AI Found an Alien Solar System With As Many Planets As Our Own

On Thursday, NASA announced the highly anticipated results of its partnership with Google to apply machine learning to data returned from the Kepler space telescope. According to NASA, Google’s machine learning algorithms found an eighth exoplanet in an extrasolar system known as Kepler-90, which means that our own solar system is now tied for the most known planets orbiting a single star.

The new star is known as Kepler-90i and is the smallest known planet in the Kepler-90 system, which is about 2,545 light years away from Earth. Kepler-90i orbits its host star once every 14 days and is likely a rocky planet with surface temperatures of around 800℉.

The Kepler-90 host star is slightly more massive than our Sun, and although our solar systems share the same number of planets, they differ in a few key respects. For starters, all eight planets in Kepler-90 are closer to their host star than the Earth is to the Sun. Moreover, all the planets in the Kepler-90 system are larger than Earth.

On the flipside, the arrangement of the planets around Kepler-90 bears a striking resemblance to our own solar system—smaller rocky planets are closest to the host star, while larger Jupiter-like gas giants orbit on the outskirts.

Launched into an Earth-trailing orbit in 2009, the main mission of the Kepler space telescope is to scan the cosmos for traces of planets orbiting other stars, known as exoplanets. Since the first exoplanet was discovered in 1992, a total of 3,720 exoplanets have been documented by astronomers in 2,787 solar systems. Of these solar systems, however, only 623 are known to host more than one planet.

Astronomers have a number of tools at their disposal to look for exoplanets and most of them don’t involve ‘seeing’ the planet directly. The most common technique, and the one used by the Kepler mission, is known as transit photometry. This is essentially looking for the shadow of a planet as it passes in front of its host star. Astronomers don’t actually see the shadow, of course, but detect the planet’s presence by capturing a decrease in the amount of light emitted from the star as the planet passes between the star and telescope.

Using this technique, astronomers have detected around 2,500 exoplanets orbiting the 150,000 stars studied by Kepler. Yet when scientists took this massive trove of Kepler data gleaned from observing these stars for four straight years, they mostly focused on the 30,000 strongest signals (that is, data that registers as a decrease in light) received by Kepler. These were the most likely to point to exoplanets.

The search was limited to the strongest signals by necessity. The amount of data generated by Kepler during its first four years is far too large to be analyzed effectively by humans. In other words, there was a very likely possibility that some exoplanets were hiding among the other 120,000 weaker signals that are waiting to be discovered.

To find these exoplanets, NASA partnered with Google to train a neural network, a type of machine learning algorithm loosely modeled on the human brain, to recognize exoplanets among the weaker signals in the Kepler data. Google researchers trained the network on 15,000 examples of exoplanet data that had been labeled by NASA researchers. This way, the neural net could learn what the presence of an exoplanet ‘looks like’ as data. By the time the researchers were done training their algorithm, it could spot exoplanets in a dataset with 96 percent accuracy.

According to Christopher Shallue, a senior software engineer at Google involved in the project, the network was then used to study data from 670 stars. Among these weaker signals, it found two exoplanets—the one orbiting Kepler-90 and another in a solar system called Kepler 80.

“It’s very possible that Kepler-90 has more planets we don’t know about,” Shallue said in the NASA teleconference announcing the discovery. “There’s a lot of unexplored real estate in the Kepler-90 system. It would almost surprise me if there weren’t more planets.”

As explained by Shallue, the application of machine learning to the analysis of Kepler data was a proof of concept to see if an algorithm really could help discover exoplanets that would otherwise have been missed by human researchers. Going forward, he hopes to use the algorithm to explore the data from the tens of thousands of other stars observed by Kepler.

This will likely cause the number of known exoplanets to increase far more rapidly than it has before. Moreover, two new space telescopes—the James Webb telescope and TESS—will be launching in the next two years, so there will soon be a deluge of new exoplanet data to analyze.

And who knows—one of those planets may be home to inhabitants using their own computers to study our Sun.

Get six of our favorite Motherboard stories every day by signing up for our newsletter .