David Salvagnini, Chief Artificial Intelligence Officer at NASA, spoke to The Innovation Platform Editor Georgie Purcell to discuss how AI technologies are shaping the trajectory of NASA’s work.
For decades, NASA has been using artificial intelligence (AI) technology to support and optimise its work across the agency, both on Earth and in space. NASA uses AI to help plan and schedule missions for planetary rovers, analyse satellite datasets, diagnose and detect anomalies, develop autonomous systems, and much more.
Some of the AI tools used at NASA involve machine learning, which uses data and algorithms to train computers on making classifications, formulating predictions, and uncovering similarities or trends across large datasets. Utilising AI tools has a range of game-changing benefits for NASA’s work, including streamlining decision-making, saving resources, and enhancing workforce efficiency.
To find out more about NASA’s AI activities and what the future holds for such technologies at the agency, Georgie Purcell spoke to David Salvagnini, Chief Artificial Intelligence Officer at NASA.
What role do AI technologies play in NASA’s work?
NASA has been very involved in the use of artificial intelligence and machine learning in several ways, particularly for more traditional means. Predominantly, these technologies have been used in our science work to assist in the discovery of objects in our solar system or distant solar systems. One example is a capability called ExoMiner, which uses AI trained by machine learning to identify exoplanets and even planets and distant solar systems. Interestingly, this used data collected as far back as 15 years ago and, through the AI models, was able to identify objects that had not previously been discovered.
Another area where the use of AI technologies is prevalent is autonomy. For example, the Mars Perseverance Rover must safely navigate the surface of Mars and avoid obstacles or hazards whilst dealing with the potential risk of a communication delay between Mars and Earth. Much like you may see in autonomous vehicles on the market today, an on-board AI system helps the Rover to process the environment around it and render decisions as to how the vehicle should manoeuvre. The autonomous nature of the system removes the issue of a propagation delay if the rover had to communicate with a controller on Earth. This technology has been heavily tested as part of the systems engineering process here at NASA, mitigating and derisking the use of AI in this case.
In collaboration with IBM, we recently released an AI Prithvi-weather-climate foundational model for a variety of weather and climate use cases. The model offers a flexible and scalable way to address a myriad of challenges related to short-term weather and long-term climate projection. The resulting data from the model is published openly, so anyone can access and use it. A lot of peer review and collaboration is involved in this type of work.
The use of AI in space is very exclusive. With traditional AI that a company may use, a large-scale compute backend, like a Cloud provider, is the processing engine for much of that work. In space, we do not have access to a Cloud platform. When we think about the future of AI and how we can continue to enable AI on space-based systems operating in extreme environments, we must account for several different factors. Such things include the extreme environment itself, the radiation that the electronics are exposed to, power budget, and compute limitations. Unlike here on Earth, it is not as simple as buying more power when it comes to a space-based vehicle – there are tight budgets and weight limitations to honour. There are many unique elements to how NASA will use AI, especially in support of a space-based mission, that create particularly complex challenges for us. However, NASA has a long history of overcoming such challenges.
What potential does AI have for the future of NASA’s missions?
One area where AI will play a key role is in helping to make things more adaptive and autonomous. In general aircraft today, for example, there is a crew but there are also autopilot systems that control the aeroplane for most of the flight. This technology is very deterministic – if the speed changes, a change will be made to the control. Similarly, NASA is currently carrying out work to be able to more dynamically adapt to different environments – whether it be in orbit, or on the surface. We are working to gain a much higher degree of situational awareness of an environment, which can then be fed to an autonomous system. The autonomous system can much more reliably carry out a response to a set of conditions that are more dynamic.
Orbital debris is a huge problem for NASA and is of rising concern as space activity continues to increase at a rapid pace. We recently released a Space Sustainability Strategy to measure and assess space sustainability for Earth, Earth orbit, the cislunar space, and deep space. AI will play a significant role in this, not only for the detection of orbital debris but also in some of the actions that may be taken by systems for remediation.
There is also the use of AI in adaptive communication systems. We often think of communication like a phone as a single mode. However, the space environment presents a much more complex need where multiple communication networks are used to place a call. AI can be used to understand the conditions at a point in time and then select the optimal communication path based on conditions and the timeliness requirements of that data being sent. AI will play a large part in optimising communications and increasing communication reliability.
We are also going to see AI models interacting with other AI models. Separate AI models with separate understandings of different parameters can co-operate to provide increased awareness of factors such as our climate, for example. There’s a lot of excitement around the potential there.
There’s also a lot of more personal and day-to-day things that NASA will pursue in the area of AI. For example, NASA is examining ways in which AI can help its workforce, from developing imagery to summarising notes from a meeting. It is important to see AI not as a threat that will replace the work of humans but as a tool to make our work easier and more efficient.
Can you provide some recent examples of projects focused on AI technologies?
The Prithvi climate model I mentioned earlier is a great example.
Another example is the use of AI to design structural components. AI enables a much more fluid and non-linear design than those made by humans. AI can also generate the component much quicker than humans, yet achieve equal or better results than the human-generated version when the component is stress tested.
The Mars Perseverance Rover and the Ingenuity Mars Helicopter that flew on the surface of Mars are some key examples of how adaptive onboard AI is helping in our research. In addition, we are also looking at longer-term human space flight – beyond the moon. For instance, we are developing technology designed to deal with medical crew emergencies. Typically, astronauts are not medical doctors and, in the case of such factors as a communications blackout, they may be unable to contact an expert back on Earth for assistance. It is important that they are equipped to deal with the emergency in real time. We are working to provide crew with an AI capability to help quickly diagnose medical conditions and suggest potential treatment actions.
What challenges have you found when implementing AI technologies and how are these being addressed?
As I mentioned earlier, there is the issue of electronics on, for example, a space-based system operating in an extreme environment and exposed to radiation. There is also the power budget and weight constraints. They are enormous challenges that those of us working with AI on Earth don’t have to think about. There is also the added challenge of having to bring that compute with you – you can’t just reach out to the Cloud to help process a large pile of data. Space brings a whole different, unique set of challenges.
There is a lot of fear surrounding the implementation of AI. Rightfully so, people are concerned about such things as ethical and responsible use, privacy, and transparency. It will take time for us to get comfortable with the technology and the potential that it brings. Thankfully, unlike other federal agencies, NASA is not involved with providing public services, such as processing claims for health benefits. Such agencies have a lot more work in terms of addressing the responsible, privacy, and ethical use concerns that exist.
There are also cultural difficulties. Like any organisation, we have to address workforce upskilling and reskilling. Some people are eager to jump into AI at full speed, while others are quite reserved about it. Every organisation has sceptics who are a little alarmed about new things and aren’t yet comfortable with them. If I didn’t account for that as a challenge, I would be remiss. We have to meet people where they are, and then help them to be equipped to effectively use these tools, to know what tool is best for which need, and help them to use the tools responsibly. It’s going to take time for the entire workforce to be ready to fully embrace all that AI offers. However, I must stress that the culture at NASA is very trailblazing. When people at NASA see an opportunity that is enabled by the use of technology, they are often eager to embrace it. Although there’s risk inherent in adopting any new technology, NASA is quite effective at managing risk. NASA will balance innovation with prudent risk management as we adopt and implement more and more AI, especially when complementing the entire workforce with Generative AI tools.
Please note, this article will also appear in the 20th edition of our quarterly publication.
