Data is a commodity that fuels the technological world. Data powers machine learning, drives advances in artificial intelligence and helps corporate and government leaders make better, more informed decisions.
One of the largest owners of data, the Department of Defense, known for its steadfast protective oversight of this digital gold, embarked on a historical first this year: a first with the ability to improve the effectiveness, agility, affordability and speed of operations for a myriad of stakeholders. The unprecedented idea started at another new DoD venture located in the heart of academia.
Members of the USAF-MIT Artificial Intelligence Accelerator, a new unit under the Department of the Air Force, located at the Massachusetts Institute of Technology, tackled some of the most difficult challenges facing our nation and the Air and Space Forces.
The AI Accelerator, a small multi-disciplinary team of Airmen, scientists, researchers engineers and students, works towards creating functional AI systems that can be rapidly adopted by the DOD and society.
A functional AI system requires large amounts of data. Making eligible Air Force datasets publicly accessible is an important aspect of researching and developing any AI system. Offering this accessibility was an undertaking that the Airmen at the AI Accelerator deemed necessary.
“By making data more accessible and available, we are able to bring the smartest minds to bear on AI challenges that directly impact the Department of the Air Force and the United States,” said Maj. David F. Jacobs, chief legal advisor to the AI Accelerator. “One of the first problems we had to tackle at the AI Accelerator was how to make eligible Air Force datasets publicly accessible so that researchers at MIT and other academic institutions across the United States could work with real data to tackle real problems.”
This problem prompted Jacobs to develop a first-of-its-kind Air Force data sharing agreement for research and collaboration. Built upon other government, industry and academic data sharing agreements, this paved the road for terms and conditions under which the Air Force can share data to the broader AI community and contribute to advancements in the field.
“Data rights and intellectual property rights are extremely important considerations when developing any AI technology,” Jacobs said, who works closely with Air Force Judge Advocate General’s Corps, Secretary of the Air Force General Counsel and the Air Force Research Laboratory Legal Counsel on legal issues involving intellectual property, data rights, contracting, cyber law, human subject research and ethics.
Accessing large amounts of Air Force data in a centralized and secure way is another challenge for the AI Accelerator.
By partnering with the Air Force Chief Data Office, the AI Accelerator uses the Visible, Accessible, Understandable, Linked and Trustworthy data platform, or VAULT, as a means to provide a centralized, secure and accessible place for hosting both public and controlled Air Force data sets.
“The VAULT is an enterprise level collaboration environment that resides on both the high side and low side, the NIPR and SIPR net,” said Derek J. Eichin, lead research analyst at the Air Force Chief Data Office. “The VAULT provides multiple tenants, secure spaces or ‘sandboxes’ for various tenants to bring in their data, use our enterprise tools, whether they be open source or licensed software, and store and curate data driven analytics within that architecture.”
“The VAULT is used as a catalyst for helping many organizations in the Air Force realize the potential technology and capability that they have,” said Lt. Col. Ronnie J Synakowski, director of data capabilities at the Air Force Chief Data Office. “The VAULT is available to them to apply to whatever their current job is.”
The AI Accelerator’s partnership with the Air Force Chief Data Office is a catalyst for change in the way data is shared and a huge win for DOD AI research.
“The relationship with MIT and the AI Accelerator has been really great,” Eichin said. “We are partnering overall in an effort to make data accessible to not only internal partners within the department, but also universities and small businesses.”
The AI Accelerator and the Chief Data Office collaborated in addressing AI problems relating to scheduling using AI by putting forth Air Force datasets to a large DOD AI community during their inaugural datathon.
The second challenge problem addressing an alternative to GPS through magnetic navigation was presented at the international conference JuliaCon, a conference dedicated to the Julia programming language.
“At the conference, we presented the challenge problem,” Jacobs said. “Here’s what we’ve done so far. Here is some very valuable Air Force data, have at it and see what we get.”
In the AI community, winning a challenge problem is for bragging rights. It is for smart minds who want to be leaders in the field and advance technology for everyone. The C-17 Globemaster III scheduling challenge was the first of its kind in the Air Force and brought together more than 50 Airmen and DOD civilians to tackle the problems given.
The magnetic navigation challenge reached outside traditional Air Force channels to embrace a broader group of academic minds. The team from Arizona State University put forth a very solid submission. Some of their proposed solutions and results won the AI Accelerator’s challenge, and showed Air Force leaders that sharing their data can bring about faster, more innovative results.
“Data is now recognized as a strategic asset to the Department of Defense,” Jacobs said. “Datasets for AI training and algorithmic models will increasingly become the DOD’s most valuable use of this strategic asset. The Department of the Air Force is setting the groundwork for smartly sharing data to train AI for immediate and lasting military advantages.”