A new generation of high-powered telescopes is creating petabytes of data about comets, stars, and distant galaxies—forever altering the way that we understand the cosmos. The result is an embarrassment of riches, and the inherent challenge of sorting, analyzing, and creating meaning out of it. Next year’s opening of the Vera C. Rubin Observatory in Chile, which will have the largest digital camera ever built—a 3,200-megapixel device the size of a car, yielding unprecedented images of space—will only exacerbate this challenge. “We’re swamped with data, too much for anyone to go through, which means a lot of it doesn’t get used,” says Karl Gebhardt, the department chair of astronomy at the University of Texas. “That’s exactly where AI comes in.”
Gebhardt and his team are using AI to spot patterns in some 1 trillion elements within 1 billion images of space taken by the McDonald Observatory in West Texas. To help train the AI model, Gebhardt created an app for “citizen scientists,” inviting volunteers to view the department’s telescopic photos and swipe right if they think the image contains a valid star or swipe left if not. “We went from five people doing this initial sorting to 30,000 people working on it,” Gebhardt says. He and his team then fed those vetted images into an algorithm that searches for probable stars and galaxies. “It’s really improved our work,” Gebhardt says. “It’s a hybrid approach, with crowdsourced humans training the model and then AI taking it from there.”
Humans defining the training set is key to the success of the AI algorithm, as astronomers must ensure that AI findings are valid. For instance, an algorithm may not be able to detect wonky data that results from an error by a telescope’s digital camera. The human eye is also better than AI at spotting some features in the imagery, such as when a fast-moving particle creates a splash of light. “The eye just picks that up instantly,” Gebhardt says. “But it’s really hard to write code that identifies that on a consistent basis.”
In July, astronomers at the University of Texas used generative AI to develop an algorithm to discover stars in the final stage of life, known as dwarf stars, which contain important clues to the elements that make up the planets in our galaxy. Dwarf stars have historically eluded astronomers. They are difficult to discern and identify because they don’t emit much light. By using an algorithm to group visually similar items together, the UT astronomers pinpointed 375 promising-looking stars out of 100,000 possible white dwarfs—and then followed up with their telescopes to confirm the findings.
Ultimately, Gebhardt and other astronomers hope to use AI to probe one of the big questions looming in the night sky: Why does the universe expand? So far, the data collected by astronomers hasn’t been able to yield an answer. The astronomers could be missing a key insight in the data they’ve gathered so far, or their hypotheses about how the universe is expanding could be fundamentally wrong. Either way, AI may help astronomers find overlooked patterns in the existing data that shed light on universal expansion or point them to an unexplored region of the sky that could finally provide the breakthrough they need. In the process, an AI-assisted insight could transform our theories of the Big Bang. “We don’t understand the physics of this expansion,” Gebhardt says. “We don’t even understand how gravity works. So hopefully we can use AI to answer some of these questions and better understand the universe at a fundamental level.”
