benefit – from the introduction of AI-enabled capabilities from the back office, to the battlefield, from undersea to outer space, in cyberspace and all points in between.

Read the entire transcript of General Shanahan’s talk here and watch the video below.

If you can’t see the video of General Shanahan’s talk click here

Lessons Learned

  • The DOD recognized that AI was one of the potential game changers
    • They set up the JAIC (Joint Artificial Intelligence Center) to see if they could Leverage AI across the DOD
  • They built products using modern processes – continuous integration, continuous development and continuous delivery
  • The group required a “Disrupter,” someone willing to break glass to push these ideas
  • Over time, JAIC has realized that scaling AI across the DOD will not come from delivering individual solutions
    • but by having a Joint Common AI Foundation that’s accessible by all DOD app developers

One Response

  1. Great read Steve — while I agree with all points, I would have loved to see the cons of AI and how it could negatively impact the future as well. Just to get a balanced view of both sides of the coin.

Leave a Reply

%d bloggers like this: