USAF-MIT AI Accelerator: collaboration for new AI solutions

USAF-MIT AI Accelerator: collaboration for new AI solutions

Michael Kanaan on the USAF-MIT AI Accelerator, and its mission to use AI to increase capabilities while addressing societal demands...

Michael Kanaan is Director of Operations, U.S. Air Force and MIT Artificial Intelligence Accelerator, having previously been at the Pentagon as the co-chair of AI for the Air Force.

The USAF-MIT AI Accelerator began in January 2020. “It’s pursuant to a cooperative agreement with MIT, MIT Lincoln Laboratory and the Department of the Air Force,” explains Kanaan. “Our efforts stretch across three main lines. The first is to execute a number of flagship AI projects and the related work to bring that into existence. The second is developing scalable AI education for the workforce - all demographics, all ages, and all ranks. And the last is to lead the dialogue in AI ethics and safety. It’s all about making AI real for our workforce.”

Aside from the three flagship projects which we are covering in depth, the initiatives include such things as natural language processing for communication with machine and foreign language training, swarming unmanned aerial vehicles for deployment on humanitarian aid missions, and using big data to illuminate weather circumstances in areas without a ground station. The projects are linked by a shared focus, as Kanaan explains. “The most important thing is to ensure that we all have a common and shared dialogue and understanding of what AI is, what it isn’t, how it works, and how to walk along that journey.”

The MIT and Air Force collaboration is of a lineage with some of the most illustrious projects in the history of the United States. “There's a triangular relationship between industry, academia, and government in the United States, that's very special and very storied throughout our past.” Kanaan emphasizes that it stems from a common language between government, industry, and academia which must be nurtured. “We have to reinvigorate the relationship that, for instance, brought the internet into our homes. Artificial intelligence is something that's going to be viewed as equivalent to electricity in our lives, because of the way it affects us every single day. What could be more important than something like electricity being shared by the greatest minds, by those who build the best technologies and by the government as representative of its people?”

The work has been enabled by the participation of a number of key partners, whose professional experts and contractors have worked alongside MIT and the USAF. “A lot of work that is necessary to bring modern technologies like cloud to bear, without which you would not have artificial intelligence. We want to make sure that it's as easy as possible for our workforce to grasp.” Kanaan emphasizes that partners run the gamut of sizes, from the smallest to the largest. “This is a team sport. It's a whole-of-nation effort, with small business innovation and research crucial to the success of the United States Air and Space Force. Meanwhile, our traditional partners understand us better than anyone else. They know how to integrate technologies with the legacy architectures that we must rely upon. We can't buy a new thing every single day, and many of those things we can't put in the cockpit of a jet, of course. And then lastly, nontraditional partners help to reinvigorate the conversations that we need to have on AI today.”

The fruits of the labor being put into these projects are not only for the Air Force’s benefit, with wider society also standing to gain. Kanaan cites humanitarian disasters, such as the wildfires and hurricanes which have had a devastating impact on the US this year, as examples of situations that could benefit from its work. “Humanitarian aid is a huge mission of the United States Air Force, as it is of the Army, the Navy, Coast Guard, and so on. AI has a role to play, and that can stretch across swarming drones to using computer vision, to predicting fire lines, to detecting people in flooded areas, and delivering telemedical health.”

In that spirit, public challenges have been established for the two-way sharing of information. “The public challenges will ultimately help develop the associated projects for use in public society. And I think what I'm excited about is our release of some of these public challenges like magnetic navigation using earth’s magnetic sphere, for which you can find the public GitHub repository today.”

The initiatives are standing the Air and Space Forces in good stead for the future by embracing digital transformation. “Once upon a time in the industrial age, you had to make trade-offs between speed, accuracy, and cost,” says Kanaan. “In the digital age, thanks to machine learning, artificial intelligence or any of the number of other automation techniques that are part of digital transformation, you can now do all three at once. For the Department of the Air Force, it’s about being more accurate, delivering better logistics, working on humanitarian aid missions while also saving the taxpayer dollars and making sure that we are good stewards of that money.”

