The famous first image of a black hole is now twice as sharp. The research team used artificial intelligence to dramatically improve the first images of 2019. In this image, the black hole at the center of the M87 galaxy is darker and larger than the first picture.
I am an astronomer who studies and writes about cosmology, black holes and exoplanets. Astronomers have been using AI for decades. In fact, in 1990, astronomers at the University of Arizona, where I am a professor, were among the first to study the shape of galaxies using a form of AI called neural networks.
Since then, AI has spread to all areas of astronomy. As technology becomes more powerful, AI algorithms are beginning to help astronomers tap into massive data sets and discover new knowledge about the universe.
Better telescopes, more data
For as long as astronomy has been a science, it has involved attempts to understand the multitude of objects in the night sky. When the only tools available were the naked eye or a simple telescope, it was relatively simple and could only see thousands of stars and a handful of planets.
100 years ago, Edwin Hubble used a newly built telescope to show that the universe is filled not just with stars and clouds of gas, but with countless galaxies. As telescopes continue to improve, the number of celestial objects humans can see and the amount of data that astronomers need to sort has also increased exponentially.
For example, the soon-to-be-completed Vera Rubin Observatory in Chile will create images so large that you’ll need 1,500 high-definition television screens to see them in their entirety. In 10 years, we expect to generate 0.5 exabytes of data. This is approximately 50,000 times the amount of information contained in all the books in the Library of Congress.
There are 20 telescopes with mirror diameters over 20 feet (6 meters). AI algorithms are the only way astronomers can hope to process all the data currently available. There are many ways AI has proven to help process this data.
pattern selection
In astronomy, we often look for needles in haystacks. About 99% of pixels in astronomical images contain background radiation, light from other sources, or darkness in space. Only 1% have the subtle shape of a faint galaxy.
AI algorithms, especially neural networks that use large numbers of interconnected nodes and can learn to recognize patterns, are ideal for identifying patterns in the galaxy. Astronomers began using neural networks to classify galaxies in the early 2010s. Currently, the algorithm is very effective and can classify galaxies with 98% accuracy.
This story has been repeated in other areas of astronomy. Astronomers working on SETI (Search for Extraterrestrial Intelligence) use radio telescopes to search for signals from distant civilizations. In the early days, radio astronomers scanned charts by eye, looking for unexplained anomalies. Recently, researchers used his 150,000 personal computers and his 1.8 million citizen scientists to search for artificial radio signals. Now researchers are using his AI to sift through sets of data much faster and more thoroughly than humans can. This enabled his SETI efforts to cover more areas while significantly reducing the number of false positive signals.
Another example is the search for exoplanets. Astronomers discovered most of her 5,300 known exoplanets by measuring the dip in the amount of light coming from the stars as the planets pass in front of them. AI tools can now identify signs of exoplanets with 96% accuracy.
make new discoveries
AI has proven to be excellent at identifying known objects that astronomers tell us to look for, such as galaxies and exoplanets. However, it is also very powerful in finding objects and phenomena that have been theorized but have yet to be discovered in the real world.
The team has used this approach to detect new exoplanets, learn about the progenitor stars that led to the formation and growth of the Milky Way, and predict signs of new types of gravitational waves.
To do this, astronomers first use AI to transform theoretical models into observational signatures (with realistic levels of noise). Then use machine learning to improve the AI’s ability to detect predicted phenomena.
Finally, radio astronomers also use AI algorithms to sift through signals that don’t correspond to known phenomena. A South African team recently discovered a unique object that may be the remnant of an explosive merger of two supermassive black holes. If this proves to be true, the data will allow a new test of general relativity – Albert Einstein’s description of spacetime.
make predictions and plug holes
Like many areas of life these days, generative AI and large-scale language models like ChatGPT are also making waves in astronomy.
The team that created the first image of a black hole in 2019 used generative AI to create new images. To that end, we first taught the AI how to recognize black holes by giving them simulations of different kinds of black holes. The team then used the AI model it built to fill in the gaps in the massive amount of data collected by radio telescopes at the black hole M87.
Using this simulated data, the team was able to create a new image that was twice as sharp as the original and perfectly matched general relativity predictions.
Astronomers are also turning to AI to defuse the complexity of modern research. A team at the Harvard-Smithsonian Center for Astrophysics created a language model called astroBERT to read and organize his 15 million scientific papers on astronomy. Another NASA-based team has even proposed using AI in prioritizing astronomy projects, a process that astronomers engage with every his decade.
As AI advances, it has become an essential tool for astronomers. As telescopes improve, data sets grow, and AI continues to improve, this technology could play a central role in future discoveries about the universe.
Chris Impey, Distinguished University Professor of Astronomy, University of Arizona
This article is republished from The Conversation under a Creative Commons license. Please read the original article.
