AI Discovers Over 10,000 New Exoplanet Candidates in NASA Data

Artificial intelligence has identified 10,091 new exoplanet candidates within existing NASA telescope data, marking the largest single discovery of potential alien worlds to date. The findings, processed through machine learning models, include a subset of over 100 rare and extreme worlds that had previously remained hidden from researchers.

The discovery relies on the application of AI to analyze data from NASA’s Transiting Exoplanet Survey Satellite (TESS). TESS identifies exoplanets by monitoring the brightness of stars. when a planet passes between a star and the telescope, it creates a slight dip in light known as a transit. While the TESS mission generates massive amounts of data, distinguishing true planetary transits from stellar noise or other astronomical phenomena has historically required extensive human vetting.

Automating the Search for Alien Worlds

The scale of the new candidate list is a result of AI’s ability to filter through thousands of light curves more efficiently than previous algorithmic pipelines. The machine learning models were trained to recognize the specific signature of a planetary transit while ignoring false positives, such as eclipsing binary stars or instrumental glitches that can mimic the appearance of a planet.

By refining the vetting process, the AI identified 10,091 candidates that were previously overlooked or categorized as noise. This surge in candidates potentially triples the number of known alien worlds if a significant portion of the list is confirmed through follow-up observations.

Among these candidates, researchers highlighted more than 100 worlds described as rare or extreme. These include planets with unusual orbital periods, extreme temperatures, or sizes that challenge current models of planetary formation.

The Challenge of False Positives

A primary hurdle in exoplanet research is the high rate of false positives. Many signals that appear to be planets are actually caused by stellar variability or the interaction of two stars. The AI used in this study was specifically designed to solve this bottleneck by identifying subtle patterns in the transit data that human observers or simpler software might miss.

Some of the identified candidates are described as impossible based on previous detection limitations, meaning they were either too small or their orbits too eccentric for standard pipelines to flag as viable planets.

“I’m really excited for the future of the field”

Researcher cited via IFLScience

Verification and Next Steps

While the AI has flagged 10,091 candidates, these objects are not yet confirmed planets. In astronomy, a candidate is a signal that strongly suggests a planet’s presence but requires independent verification.

The next phase of the research involves using other detection methods to confirm the existence of these worlds, including:

• Radial Velocity: Measuring the “wobble” of a star caused by the gravitational pull of an orbiting planet.
• Direct Imaging: Attempting to capture actual light from the planet using high-contrast imaging.
• Additional Transit Observations: Using other telescopes to verify that the light dips occur at regular, predictable intervals.

The integration of AI into NASA’s data pipelines demonstrates a shift toward computer-modeling-driven discovery, allowing astronomers to extract maximum value from existing satellite missions without requiring the launch of new hardware to find thousands of new worlds.