Kanaan views the collaboration that has enabled the accelerator as key to its success. “I can’t emphasize enough how grateful we are to MIT, to academia, to industry for being a part of this conversation and to our airmen and workforce for wanting to have the dialogue. What makes us special is that, while we are certain to make mistakes along the way, we hold a dialogue afterwards. It's all about diving in.”

Maj. David Jacobs, US Air Force: Magnetic Navigation

Having graduated from Stetson University College of law as a patent attorney, Maj. David Jacobs, U.S. Air Force, got to ply his trade while stationed at an Air Force research lab. “I became the only active duty patent attorney, and I’ve enjoyed the opportunity to get where I am today as both the chief legal counsel to work on intellectual property, data rights, contracts, industry, and, of course, ethics, and then also a program manager for the robust neural differential models for navigation and beyond.”


Considering his role, Jacobs emphasises the focus on ethics that pervades the Air Force’s work with AI. We embed in all of our projects a consideration of artificial intelligence ethics and how it's done. The Air Force wants to be a leader in AI, and to do that you have to focus on ethics.” He gives the example of the possibility of human research subjects being implicated from AI research based upon data. “One of the things we did is work with the Air Force 711th Human Performance Wing and coordinated with the Department of Army to develop a simple two-page guide to help Air Force and MIT researchers understand when something is human subject research under AI, and when it is not, so that we're following ethical guidelines at all times.”

In line with this ethical consideration are the public challenges. “The Air Force can actually be a partner in advancing the state of the art for everyone, and be leaders in this field. While it's not new for academia to put forward challenges for other academics, it is new for the Air Force to get involved.” That new approach has required a number of advancements to make possible. “Some of the things that we've had to address include the Air Force sharing data at such a public level. On the legal side, we’ve developed a data sharing agreement which enables the Air Force to share data publicly with academia and industry.”

That collaboration is key to the Magnetic Navigation project, which Jacobs works on, and is aimed at developing an alternative to GPS systems which are vulnerable to disruption, especially in a conflict setting, which would create huge problems in both a military and civilian setting, considering the extent to which the technology is embedded in our lives. “The government is looking into what alternative GPS could look like and magnetic navigation is one of the possibilities,” says Jacobs. “Magnetic navigation takes the earth magnetic resonance and a magnetometer reader to pinpoint where you are in relation to the earth. Because this technique doesn't rely on any external sources, it becomes particularly useful in areas where other signal sources are uncommon, such as over water.”

The role of AI in the project is threefold, as Jacobs explains. “One, we’re using AI to reduce excess noise on the system. Have the AI cancel out what is coming from the plane and recognize what is interference and what are actual readings. Two, determine your position in real time with faster speeds. As we go to other vehicles like an F-16, we're breaking the sound barrier and so we need to determine location at much faster speeds. And then finally we’re combining that magnetic parameter with other systems in the aircraft to create a complete picture.”

The project has attracted a number of interested partners, such as the Joint Artificial intelligence Center and the Department of Defense. “We've had some talks with DARPA, with NASA and of course, academic institutions and industry. We're happy to work with small business, large business and other academic institutions, because the more people that tackle this problem, the better. We're approaching it as a chance to provide something that is good for the community at large.”

TSgt. Armando Cabrera, US Air Force: Synthetic Aperture Radar

“I'm first generation everything,” says TSgt. Armando Cabrera, US Air Force. “First generation American, first in my family to graduate high school, college and first to join the military.” Having graduated with a Bachelor’s in Mechanical Engineering, and after some time struggling to find a job, Cabrera joined the Air Force for Geospatial Intelligence. “I worked really hard in the Air Force technical school and graduated as a distinguished graduate.”


Having demonstrated his potential, Cabrera was eventually selected for a program usually reserved for officers to be sent to Amazon to learn best practices for machine learning. “I was there for a year, playing two kinds of roles. First, I was a student taking all their courses, starting from the fundamentals of mathematics, machine learning, all the way to neural networks. And then also I was building training guides for how to use their equipment and software.”

