Breakthrough Listen's AI uncovers 8 new alien signals from 820 stars

Breakthrough Listen, the initiative dedicated to finding signs of intelligent life in the universe, has released results from a novel AI-powered method to detect technosignatures. Published in Nature Astronomy, the research was led by University of Toronto undergraduate student Peter Ma, who started working with the team in high school. The team analyzed 480 hours of data from the Green Bank Telescope (GBT) in West Virginia, scanning 820 stars and identifying roughly 3 million signals. Using a two-step AI process—an autoencoder to learn fundamental signal properties and a random forest classifier to separate candidates from noise—they found eight signals of interest that matched the expected characteristics of genuine extraterrestrial transmissions.

Classical technosignature algorithms compare on-target and off-target scans to filter Earth-based interference, but they struggle in crowded radio bands. The new AI inserts simulated signals into real data to train the model, effectively ‘listening for a whisper in a crowded room.’ While re-observations of the eight targets did not yield persistent signals, the technique dramatically improves filtering speed and accuracy. Breakthrough Listen's Principal Investigator Andrew Siemion noted that the algorithm was even more effective than prior methods, such as the one that found the BLC1 candidate in 2021. The team plans to apply the AI to larger datasets, hoping to eventually confirm a true technosignature and answer humanity’s profound question: Are we alone?