For thirty years, the Hubble Space Telescope has gathered light from the cosmos into an archive so vast it exceeds the reach of human attention. Two ESA researchers answered this limitation not with more astronomers, but with a machine trained to notice what people miss — and in doing so, confirmed over 1,300 genuine anomalies, including 43 objects that resist all known classification. The discovery speaks less to any single strange galaxy than to a quiet turning point: the moment when humanity began delegating the act of cosmic wonder to algorithms, not out of indifference, but out of necessi
AI Discovers 1,300+ Anomalies in 30 Years of Hubble Data, Including 43 Unclassifiable Objects
Forty-three objects with morphologies defying classification
Mark
So they found thirteen hundred unusual things in data that's been sitting there for thirty years. Why didn't astronomers spot these before?
Mimi
They're good at finding anomalies, but there's just too much data. 1.7 million observations. A human team can only look at so much, and they tend to focus on what they're specifically searching for.
Luke
Right, but I want to be careful here. The AI flagged anomalies. Humans then confirmed them. So we're not saying the AI discovered these—we're saying it helped humans find them faster.
Mimi
That's fair. The AI did the sorting work that would have taken humans years.
Mark
And what were most of these anomalies? Merging galaxies?
Mimi
About half, yes. Then gravitational lenses, protoplanetary disks, lensed quasars. Things that are individually rare but scientifically important.
Luke
Though the lenses need more observation to confirm. The paper says they're "thought to be" gravitational lenses.
Mark
What about those forty-three objects that don't fit anywhere?
Mimi
That's the interesting part. They might be jellyfish galaxies—galaxies losing gas as they move through galaxy clusters. But the researchers don't know for sure.
Luke
And they're releasing them without classification, which is honest. They're saying: we don't know what these are.
Mark
So what happens next?
Mimi
More observation, more study. But also, this tool gets used on the next big surveys. Euclid, Vera Rubin. Those will produce so much data that you basically need AI to find anything unusual.
Luke
That's the real story—not the thirteen hundred objects, but that human astronomy is becoming impossible without machine help.
Le Pouls
- A 1.7-million-observation archive had quietly accumulated decades of overlooked strangeness — objects hiding in the background of images taken for entirely different purposes.
- AnomalyMatch, an AI built by ESA researchers David O'Ryan and Pablo Gómez, tore through that archive at a speed no human team could match, surfacing over 1,300 confirmed anomalies including merging galaxies, gravitational lenses, and protoplanetary disks.
- Forty-three objects defied every existing category entirely, prompting the researchers to release them as open unknowns rather than force them into frameworks that don't fit — a rare act of institutional humility.
- The AI even identified lensed quasars without being trained to do so, suggesting it had internalized something deeper than a checklist — a generalized sense of the anomalous.
- With Euclid and the Vera Rubin Observatory set to flood science with nightly data volumes that dwarf Hubble's entire legacy, tools like AnomalyMatch are shifting from impressive to indispensable.
For thirty years, the Hubble Space Telescope has gathered light from the cosmos into an archive so vast it exceeds the reach of human attention. Two ESA researchers answered this limitation not with more astronomers, but with a machine trained to notice what people miss — and in doing so, confirmed over 1,300 genuine anomalies, including 43 objects that resist all known classification. The discovery speaks less to any single strange galaxy than to a quiet turning point: the moment when humanity began delegating the act of cosmic wonder to algorithms, not out of indifference, but out of necessity.
Thirty years of Hubble observations have produced 1.7 million data points — a mountain no human team could fully search. ESA researchers David O'Ryan and Pablo Gómez built AnomalyMatch to do what people inevitably cannot: scan the entire archive for objects that don't belong, flagging anomalies for human review. When astronomers examined what the AI surfaced, they confirmed more than 1,300 genuine oddities rare and strange enough to warrant serious scientific attention.
