AI at NASA

(Excerpt from Houston We Have a Podcast, Episode 424: AI at NASA)

Nilufar Ramji: Houston We Have a Podcast. Welcome to the official podcast of the NASA Johnson Space Center, Episode 424: AI at NASA. I'm Nilufar Ramji, and I'll be your host today. On this podcast we bring in the experts: scientists, engineers, and astronauts, all to let you know what's going on in the world of human spaceflight and more.

Artificial Intelligence, or AI, continues to be a hot topic across several industries and communities of practice. As humanity continues to innovate, the technology follows. And for decades, and even centuries, that evolution has brought us to the next big leap — computing systems that can perform complex tasks normally done by humans, as it relates to human reasoning, decision making, and creation of content. At NASA, AI helps support missions and research projects, analyzes data to reveal trends and patterns, and develop systems capable of supporting spacecraft autonomously.

With safety and security as a top priority, NASA researchers and scientists have been leveraging artificial intelligence for decades. Most recently, they are looking at emerging AI and how it can best serve missions. What this could mean is sifting through satellite imagery, developing technology for autonomous vehicles, and searching for planets outside of our solar system using our deep space telescopes to name some examples.

The question remains though, how can NASA continue to adopt these emerging methodologies and work with industry to ensure that advancement continues safely and securely? Joining me today, we have Kevin Murphy, NASA's Acting Chief Data and AI Officer to tell us about what that landscape looks like. Let's get started.

<Intro Music>

Nilufar Ramji: Kevin, thank you so much for joining us on Houston We Have a Podcast. Before we go ahead and get started, can you tell us a little bit about yourself and what you do here at NASA?

Kevin Murphy: Sure. Thanks. Really happy to be here and on the podcast. I have a couple of jobs currently, just like a lot of people across NASA, and really happy to do all of them. It's really, really interesting work. My primary job right now is, I'm the Chief Science Data Officer for the Science Mission Directorate, and that job allows me to look across all of the wonderful scientific information that we have and try to use advanced techniques to understand that gold mine of scientific information. I'm also the Acting Chief AI officer and the Chief Data Officer for the agency, which has given me a really interesting perspective on not only how we use modern data science and AI techniques for our scientific data, but how that applies across engineering, how it applies the business processes, and how that applies just generally for our mission for discovery.

Nilufar Ramji: So if you could break it down for us, in your own words, can you explain artificial intelligence to us and then take it a step further by explaining why it's important for us to understand this as an agency.

Kevin Murphy: Sure, artificial intelligence is typically used as a pretty broad term. It means a lot of different things, depending on who you're talking to, but at its basis, artificial intelligence is software that learns patterns from examples instead of following rules written by hand, like you don't have expert logic, right? You feed it a lot of inputs. You give it text or images or sensor data, and it builds a mathematical model of what those patterns look like, so it can recognize or generate something that fits those models, right? So the intelligence isn't reasoning or understanding. It's a very sophisticated pattern of machine learning, things at scale. And there are a number of different types of artificial intelligence, because people are kind of experiencing it today in one realm, which is basically through large language models. But if you go back to the 80s and 90s, we had expert systems where you hand coded all of the rules by hand. And let's say the machine or software went outside of those controls, it would sometimes misbehave or not do what you expect it to do. We kind of transitioned into machine learning models, deep learning models and finally, we're kind of using these things called transformer architectures, which really power large language models today.

So you know, we need to use those techniques, because NASA generates enormous amount of information, but not only that, we have some of the hardest problems in the world to solve, right? How do we go explore a distant planet, how do we develop systems to be resilient on the moon, keep astronauts safe — we need to take a lot of that information together. We need to look for patterns that would be difficult to do through teams of people, even, and use that information to make decisions more quickly and more comprehensively.