The search for alien life is an ongoing and continually time-consuming process. From analyzing the surface of planets/moons to detection of exoplanets, these processes can take weeks, months, or even years. However, the advancements of artificial intelligence over the coming years can significantly speed up these processes. In fact, there are already rudimentary machine-learning processes helping us in astrobiology right now.
In late 2017 Google developed a form of machine-learning that was able to detect several new exoplanets. It was described as a neural network capable of analyzing data collected by NASA’s Kepler Space Telescope. This “AI” found two new exoplanets, and many were excited due to their size-similarities to Earth, as these are normally more difficult to detect. The planets are in the previously-known solar systems of Kepler-80 and Kepler-90, but were overlooked by scientists for years. The data sent back from Kepler was 3 years old, yet no one had detected these exoplanets. It wasn’t until a machine-learning algorithm carefully combed through the data that these discoveries came to light.
The Kepler Telescope uses the transit technique when searching for exoplanets. This means that it observes the slight dimming of the star’s lights as the planet passes in front of it. Large planets like Jupiter would cause a dimming around 1%, while Earth-sized planets would only dim by a fraction of that. Thus, making planets of this size extremely difficult to detect. In fact they are on the verge of being undetectable, and it is likely the reason why these exoplanets were not discovered sooner. However, this machine-learning network has been fed thousands of examples, and can then find minute similarities in patterns and phenomena from the transit information. During the first test, it found 30 possible planets and 4 which had a 90% chance of being an undiscovered exoplanet. The first one was part of a binary system and the second had a light dip that continually decreased. These are scenarios which had not previously been fed into the machine-learning system, but could be used to further increase its accuracy in the future. The other two were of course new exoplanets.
The technology is still not perfect, but there is definite promise in this field. As the AI gains more and more information, it can continually refine its exoplanet-searching algorithms. With the speed in advancements, it seems very plausible that the use of AI will significantly surpass any group of humans searching for exoplanets in speed and accuracy. Right now, humans need to make corrections and double-check the AI, but it doesn’t seem too far in the future that a neural network could have increased accuracy and speed than any person on Earth. The implications of this could extend to all aspects of astrobiology. As AI makes more and more nuanced and complex connections, it could potentially even develop life-detection methods that humans haven’t even considered. I know this is all speculation, but I don’t currently see any limits for the “intelligence” or computational power of AI. I would like to hear what anyone else thinks about the future of AI in astrobiology as well.
Very informative paper! I did not know that AI was being implemented in astrobiology currently, thanks for bringing me up to speed.
I used to be against AI until taking this class. My reasoning for changing my mind is two fold. First, I realized that the invention of calculators and algorithms have greatly reduced the time it takes to solve complex problems. I can only imagine what a learning calculator/algorithm can accomplish. Second, I realize that the inability to adapt causes death. If we as a human species want to survive, incorporating AI into our everyday life may become more and more necessary. True it could back fire, but why does everything have to be seen as a negative. Be like me, stay optimistic my friend!
Kenneth C Carroll