A team of astronomers led by an undergraduate student in Texas has discovered two planets orbiting stars more than 1,200 light-years from Earth.
Astronomers already knew of about 4,000 exoplanets, so finding two more might not seem like huge news. But it's who found them and how that's getting attention.
Anne Dattilo, a senior at the University of Texas, Austin, found the planets by using an artificial intelligence program to sift through a mountain of data collected by NASA's Kepler space telescope. By using AI, the 22-year-old is helping to usher in a new era in astronomical research.
Dattilo got involved in the project about a year and a half ago, after astronomer Andrew Vanderburg gave a talk during one of her classes. Vanderburg uses data from Kepler to hunt for planets orbiting distant stars.
"And at the very end, he said, 'I'm taking undergrads if any of you want to do research on this subject finding planets,' " Dattilo recalls. "I decided that's what I wanted to do, so I emailed him."
Kepler, which launched in 2009, was designed to point at a small patch of sky and measure the light from some 100,000 stars in its field of view. If there is a planet orbiting one of the stars, the light coming from that star will dim slightly when the planet passes in front of it.
Toward the end of Kepler's life, mechanical failures meant it couldn't measure the light quite as accurately, so the data it collected were harder to interpret.
Dattilo modified an artificial intelligence program called AstroNet-K2 to work on data from the latter part of the Kepler mission. Once the modified program found stars that appeared to have planets, Dattilo and her colleagues used ground-based telescopes to confirm the findings.
The newly discovered planets, officially named K2-293b and K2-294b, orbit stars in the constellation Aquarius and are both slightly larger than Earth.
"If we want to know how many planets there are in total, we have to know how many planets we've found," Vanderburg said in a statement from the University of Texas McDonald Observatory. "But we also have to know how many planets we missed. That's where this [AI] comes in."
Vanderburg and Google engineer Chris Shallue "first used AI to uncover a planet around a Kepler star" in 2017, according to the statement.
Jessie Christiansen, a research scientist with NASA's Exoplanet Science Institute at the California Institute of Technology, says she's impressed by what Dattilo's team accomplished. And in a way, she's not all that surprised.
"NASA makes all of the data publicly available," Christiansen says. "You just have to think of a new idea of what to do with the data that no one has done before."
Michelle Ntampaka of the Harvard-Smithsonian Center for Astrophysics expects to see lots more astronomers using artificial intelligence techniques to analyze their data in the future. That's because newer telescopes don't so much collect images of stars and galaxies as digital data about these celestial objects.
"We're just going to see unprecedented data volumes, and we're going to have to come up with new ways to deal with that," Ntampaka says.
So writing a machine learning program as an undergrad is good preparation for Dattilo as she heads off to get her graduate degree in astronomy.
The discoveries have been accepted for publication in The Astronomical Journal.
STEVE INSKEEP, HOST:
Astronomers know of about 4,000 planets orbiting stars outside our solar system. Now they know of two more, thanks to an undergraduate college student using artificial intelligence. Here's NPR's Joe Palca.
JOE PALCA, BYLINE: Anne Dattilo is a senior at The University of Texas at Austin. Last year, an astronomer talked to her class about his research using a NASA satellite called Kepler to hunt for planets orbiting distant stars.
ANNE DATTILO: And at the very end, he was like, I'm taking undergrads, if any of you want to do research on the subject, finding planets, and I decided that's what I wanted to do. So I emailed him, and a year and a half later, here I am.
PALCA: She led a team that discovered two Earth-sized planets orbiting stars more than 1,200 light-years from Earth. To find the planets, Dattilo used an artificial intelligence approach called machine learning to comb through a Kepler data set called K2; K2 contains measurements of the light coming from tens of thousands of stars. Dattilo says when a star is what she calls boring, the light coming from it is constant.
DATTILO: But if you can imagine something passing in front of that star, the light we receive would dim. And so if you see that periodically, that would be a signal that a planet is in front of that.
PALCA: The artificial intelligence program looks for these fluctuations in a star's light that might be associated with a planet passing in front. Now, you don't have to be a NASA scientist to use data from a NASA satellite.
JESSIE CHRISTIANSEN: NASA makes all of the data publicly available. You just have to think of a new idea to do with the data that no one's done before.
PALCA: Jessie Christiansen is a research scientist at the NASA Exoplanet Science Institute at Caltech in Pasadena.
CHRISTIANSEN: This is the first time someone's gone through and done a machine learning process on the K2 data to come up with a uniform list of planet candidates.
PALCA: And that will be valuable beyond just getting a good grade on an undergraduate class?
PALCA: In fact, Michelle Ntampaka at the Harvard-Smithsonian Institute for Astrophysics in Cambridge says she's seen something remarkable happen in the last five years or so.
MICHELLE NTAMPAKA: And that is that there has been a dramatic increase in the amount of machine learning research that's happening for astronomy applications.
PALCA: That's because newer telescopes don't so much collect images of stars and galaxies as digital data about these celestial objects.
NTAMPAKA: We're just going to see unprecedented data volumes, and we're going to have to come up with new ways to deal with that.
PALCA: Ntampaka says the next generation of astronomers will have to be comfortable working with artificial intelligence to make sense of all this data. So writing a machine learning program as an undergrad is good preparation for Anne Dattilo as she heads off to get her graduate degree in astronomy. Joe Palca, NPR News.
(SOUNDBITE OF FYZE'S "WANDERER") Transcript provided by NPR, Copyright NPR.