On April 10, 2019, a worldwide choir of wonder welcomed the unveiling of the first image of a black hole, the one in the center of the M87 galaxy, the shadow of a colossus as massive as seven billion suns. A ring of fire, recognizable but very blurred, of an impalpable evanescence like its presence, up to that moment, 55 million light years away from us.
It was the result of processing petabytes of data, to contain which the hard disks and thousands of PCs would be needed. To navigate that sea of information gathered by the Event Horizon Telescope (EHT), eight radio telescopes scattered all over the world who have all pointed to the same remote corner of the sky, they are served months and months of work. However, someone, of that image, has built a meme and made fun of it. Maybe because that black hole didn’t look enough like Gargantua from Interstellarthe masterpiece by Christopher Nolan, with whom the Nobel prize winner Kip Thorne also collaborated.
Four years after that event, some scientists who were part of the Eht team, have entrusted that data to Primo, a new algorithm “based on dictionary learning using high-fidelity simulations of black hole accretion as a training set”. And the generative artificial intelligence (as ChatGpt is, for example) has returned an image twice as defined.
Chris Impeyprofessor of astronomy at the University of Arizona, told in an article about The Conversation that it was he and his colleagues, already in 1990, among the first to use artificial intelligence with neural networks to study the shapes of galaxies. That technology was needed, Impey explains, because after Edwin Hubble’s discovery of the “geography” of the Universe, the fact that it is much, much more extensive beyond the confines of the Milky Way, and the ever greater power of instruments, the quantity of data has become so enormous that we need, let’s say, an extra “brain” at the service of astronomers. And which has the advantage, in addition to speed, of having an increasingly important reliability in avoiding “false positives”.
Order in disorder
In fact, it is difficult to think of something that produces more data than the whole universe, excluding ourselves. Which, in addition to generating an obvious memory overload, is also an opportunity. Why AI and machine learning are hungry for data, with those they train, learn and are able to give answers with a speed unthinkable for a human. It’s at discovering things that may have escaped, lost in the “noise”, recognizing patterns and classifying objects (and as we know, stars and galaxies number in the billions). Space telescopes such as Hubble (and now the James Webb space telescope), Gaia or the now retired Kepler, those on Earth, dozens scattered around the globe, conduct surveys that even from apparently dark corners of the sky they capture the light of millions of sources, each is a galaxy. Since the beginning of the 2000s artificial intelligence has been helping astrophysicists to put order and classify them according to their shape by handfuls of hundreds of thousands at a time, also helped by the human eye with projects which, unlike neural networks, exploit also the voluntary help of thousands of people to recognize if a galaxy is spiral, barred, or the strange shape is due to the collision between two of these objects.
Hunting for planets and gravitational waves
The space telescope Kepler, a glorious machine now out of use, during its operational life it has discovered over 4,000 exoplanets, i.e. planets that orbit around other stars. In his grande dataset, however, in a cloud of unclear observations and data, since 2017 the specially trained algorithms, sifting, have begun to discover new “anomalies” (Kepler observed the decrease in brightness of a star to associate it with a possible planet that transits in front of its star). These events are often regular, because the orbit is cyclical, therefore with patterns that must be sought and recognized. Google’s AI made the first of these discoveries, a new world, followed by dozens and then hundreds of planets in danger of escaping. Where the human eye or simple software cannot reach, deep learning, which typically learns from theoretical models on which studies are based, and from large database training sets, manages to bring out something that was perhaps latent ‘in watermark’.
If AI discovers 50 new planets by digging into NASA data
by SIMONE COSIMI
The AI is therefore useful for looking for weak anomalies, unclear, fuzzy signals, to be brought out more clearly. Just like the black hole image of M87. One of these is microlensing, the phenomenon of space-time distortion that a massive object produces by passing in front of another. It happens, very clearly, for galaxies, which amplify the light of other objects that are in the background behind them (observing from our point of view), objects whose light is bent and amplified by the gravity of those in the foreground. This is how Einstein’s crosses and rings originate, for example. And it happens, but with infinitesimal intensity, even for stars, planets, and small black holes as they pass in front of another star. It is a method of finding objects that wander in space such as planets and solitary black holes or, again, exoplanets.
Speaking of extreme instrumental precision and very weak signals arriving from afar, one immediately thinks of gravitational waves. For which artificial intelligence, trained with data deriving from models, will in the future be able to better recognize and describe the objects which, by colliding, give rise to the ripples of spacetime. Predicted by the general theory of relativity and “listen” for the first time in 2015.
It’s a matter of telling the machine what to look for, train it with synthetic examples of what it might look like, and then challenge it with the real cosmos. And it also works to delete what is not needed, indeed, disturbs. Like the interference and incursions of satellites scratching the sky while you are observing.
Space
Strange lights in the sky over Northern Italy? It was just Starlink satellites going into orbit
by Emanuele Capone
The future in the hands of AI
It is neither a prophecy nor catastrophism to say that the future is entrusted to artificial intelligence. At least from the point of view of scientists, astronomers in particular, it is reality. And that’s actually good news. “For example, the Vera Rubin Observatory in Chile, which will be completed soon, will take images so large that it would take 1,500 high-definition TV screens to see them all in their entirety. In 10 years it is expected to generate 0.5 exabytes of data, about 50,000 times the amount of information contained in all the books contained in the Library of Congress” Impey underlines again. The new campaigns for the search for gravitational waves will lead, especially after the construction of the Einstein telescope, to listen to the cosmos with a thousand times greater sensitivity. It will be up to the AI, through all this data, to classify objects, find anomalies and patterns and decode the Matrix of the Universe.
Even the Setithe institute for the research of extraterrestrial intelligence, it entrusts the scanning of radio signals to the algorithms that come from ears stretched towards the stars. For now, no ring, but one day perhaps it will be the signal, an alert or perhaps the voice lent to a synthetic intelligence to warn us, first, that there is a planet out there that has all the characteristics to host life. Or that there’s mail for us, coming from some remote corner of the sky. A message from a very advanced civilization, perhaps sent by another artificial intelligence trained for it.