Artificial intelligence
With computers growing smarter than ever, training them how to spot and categorise astronomical objects could lead to a plethora of breakthroughs
The world is currently in the grips of an artificial intelligence (AI) frenzy. Everyone is trying ChatGPT, AI stocks are surging T to record highs and people are worried about whether machines are coming for their jobs. Yet astronomers have been using similar techniques to unlock the universe’s secrets for decades. “It goes back over 30 years,” says Chris Impey from the University of Arizona. In 1990, Impey’s colleagues at the university’s Steward Observatory used an artificial neural network (ANN) to both divide galaxies into groups based on their appearance and to distinguish between stars and galaxies in images of the sky. An ANN is fed a series of training images and is then set free to analyse new data. It looks for connections between data points and is based on a simplified version of the human brain.
Classifying galaxies is intricate work. “Humans can’t do it well enough,” Impey says. Deciding whether a galaxy is a spiral or an elliptical may be fairly straightforward, but smaller details can be crucial. The galaxy may or may not have a bar structure in the centre, be asymmetrical or have distortions close to its outer edge. “That might tell you that the galaxy had an interaction with a neighbouring galaxy that has long since sailed off to a different part of space, but the distortion remains,” Impey explains. “The distortions are subtle, a per cent or less.” It’s a bit like human scars. These small marks on our bodies tell a story about where we’ve been and the things that have happened to us.
Galaxy classification software has even had an impact far beyond the realms of astronomy. A decade ago, astronomers at the University of Cambridge teamed up with oncologists to use their techniques to more accurately diagnose breast cancer. The software was adapted and taught to look for cancer cells instead of galaxies. Tests showed that the software was at least as accurate as a doctor looking down a microscope – and often far quicker, too.
Technology has come a long way since 1990, and today’s computers are far superior. But then telescopes have also drastically improved. There’s currently an arms race between the amount of data flooding in from all these observatories and the computing power needed to analyse it all. “We now have 10 to 12 metre [32 to 39 foot] telescopes that didn’t exist a decade ago,” Impey says, “and we’re about to have 20 to 30 metre [65 to 98 foot] telescopes in the next six years.” The 30-metre Extremely Large Telescope, for example, is due to see first light in 2028. It has around 250 times the light-gathering area of the Hubble Space Telescope and will provide
