While quite busy looking to the skies, NASA is also kept occupied looking to the future. And, as more and more data is collected, the agency is working to overcome barriers that once blocked it from fully utilizing innovations in artificial intelligence and machine-learning technology.
NASA has previously used AI in human spaceflight, scientific analysis and autonomous systems. It currently has multiple programs that use AI/ML: CIMON, which is “Alexa for space”; ExMC, medical care AI assistance; ASE, autonomy for scheduling in space; Multi-temporal Anomaly Detection for SAR Earth Observations; FDL, partnership between industry and NASA through AIMs; and robots, and rovers.
In a June 26 presentation at the AI World Government Conference in Washington, D.C., Brian Thomas, a NASA agency data scientist and program manager for Open Innovation, spoke about how to get the best results for AI/ML while considering important policy and culture implications at the agency.
Machine learning has been with us for 60 years, Thomas said, “so really the question is, why haven’t we been using this all along?”
“We’ve already seen the value in these technologies," Thomas said. "They are enabling NASA’s mission now. The problem is that we’re being held back, believe it or not, and so how can we do better.”
Whereas NASA was once held back by insufficient means to collect and process data, "we are now getting to the point where we have collected enough data that data is accessible ... and having enough of it to make that [AI/ML] happen,” Thomas said.
The agency still faces a challenge when it comes to processing power of these volumes of data, however, which is crucial to being able to train your solution.
“Many of these things can not be trained on your laptop. We’re not at that point anymore. We moved away from that probably a decade ago."
After addressing how to handle the data, the next question to answer is why.
“What is our ultimate goal here? The goal is to apply value, right? We’re not here to do science fiction. We’re not here to say, ‘oh wow, that’s really cool.’ We’re here to apply this to NASA’s mission and make things actually happen.”
In this regard, Thomas said the four categories where NASA is currently looking to utilize AI/NL are: aeronautics, operations, human capital and IT support.
When it comes to aeronautics, the belief is using AI/ML could allow NASA to build better vessels and equipment.
“The last thing we want is to have that tested wing or model blow up in a wind tunnel,” Thomas said. "So, can we model a very complex engineering structure to understand when we are getting really close, because we can’t do it now with the traditional model and technique.”
In terms of operations, NASA could use AI/ML technologies to assist in antenna positioning to make maximum contact with satellites, Thomas said.
Workforce uses of AI/ML are a top concern for most agencies, and NASA hopes to use these new technologies to learn more about who is working for them.
“Can we look at our position descriptions and analyze those using text analytics and try to classify that out and then make predictions about what we need to do or where our workforce is at,” Thomas said.
NASA is also looking to AI/ML to help understand network intrusions and how to avoid them, Thomas said.
As it moves forward with AI/ML innovations, Thomas said, NASA is looking at the possible implications of these new technologies.
“Machine learning, every solution we create is bespoke, particularity in supervised learning; it’s all tied to the data."
Kelsey Reichmann is a general assignment editorial fellow supporting Defense News, Fifth Domain, C4ISRNET and Federal Times. She attended California State University.
