The newly discovered exoplanets, or planets outside of the earth’s solar system, were found after researchers applied the same AI techniques that help computers recognize images like cats in photos to data gathered from the Kepler space telescope.

NASA launched Kepler in 2009 to discover new planets orbiting other stars, Jessie Dotson, a Kepler project scientist at NASA’s Ames Research Center, said during a media briefing on Thursday. Prior to 2009, she said that NASA only knew of 326 exoplanets, but that Kepler has helped the agency discover 2,500 more.

In talking about the two new planets, NASA focused less on Kepler-80g and more on Kepler-90i because it was found to be the eighth planet orbiting the only star in its solar system. That’s significant because it shows that this particular solar system somewhat mirrors our own in which eight planets orbit a sun, the researchers said.

Get Data Sheet, Fortune’s technology newsletter.

To discover exoplanets, astronomers must comb through data collected by Kepler and identify “signals” that could indicate possible planets. Although researchers currently have software that helps spot “strong signals” so that humans can then investigate, there are many “weak signals” that are not followed up on.

Through a partnership with Google (GOOG), NASA applied so-called neural networks to help parse through the Kepler data and spot possible exoplanets that humans may have overlooked.

NASA

Neural networks are essentially software designed to loosely mimic how the human brain learns, explained Christopher Shallue, a Google senior AI software engineer. To train these neural networks to recognize images of cats in photos, for example, Shallue said Google fed neural networks enough cat photos so that the software eventually could discover cats in new photos on its own based on patterns it discovered.

Instead of feeding NASA’s neural network cat photos, however, the researchers fed it “15,000 signals” taken from the Kepler data that scientists confirmed were from exoplanets. This so-called “training” period, which took two hours, helped the neural network discover unseen patterns that it then used to discover exoplanets when the researchers fed it new Kepler data.

The researchers used the neural network to parse Kepler image data taken from 670 stars, and the neural network then indicated that two “weak signals” were likely to be planets.

“Even though the signals were weak, the model was convinced,” said Shallue of the software’s confidence in its accuracy. Using more traditional methods, NASA said it was able to verify that the neural network’s picks were indeed exoplanets.

“The key contribution of machine learning here was that it was able to search much larger number of signals than humans would have been able to do within a reasonable amount of time,” Shallue said.

Despite the noteworthy use of artificial intelligence to speed up the planet discovery process, humans are still needed, the researchers explained. NASA astronomer Andrew Vanderburg said that citizen scientists that independently help NASA find planets excel at “finding things that are weird” in the data and that “neural networks struggle with.”

Some anomalies that the neural networks may have not recognized may, in fact, turn out to be planets.

Shallue said that Google plans to release the software used in this project to the public for free. Doing so could spark interest among machine learning practitioners to become citizen scientists, “or encourage citizen scientists to use machine learning in their efforts as well,” he said.