ESA Scientists Find Hundreds of New Galaxies in 36 Years of Hubble Data Using AI
Updated
Updated · Financial Times · Jun 30
ESA Scientists Find Hundreds of New Galaxies in 36 Years of Hubble Data Using AI
1 articles · Updated · Financial Times · Jun 30
Summary
David O’Ryan and Pablo Gómez uncovered hundreds of previously unseen astronomical objects in Hubble’s 36-year image archive by using AnomalyMatch, a neural network trained to flag unusual patterns.
The finds show AI can extract new science from old data, with Gómez saying the system ran on a single GPU in a few days rather than requiring massive computing power.
Oxford physicist Héloïse Stevance reported a separate AI tool cut human screening of supernova alerts by 85%, using about 15,000 examples and laptop-scale computing.
Astronomers are also using AI emulators to model stellar evolution and exoplanet atmospheres faster, ahead of a data surge from new observatories.
That push is becoming urgent as Chile’s Vera Rubin Observatory prepares to capture a 3-billion-pixel image every 30 seconds for 10 years.
Is AI in astronomy truly democratizing science, or will computing demands create a new digital divide?
As AI automates analysis, what is the new role for human intuition in the next era of cosmic exploration?
Could AI's reliance on known data cause it to miss the universe's most revolutionary discoveries?
AI Uncovers 1,300 New Cosmic Anomalies in Hubble’s 36-Year Archive: The AnomalyMatch Revolution
Overview
In early 2026, the AnomalyMatch AI tool, developed through a collaboration between ESA and NASA, was deployed to systematically analyze the Hubble Space Telescope’s vast 36-year data archive. By processing nearly 100 million image cutouts, AnomalyMatch identified over 1,300 anomalous objects, with more than 800 never seen before by human astronomers. This breakthrough instantly expanded the known catalog of cosmic curiosities and set the stage for a new era of astrophysical investigation, demonstrating how AI can rapidly uncover hidden wonders in massive scientific datasets.