The project began with a focused goal — finding edge-on protoplanetary disks that tend to appear incidentally in the background of images rather than as intended targets. But the search expanded to include merging galaxies, gravitational lenses, and jellyfish galaxies, whose trailing streamers of gas and dust are stripped away as they move through the dense medium of galaxy clusters. About half of all confirmed anomalies turned out to be merging galaxies; gravitational lenses were the next most common, pending confirmation.
Most striking were 43 objects that fit no existing category at all. Rather than force them into familiar frameworks, the researchers released them to the broader scientific community as open questions. The AI also independently identified lensed quasars — objects it had never been trained to find — suggesting it had learned something more fundamental than a list of shapes to match.
The deeper significance is forward-looking. Euclid and the Vera C. Rubin Observatory will soon generate data volumes each night that make Hubble's archive look modest. Manual inspection at that scale is simply impossible. AnomalyMatch points toward a future where AI is not a convenience but a requirement — the only reliable way to ensure the universe's strangest objects don't quietly disappear into the noise.
Thirty years of observations from the Hubble Space Telescope have yielded 1.7 million separate data points—a mountain of information so vast that human astronomers, for all their skill at spotting the unusual, inevitably miss things. Two researchers at the European Space Agency decided to build a machine that wouldn't.
David O'Ryan and Pablo Gómez created AnomalyMatch, an artificial intelligence system trained to recognize patterns in astronomical data and flag objects that deviate from the norm. The software worked through millions of Hubble observations in a fraction of the time it would take a human team to accomplish the same task. When astronomers then manually reviewed what the AI had flagged, they confirmed over 1,300 genuine anomalies—objects rare enough and strange enough to warrant scientific attention.
The motivation for the project was specific. Astronomers had long known that edge-on protoplanetary disks—the dusty rings around young stars, viewed from their thin edge—were worth studying, but these objects rarely appeared as the primary target of an observation. They showed up in the background, easy to overlook. O'Ryan and Gómez expanded their search to include other categories of unusual objects: galaxies in the process of merging, gravitational lenses where massive objects bend light from distant sources, and jellyfish galaxies, which have a distinctive appearance. The AI, once trained, began pulling these objects from the archive.
About half of what AnomalyMatch found were galaxies actively merging with one another. The next most common category appeared to be gravitational lenses, though these would require additional observations to confirm with certainty. The system also recovered several edge-on protoplanetary disks and identified lensed quasars—a type of object it had not been explicitly trained to recognize, suggesting the AI had learned something more general about what constitutes an anomaly.
But then there were the forty-three objects that refused to fit into any existing category. The researchers examined them carefully and found no precedent in the literature. Some might not be galaxies at all, the team noted, but rather objects from a class of phenomena that astronomers have limited experience classifying. Rather than force these into an existing framework, O'Ryan and Gómez released them to the broader scientific community as unknowns, inviting further study. They speculated that many could be jellyfish galaxies—objects that appear to show signs of ram pressure stripping, a process in which galaxies moving through the gas that fills galaxy clusters lose their own gas and dust, leaving trailing streamers behind them like a comet's tail.
The real significance of the work, however, lies not in any single discovery but in what AnomalyMatch demonstrates about the future of astronomical research. The next generation of surveys—Euclid and the Vera C. Rubin Observatory among them—will generate such enormous volumes of data each night that manual inspection becomes impossible. Gómez noted that the tool represents a powerful way to extract scientific value from archives that would otherwise overwhelm human capacity. As these new surveys come online and begin collecting data at scales that dwarf even Hubble's legacy, AI systems like this one may become not a luxury but a necessity, the only practical way to ensure that the universe's strangest objects do not simply vanish into the noise.
Citations marquantes
This is a powerful demonstration of how AI can enhance the scientific return of archival datasets. The discovery of so many previously undocumented anomalies in Hubble data underscores the tool's potential for future surveys.— Pablo Gómez, European Space Agency
Envie de l'histoire complète ?
Lire l'original sur IFLScience ↗
Technology