As reported by ESA Hubble, more than 800 objects that had never appeared in the scientific literature, pulled from data that had been sitting in an archive for years. It took two and a half days. Not two and a half years of grant applications and follow-up work. Two and a half days of computer time.
The work came from two European Space Agency astronomers, David O’Ryan and Pablo Gómez. They pointed a new AI tool at the Hubble archive and let it sort through nearly 100 million small image cutouts. The results were published in the journal Astronomy & Astrophysics.
What strikes us is not that a machine found strange things in the sky. It found them in pictures we had already taken.
The number that stops you
As mentioned, the tool worked through nearly 100 million image cutouts in about two and a half days. O’Ryan and Gómez then looked by hand at the top candidates and confirmed more than 1,300 genuine oddities. Of those, more than 800 had never been documented before.
That last figure changes how you read the whole thing. Hubble is one of the most studied instruments in the history of science. Its archive has been picked over by professionals and volunteers for decades. You would expect the obvious oddities to have been found already. Gómez said as much when he called it a “great result”, noting that in Hubble data “you might expect many to have already been found.”
What the tool actually is
The tool is called AnomalyMatch. It helps to be plain about what it is and is not. It is not a robot astronomer that understands galaxies. It is a program that learns to spot patterns. It picks up what “unusual” looks like from a small set of examples plus a much larger pile of unlabelled images, and it keeps asking a human to judge its most uncertain guesses, then folds those answers back in.
Gómez called it a “fantastic use of AI to maximise the scientific output of the Hubble archive.” That is a researcher praising his own method, so read it as a view from inside the work rather than settled opinion. The idea behind it, though, is modest: a fast filter, with human review at the end.
What it found
As reported by Universe Today, the biggest group by far was 417 merging and interacting galaxies, pairs and clusters caught mid-collision. The search also turned up 86 new potential gravitational lenses, where a massive object in the foreground bends and magnifies the light of something behind it, plus ring galaxies and 35 jellyfish galaxies, named for the streaming tails of gas they trail as they move through a crowded cluster.
And then the odd bin at the end: several dozen objects that did not fit any category.
Why “anomaly” doesn’t mean what you think
In everyday use, “anomaly” suggests something inexplicable, something that breaks the rules. Here it means something drier and more useful: statistically unusual. An object that looks different enough from ordinary galaxies and stars that it stands out to a pattern-matcher.
Most of these objects are well understood as phenomena. Merging galaxies are not mysteries. Astronomers know roughly what happens when two galaxies fall together. What made them stand out is that they are rare in the frame, not that they defy explanation. The value is not that the tool found things science cannot explain. It found rare, real objects nobody had catalogued, fast enough to make systematic searching practical for the first time. O’Ryan put the raw resource plainly: “Archival observations from the Hubble Space Telescope now stretch back 35 years, providing a treasure trove of data in which astrophysical anomalies might be found.”
The real point is the method
Perhaps the most interesting thing here is not any single ring galaxy or unclassifiable smudge. It is what the exercise says about data we already have. These 800-plus objects were not newly observed. They were photographed years ago and then sat, unremarked, inside a public archive, until something could read all of it at once.
That reframes a quiet assumption in a lot of science. The instinct is usually to build the next, bigger instrument to see more. This result points the other way, toward mining what has already been collected. The timing matters, because the pile is about to get much larger. The Vera C. Rubin Observatory alone is expected to gather more than 50 petabytes of images over its ten-year survey, far more than any team could inspect by eye. Euclid and the Nancy Grace Roman Space Telescope will add their own floods of data.
Gómez expects the approach to travel. Finding so many oddities where you would assume the obvious ones were long gone, he said, “shows how useful this tool will be for other large datasets.” In a later comment he said the undocumented finds underscore “the tool’s potential for future surveys.” That is a prediction from the person who built the method, not a guarantee, but it fits the shape of what is coming.
What we keep returning to is how ordinary the source was. No new telescope, no new night of observing. Just a public archive, a pattern-spotter fast enough to read all of it, and two people willing to check the shortlist. The discoveries had been there for years. What changed was the ability to notice them.