The National Geospatial-Intelligence Agency is harnessing the power of artificial intelligence and machine learning to overcome the challenge of taking in, processing and managing large amounts of data in a short amount of time.
One way NGA is using AI and machine learning is to detect and decipher satellite imagery while actively developing new ways to manage the data these tools make available. In this mission, the people who use AI and machine learning are just as important to the process when it comes to the challenges of bias in the collected data, managing drift in the output, and maintaining security while advancing technology.
“I think that one of the big goals, as well as exciting opportunities about AI is helping people do their jobs better and faster, and even day-to-day things like supporting individuals who code and using large language models to help them create code. It’s not that these models are going to replace us writing the computer code, but they can be a tool,” said Natasha Krell, a computer vision, machine learning, research and development scientist at the National Geospatial-Intelligence Agency, on Federal Insights – AI/ML. “I think one way that somebody put it recently, it’s almost like a word document processor. The word document is not going to write the text for you. You still have to input it, but it’s a tool that helps you complete your job.”
NGA is leaning on these emerging tools to find that proverbial “needle in the haystack” when analyzing imagery.
“It’s really helpful for detecting those objects and then also classifying what they may be and considering the deluge or large amounts of data coming from satellites, AI is really one tool to process all that imagery and again insights from it,” Krell told the Federal Drive with Tom Temin. “We look at both on premise as well as cloud opportunities for storage, infrastructure and compute.”
NGA classifies it’s large multimodal models for imagery into several different types including electro optical, synthetic aperture radar (SAR) and thermal. Bringing all of those sets together is what Krell called “the next stage, and really the cutting edge of what’s happening within AI.”
Krell said NGA and its partners are using AI and ML to impact the humanitarian assistance disaster relief mission space.
“We do have certain AI and machine learning initiatives that we’re able to share publicly and work with other government partners in different organizations, ” she said. “We’re certainly sharing output models, data, insights. There’s definitely a good track record of that.”
Like most agencies, NGA faces its share of challenges in bringing AI and ML into the fold.
She said training the workforce, managing the volume and velocity of data and protecting against bias as just a few of the top concerns.
“AI, in general, is very data hungry for data management systems and processes are really important. There’s also kind of interesting developments and I would even say paradigm shifts happening with these foundation models. These are large pre-trained models that are the starting points for natural language processing (NPL), and you basically fine tune it on a specific use case. . .the vantage point of taking in a large already pre-trained model and then fine tuning it on a smaller dataset,” Krell said. “Bias can come in a lot of places, whether it’s from the data sets, the models, and then the algorithms themselves. So, it’s definitely something to be aware of and something that we have to balance at the end of the day.”
Additionally, NGA must be aware of the security risks with these tools.
“You can’t just take any machine learning or algorithm off the shelf. You really need to do rigorous verification and validation on those models and data that you’re bringing in,” she said.
NGA is relying on guidance and documentation from the chief data and AI officer (CDAO), the Department of Air Force chief data and AI office as well as the Joint Artificial Intelligence Center to ensure AI and ML tools are used appropriately.
Krell said while NGA is quite aware of the limits and challenges of AI/ML, the growth in use of tools like ChatGPT has helped the general public and its workforce gain a better understanding of the technology.
“Natural language processing is gaining a lot of traction and attention out in academia and industry… And there’s some really interesting advancements happening in industry and academia, especially in the space within large multimodal models. So this is where you’re using both text and imagery to perhaps query imagery and gain more insight from the machine learning generated output,” she said.