Astronomers used a neural network to clean up an image of the supermassive black hole Sagittarius A* at the center of our galaxy. The results were incredible. However, the Nobel laureate warns: AI — don’t rely too heavily on artificial intelligence, because it’s not a “magic wand.”
Neural network against cosmic noise
The supermassive black hole Sagittarius A* lurks at the heart of the Milky Way, 26,000 light-years away. Its mass is 4.3 million solar masses, and the diameter of its event horizon reaches 25 million kilometers. Proof of its existence brought Reinhard Genzel and Andrea Ghez the Nobel Prize in Physics in 2020. But even after the first direct image obtained by the Event Horizon Telescope (EHT) in 2022, its internal structure and behavior remained unclear due to the blurriness of the images.
To make images of Sagittarius A* clearer, an international team of scientists turned to artificial intelligence. They trained the neural network on millions of synthetic simulations of black holes, using even EHT data that had previously been considered unusable due to strong interference, primarily from the Earth’s atmosphere. After training, the AI turned to real observations of Sagittarius A*.
“It is very difficult to deal with data from the Event Horizon Telescope. A neural network is ideally suited to solve this problem,” explains study co-author Michael Janssen from Radboud University.
The image cleaned up by artificial intelligence led to some awesome findings. It suggested that Sagittarius A* rotates at an incredible speed — 80–90% of the maximum possible, which is a significant fraction of the speed of light. Its axis of rotation is probably directed toward Earth. In addition, the model indicates that the radiation from the hole does not originate from powerful jets, which is often the case, but directly from the superheated accretion disk.
Skepticism of the Nobel laureate
However, the team’s enthusiasm was met with a cold shower of skepticism. Reinhard Genzel, whose work laid the foundation for the study of Sagittarius A*, expressed doubts.
“I am very sympathetic and interested in their work. But artificial intelligence is not a panacea,” he said. Genzel believes that low-quality input data could have led to significant distortions in the AI results. What the model showed may be an artifact of processing rather than a true picture.
AI’s prospects in favor of science
Despite Genzel’s skepticism, no one denies the potential of AI for astronomy. The volume of data from modern telescopes, such as the Vera Rubin Observatory, is simply impossible to process manually. AI is already an integral part of object classification. However, scientists emphasize that it should be used with extreme caution, especially when applied to complex data from sources such as the EHT.
The struggle between innovative AI and conservative scientific skepticism continues. New data and improved models should finally determine whether the heart of our galaxy is really spinning at incredible speeds, or whether this is just an illusion created by the algorithm.
Earlier, we reported on how the image of the black hole Sagittarius A* turned out to be erroneous.
According to Live Science
