AI Discovers 10,000 Hidden Planets in NASA Telescope Data

The Largest Exoplanet Discovery in Decades

In a groundbreaking achievement that could reshape our understanding of planetary abundance in the galaxy, researchers have used machine learning to identify over 10,000 previously unknown exoplanet candidates in a single survey, uncovering exactly 10,091 candidate planets that had never been seen before . The discovery represents more than were detected in the entirety of NASA's Kepler mission and its follow-on K2 and more than double the existing planet candidates from TESS that await confirmation .

This find could more than double the number of worlds currently known to science, and the sheer volume suggests that our galaxy is far more crowded than previously confirmed . Currently, astronomers have confirmed the existence of just under 6,300 exoplanets , making this potential addition monumental for planetary science.

Revolutionary AI Technique Uncovers Hidden Worlds

The breakthrough came from using machine learning to survey over 83 million stars that were observed during TESS' first year of observations, revealing 10,091 appeared to have transiting, planet-like objects never seen before . Unlike traditional searches that focus on bright stars, this survey analyzed stars 16 times fainter than those typically targeted by TESS, deliberately shifting attention toward much fainter targets .

The algorithm was trained to distinguish the specific "U-shaped" dip of a planetary transit from the erratic fluctuations of a star's natural activity, learning to distinguish subtle clues that a transit had potentially occurred . This is a classic data-mining success story: the information was already there, stored in NASA's archives; we just needed a more efficient way to read it .

First Confirmation Validates the Method

To test their approach, researchers confirmed one candidate from this batch, revealing that the planet, named TIC 183374187, is a gas giant with a mass similar to Jupiter and is what astronomers call a hot Jupiter, because it orbits close to its star, making it blisteringly hot . The confirmation of TIC 183374187 b hints that at least a few of the other exoplanet candidates will also end up being confirmed, though first these planets must be verified by independent surveys and studied in greater detail, which can take months or years to do properly .

According to the paper, 97.7% of the planets discovered were gas giant-like planets, as bigger planets are easier to see and planets with shorter orbits were also easier to detect because they caused more frequent dips in the light curves .

Transforming the Future of Planet Hunting

These studies highlight how advances in artificial intelligence are transforming astronomy, as combining massive datasets with machine learning allows researchers to uncover new planets while also improving the tools themselves through challenging real-world data . This demonstrates that entire populations of celestial objects may be hiding not in distant galaxies, but in datasets already sitting on Earth, waiting for better tools to unlock them .

The timing is particularly significant as NASA is preparing the upcoming Nancy Grace Roman Space Telescope, currently scheduled for launch by May 2027, which will employ a different technique called gravitational microlensing . This next generation of space telescopes promises to move beyond simply detecting planets to characterizing their atmospheres and potential habitability. As AI continues to revolutionize how we analyze cosmic data, we may be on the verge of answering one of humanity's most profound questions: just how common are planets like Earth in our galaxy?