Experts discuss training data, GEOINT assurance, and more at USGIF’s Machine Learning & Artificial Intelligence Workshop
USGIF’s Machine Learning and Artificial Intelligence Workshop in November attracted more than 300 GEOINT professionals—one of the Foundation’s largest workshops to date.
Interest in the workshop, which was hosted by USGIF’s Machine Learning and Artificial Intelligence Working Group as part of the Foundation’s 2018 GEOINT Community Week, has grown each year, demonstrating steadily increasing demand for discussion and answers surrounding AI technology.
The two-day workshop was held at the National Geospatial-Intelligence Agency (NGA) in Springfield, Va., and included both an unclassified and classified agenda.
The Chinese “AI Security Dilemma”
Elsa Kania, an adjunct fellow in the Technology and National Security program at the Center for a New American Society, gave the first keynote of the unclassified day with an accompanying presentation titled, “China’s AI Revolution.”
Kania described the notion of an AI “arms race” or “Cold War” as hyperbolic, instead opting to use the term “AI security dilemma.”
“Beyond and despite the hype, China’s quest to lead in AI could change the balance of power,” she said. “China is catching up.”
The Chinese government is harnessing the country’s tech companies in a “national team” Kania described as a unique, and morally questionable, “military-civil fusion.” The country has set ambitious goals to achieve a series of landmark AI products by 2020 and to be the world’s premier AI innovation center by 2030.
While the United States still outpaces China with its number of AI patents and experts, other metrics demonstrate China is eclipsing the U.S.—in 2017, Chinese AI startups received more funding than their U.S. counterparts.
“If data is the new oil, China is the new Saudi Arabia,” Kania said.
Preparing and Simulating Training Data
The unclassified programming also featured two panel discussions on data—one on preparing training data for AI and machine learning systems, and another on simulating data to train artificial neural networks.
Dr. Shay Strong, director of data science and machine learning at EagleView, discussed the tradeoffs between data quantity and data quality—the former being necessary to scale and the latter being necessary to do so accurately.
Kate Werling, a senior solutions architect with Amazon Web Services, echoed Strong’s sentiments, saying quality data equals a quality model—in other words, “garbage in, garbage out.” She emphasized the importance of patience when building a quality dataset.
Werling also pointed to the need for specificity when instructing team members how to annotate data—for example, whether they are labeling cars or vehicles, or how tight the bounding boxes should be around the cars
Panelists who discussed the notion of simulating training data acknowledged the concept is nascent. According to the experts, the market for synthetic data sets is still immature, and though it is attracting some venture capital interest, no major companies have emerged yet in this niche.
Peter Bronez, a senior member of the technical staff at In-Q-Tel, said the motivation for developing simulated training data is that not everyone can gain access to big data, for financial reasons or otherwise. He added demand for simulated training data is likely to rise in correlation with the advent of autonomous vehicles—an industry facing challenges collecting adequate training data due to physical restrictions.
Synthetic training data also has the potential to help AI and machine learning experts solve their data diversity challenges, according to panelists.
NGA & AAA
The afternoon unclassified agenda featured two NGA keynotes followed by a special edition of USGIF’s GEOINTeraction Tuesday networking event in which remarks came from David Gauthier, director of NGA’s new Commercial and Business Operations Group.
Dr. Steven Hall, NGA’s mission leader for AI, Automation, and Augmentation (AAA), gave the first keynote. Hall clarified that automation does not always equal AI or machine learning and highlighted the importance of applying these new solutions to augment human analysts.
Hall said his team is pursuing goals brought to them by analysts who want to improve their workflows.
“We’re putting analysts in the driver’s seat and they’re telling us what they need in order to do their jobs better,” Hall said. “In the last two to three months we’ve seen more adoption and feedback than in the last two years.”
Benjamin Tuttle, NGA’s director of Outposts, said the agency’s outpost in Silicon Valley is also working with analysts to define mission needs early in the research and development process; a machine learning ground order of battle prototype the group recently developed will be tested for the next six months on analyst desktops.
Tuttle shared that NGA is working to set up a second outpost location in Austin, Texas, and said the agency is focused on attracting new talent and partners in areas of interest such as AI, commercial technology, analytic modernization, automated analysis, and advanced security.
He added GEOINT assurance has recently become an important issue for NGA: “How do we track provenance as all of this information becomes available?”
Gauthier also pointed to the need to secure the AI supply chain and study data provenance during his GEOINTeraction Tuesday remarks, saying industry can help the agency “assess commercial services flooding the market.”
“AI is changing our lives right now,” Gauthier said. “The AI thing is something we have to solve, not just for GEOINT but for all of society.”
Headline Image: Experts discuss machine learning and artificial intelligence tradecraft at NGA Campus East. Photo courtesy of NGA.
The overlapping threats presented by climate change, including instability both internationally and domestically, are a new focal point for federal, nonprofit, and private entities. While technology rapidly advances, bringing about innovative possibilities, the reality remains that these issues require thoughtful, collective action, considering both short-term solutions and long-term sustainability.
The evolution of analytic modeling and the expansion of its purpose to capture mission knowledge relating evidence and indicators to answers to key intelligence questions