That background has led him to MIT, where he is now responsible for the Multimodal Vision for Synthetic Aperture Radar project. “What I bring is the operational experience of how to use these types of sensors, so I can field questions.” The goal of the project is to turn the images taken by specialized sensors into more human readable and interpretable photos. “That way you don’t need experience as an image analyst, to understand what the images show. SAR sensors can penetrate things like weather or smoke, but the drawback is it's hard to interpret the image itself. I'm hoping that with this capability that we're creating, it can be used during events that usually don’t deploy it for. We can automatically use the sensor to penetrate through smoke or clouds and easily get additional information that we normally wouldn't have.”

The project uses learning algorithms trained on paired SAR and more easily understood electro-optical (EO) images. “It’s able to learn what a SAR image will look like compared with an EO image, and over time it will pick up the characteristics of the SAR image that are equivalent to EO image. That way, it can create new images with the synthetic EO image attached.” Cabrera points to the usefulness of such technology in response to disasters such as the California wildfires, making previously obscured areas visible to build up a better picture of what is happening on the ground.

Cabrera hails the open nature of machine learning as meaning that it’s now accessible to far more people. “When I first heard I was going to learn machine learning, I didn't know what it was. But AI is so democratized now that I could learn a lot of information just from searching the internet. The two years that I was able to spend with a non-stop focus on machine learning and AI have meant I’ve been able to bring back a lot of benefits.”

Capt. Ronisha Carter, U.S. Air Force: C-17 scheduling

Having enlisted in the Air Force directly out of high school, Capt. Ronisha Carter started off in the field of server maintenance and boundary protection, before becoming an officer and receiving a Master’s in Computer Engineering. “I was selected for an Education with Industry fellowship at VMware, where I was able to work within an Artificial Intelligence Machine Learning development team,” she says. “It was at this time when I developed a foundation in artificial intelligence and machine learning.”


Her current role is as a Cyberspace Warfare Operations officer. “My career field covers the entire communications spectrum,” says Carter. “Everything from network defense to base communications structures, to tactical communications. This background along with my AI foundation led me to be one of 11 selected to collaborate with MIT on the integration of artificial intelligence technology into Air Force platforms.”

Under Carter’s remit falls the C-17 scheduling project, with the intention of bettering the lives of pilots and airmen using AI to make the process of scheduling less time consuming while increasing efficiency and minimizing errors. “Creating an Air Force flight schedule today, the scheduler has to account for a multitude of variables we identify as constraints. This includes qualifications or the training a pilot requires for that seat and crew rest - the time they must take off in between each flight. Also the amount of flights that need to be scheduled, and the time intervals between those flights. This process is currently being accomplished through various manual channels. Separate data systems, phone calls, and even whiteboards, which causes scheduling to be immensely complex and time consuming.”

The remedy to that involves using AI to take up the burden. “What we hope to achieve is to create a data driven model that can produce the best or most optimized schedule for multiple objectives and constraints,” says Carter. “We provide decision-makers with a mathematically aided assessment that predicts schedules weeks in advance and it gives them back time in their day.”

Wider implications for the project involve the gaining of efficiencies across the board, from supply chains to maintenance. “For instance, the work that we’re doing today could allow for advancements in scheduling for hospital staffing, shift workers, cargo and mail distribution, logistics operations, and even commercial airline crew scheduling or flight maintenance.”

Carter emphasizes the extent to which ethical considerations guide everything which is done with AI. “Within all of our projects we are considering the implications of ethics. In February, the DoD adopted ethics principles for AI based on recommendations from the Defense Board of Innovation. This mandates that all DoD AI capabilities must be responsible, equitable, traceable, reliable, and governable and meet the same legal, ethical, and policy standards across the department.

Partnerships have again made the project possible. “Our partnership with MIT and Lincoln Lab is essential to developing these technologies, and we also work hand-in-hand with Tron and Airmen Coders, Air Force Software development teams that vector internal Air Force talent to solve and engineer solutions for the really tough Air Force problems. Our team of MIT principal investigators, grad students, software developers, human-computer interaction designers, and Air Force software development teams ensures we create better solutions for our Airmen.”