Searching for Extraterrestrials with Artificial Intelligence

Many have seen the viral “Alien Abduction” Halloween costume of a bright green, saucer-eyed alien trying to kidnap a human. It humorously and frighteningly pokes fun at the potential dangers of sharing our universe with extraterrestrials. 

Extraterrestrial beings, or aliens, are life forms that may exist outside of Earth. The Fermi Paradox points out that there is a high probability of the existence of extraterrestrial life, but scientists have no direct evidence to confirm this hypothesis. In fact, there are as many as 300 million potentially habitable planets that are similar in size to Earth, revolve around Sun-like stars, and could support liquid water. Considering this and the fact that the universe is 13.7 billion years old, we are faced with the question of why aliens have not visited us yet. 

One possible answer is that aliens have visited our solar system, and are communicating with us through radio emission signals or radio waves. The Search for Extraterrestrial Intelligence (SETI) Institute is using artificial intelligence (AI) to attempt to locate these signals. 

SETI began using AI to search for extraterrestrials in the Breakthrough Listen Initiative, a decadal project funded by billionaire Yuri Milner to search one million stars for intelligent life. The goal was to find a steady frequency change in radio emissions coming from a star that could be the host of a habitable planet. The consistent change in the frequency could signal an alien transmitter moving away from Earth.

To simplify, think of all radio emissions and radio interference as data. Scientists are now able to sort through this data using an AI model developed by University of Toronto researcher Peter Ma. This allows them to separate insignificant data—like radio interference—from significant parts for further investigation. The AI model was developed using data from Green Bank Telescope, which contains information from 820 stars. 

To teach the model how to distinguish between significant and insignificant data, the researchers used semi-supervised learning, the middle ground between supervised learning and unsupervised learning. Supervised learning trains models to produce a desired outcome by feeding them data, but the human interference required made the researchers feel this method restricted their model. Unsupervised learning was used to train the model to separate unlabeled datasets without human interference, but this removed a necessary level of human control over the model. Semi-supervised learning took the pros of each system without the downsides, making it the most effective way to train the model. 

The model was fed data of “normal signals” (like radio interference) so that it could separate them from the non-interference data. Any data the model wasn’t trained on was automatically flagged. The model picked up 489 hours worth of data and 30,000 results that were not radio interference. Ma then went through each signal manually to ensure accuracy. Both the model and Ma flagged eight unusual signals. 

These eight signals had some key characteristics SETI looks for when searching for extraterrestrial intelligence. One is the presence of narrow bands, which cover a small range of frequencies in hertz; natural phenomena tend to be broadband and cover a wide range of frequencies. Another is if the signal’s frequency changes over time, suggesting its source is not local to the radio observatory and could be a distant celestial object. The last quality they look for is if the signals are present in ON-source observations, meaning that if they were to point a telescope at the source, they could observe the radio emissions. It is important that the signals are not present in OFF-observations, in which the source’s radio emissions disappear when the telescope turns away from the object. Determining whether these sources are radio interference is critical because it usually appears in both ON and OFF source observations.

Before concluding definitively if these signals are extraterrestrial, the researchers must obtain the same signals multiple times—consistent signaling could point to extraterrestrial communication. During the brief follow-up observations using the Green Bank Radio Telescope, the eight signals were not found. Thus, more observations and analysis are necessary before drawing any conclusions. These eight signals were not picked up on by traditional, non-artificial intelligence methods, proving the AI algorithm obtains more accurate results when scanning data. Ultimately, whether humans are alone in the universe or not, the search for extraterrestrial life evokes a science-fiction-like fear with possibilities that can only be imagined through things like the “Alien Abduction” costume.