There are moments in the history of science when we are reminded, with humbling clarity, that what we thought we knew barely scratches the surface. This is one of them.
An artificial intelligence tool has combed through 35 years of archived images from NASA's Hubble Space Telescope and found more than 1,300 previously overlooked cosmic anomalies — including merging galaxies, gravitational lenses, and objects so strange they defy classification altogether. More than 800 of these had never been documented in scientific literature.
The sheer scale is staggering. Nearly 100 million image cutouts from the Hubble Legacy Archive, each just a few dozen pixels across, were analysed in two and a half days. A task that would have taken human astronomers years — perhaps decades — was accomplished before the weekend was out.
The Tool Behind the Discovery
The breakthrough was made possible by AnomalyMatch, a neural network developed by ESA researchers David O'Ryan and Pablo Gomez. Trained to recognise rare and unusual astronomical objects by detecting patterns in visual data, the AI mimics the way the human brain processes images — but at a pace no team of scientists could hope to match.
"Archival observations from the Hubble Space Telescope now span 35 years, offering a rich dataset in which astrophysical anomalies may be hidden," said O'Ryan, lead author of the study published in the journal Astronomy and Astrophysics.
Traditionally, such discoveries rely on manual inspection or pure serendipity. Expert astronomers are superb at spotting the unusual, but the sheer volume of Hubble data makes comprehensive human review impractical. Even citizen science projects, which have contributed enormously to galaxy classification, cannot keep pace with archives of this magnitude.
What They Found
The catalogue of anomalies reads like a cosmic bestiary. Most were galaxies caught in the act of merging or interacting — trailing long, elegant streams of stars and gas as they collide in slow motion across millions of years.
Others were gravitational lenses: cases where the gravity of a foreground galaxy warps spacetime itself, bending light from a more distant galaxy into haunting arcs and rings. The team also uncovered galaxies with enormous star-forming clumps, so-called "jellyfish galaxies" with gaseous tentacles streaming behind them, and edge-on planet-forming disks in our own Milky Way that bear an uncanny resemblance to hamburgers.
Perhaps most tantalising of all, several dozen objects refused to fit any existing classification scheme. They remain, for now, genuinely unidentified.
"This is a powerful demonstration of how AI can enhance the scientific return of archival datasets," said Gomez. "The discovery of so many previously undocumented anomalies in Hubble data underscores the tool's potential for future surveys."
A New Era of Discovery
The implications stretch far beyond Hubble. A new generation of telescopes is preparing to flood astronomers with data on an unprecedented scale. NASA's Nancy Grace Roman Space Telescope, ESA's Euclid mission, and the Vera C. Rubin Observatory — which will collect more than 50 petabytes of images over its ten-year survey — will all generate archives that dwarf even Hubble's.
Tools like AnomalyMatch will not replace the astronomer's trained eye, but they will become indispensable companions — sifting through mountains of data to flag the needles in a cosmic haystack the size of the universe itself.
It is a partnership that feels almost poetic: human curiosity built the telescope, human ingenuity built the AI, and together they have revealed that even our most celebrated instruments still had secrets left to give up.
After 35 years, Hubble is still full of surprises. One rather suspects the universe always will be.
