Space probes, Credit: NASA
During the Cold War, the Soviet Union and the United States were in a breathless competition, battling over supremacy in spaceflight. Russian cosmonaut Yuri Gagarin became the first man to escape orbit and reach space in 1961, while Americans Neil Armstrong and Buzz Aldrin became the first to walk on our moon in 19691 . The fierce space race, which defined an entire era, was certainly a competition, but it also led to mankind studying and theorizing about the endless expanse like never before. Unmanned craft such as Voyager 1 were sent out into space, incredibly advanced data gathering instruments such as the famous Hubble Telescope were built, and countless researchers and scientists constantly made new discoveries about our solar system and the observable universe. These discoveries have only gained momentum over time and become ever more fascinating. They range from our discovery of potentially the first artificial object to emerge from deep space, Oumuamua, to the recently detected giant black hole in the center of the Abell 2261 cluster2.
Whatever the specific nature of these observations and discoveries might be, one fact about space exploration remains unequivocally true: more data is being gathered now than ever before. This data streams in from radio and optical telescopes and sensors of all types and powers our pursuit to understand the wonders contained within our universe. According to Kevin Murphy from NASA’s earth science data systems program, “NASA was generating 12.1TB of data every single day from thousands of sensors and systems dotted across the world and space.3 ” Raw data produced by NASA and other space organizations across the globe holds discoveries just waiting to be revealed. The quantity of this data is so vast, however, that simply combing through the data manually is entirely unviable. So how can we unlock new discoveries within this treasure trove of astronomical data and raise our understanding of the endless reaches of space? By using the power of artificial intelligence, and machine learning, of course! AI techniques such as deep learning, a subfield of machine learning inspired by biology and focused on building artificial neural networks with many layers, have been particularly effective. Deep networks have already been used to detect planets or identify unexpected occlusions in distant stars. Anomaly detection algorithms can help spot unexpected phenomena and focus our attention on objects that would otherwise escape study. Clustering algorithms can group like objects by behavior, luminosity and other similarities so that we can supply examples of objects of interest and then have similar objects automatically identified in massive data sets. These are only a few of the tools from machine learning and artificial intelligence that can help us derive the answers to the mysteries of space at great speed and scale.
Mars probes and data retrieved, Credit: NASA
Machine Learning is a major focus for NASA’s Frontier Development Lab (FDL) which brings together researchers from the SETI Institute and NASA’s Ames Research Center, both based in Silicon Valley. The incredible scientists and researchers on this team have publicly shared their groundbreaking findings4 for the benefit of the larger academic community. Commercial companies partnered with the FDL contribute powerful resources such as supercomputer access, funding, and of course, artificial intelligence powered software. For example, working with Google, the NASA team developed a neural network with a “Bayesian” structure and deployed it to sort through data gathered from the atmosphere of the WASP-12b exoplanet. The neural network generated very promising results that were more accurate than traditional statistical approaches. According to Oxford University graduate Adam Cobb, who was involved with deploying the neural network model, “We found out right away that the neural network had better accuracy than random forest in identifying the abundance of various molecules in WASP-12b’s atmosphere.5 ” More important than the accuracy of the findings however, is that the model can show the uncertainty around certain parts of its predictions. This helps researchers by alerting them to those parts of the algorithm’s findings that may require more data sampling.
It is clear that machine learning has great potential in extrapolating meaningful findings from raw data, but AI’s potential in discovering space extends far beyond this. Instead of just finding correlations and connections in captured data, AI can also be used to gather more data, more effectively. At the Goldschmidt Geochemistry conference in June 2020, NASA revealed its plans to develop probes to scan Mars for potential signs of life. Amazingly, these probes will autonomously process gathered data and decide on which subsets to transmit back to earth6 . How is it possible for a probe to automatically decide what data is fit to be sent? Once again, by employing deep neural networks. Not only will this automated analysis be more efficient as it cuts down the amount of data transmitted over lowbandwidth inter planetary links, it also eliminates laborious manual processes and is far more cost effective. According to Eric Lyness, software lead at the Planetary Environments Lab at NASA’s Goddard Space Flight Center, “Data from a rover on Mars can cost as much as 100,000 times as much as data on your cell phone, so we need to make those bits as scientifically valuable as possible.7 ” And it’s not just about efficiency. Due to the long lag times involved in sending inter-planetary messages, space probes and rovers controlled by artificial intelligence are an absolute necessity if we are to safely explore other planets, celestial bodies and deep space. One day, a swarm of such autonomous systems will explore the cosmos for strange new worlds… and perhaps even seek out new life and new civilizations!
Powered by artificial intelligence, space exploration and our understanding of the cosmos as a whole is about to enter an exciting new era. AI as a means of automated large scale data analysis, as a mechanism to process sensor data remotely and optimize communications, and finally as a means of autonomous control will make it the cornerstone technology for future space exploration. Truly, the final frontier has never been more open and accessible, and artificial intelligence promises to open the door to the world of tomorrow.
1. History.com Editors. (2010, February 22). The Space Race. Retrieved January 04, 2021, from https://www.history.com/topics/cold-war/space-race
2. Wall, M. (2021, January 01). A giant black hole keeps evading detection and scientists can’t explain it. Retrieved January 04, 2021, from https://www.space.com/abell-2261-supermassive-black-hole-missing
3. Gorey, C. (2017, October 26). The volume of data NASA has to manage is mind-boggling. Retrieved January 04, 2021, from https://www.siliconrepublic.com/enterprise/nasa-data-figures
4. Expert System Team. (2020, December 21). What is Machine Learning? A definition - Expert System. Retrieved January 04, 2021, from https://www.expert.ai/blog/machine-learning-definition/
5. Shekhtman, S. (2019, November 15). NASA Applying AI Technologies to Problems in Space Science. Retrieved January 04, 2021, from https://www.nasa.gov/feature/goddard/2019/nasa-takes-a-cue-from-silicon-valley-to-hatch-artificial-intelligence-technologies/
6. Carter, J. (2020, June 24). NASA Now Has Alien Life-Hunting Robot Explorers That Only Phone Home When They’ve Hit The Jackpot. Retrieved January 24, 2021, from https://www.forbes.com/sites/jamiecartereurope/2020/06/25/nasa-wants-less-needy-robots-thatonly-phone-home-when-they-think-theyve-found-alien-life/?sh=5fb21fda46ca
7. Brownlee, J. (2020, August 14). What is Deep Learning? Retrieved January 24, 2021, from https://machinelearningmastery.com/what-is-deep-learning/