Scientists identify 11,554 exoplanet candidates using machine learning
Updated
Updated · Livescience.com · May 2
Scientists identify 11,554 exoplanet candidates using machine learning
15 articles · Updated · Livescience.com · May 2
The survey found 10,052 previously unseen candidates in TESS data from 83,717,159 stars and confirmed one hot Jupiter, TIC 183374187 b, with a Magellan telescope in Chile.
If verified, the haul would lift the known exoplanet total to nearly 18,000, almost triple today's count, after the algorithm detected faint transit signals usually missed in dim stars.
Most candidates orbit every 0.5 to 27 days, suggesting they are probably too close to their stars for life, and independent confirmation could still take months or years.
AI flagged 10,000 potential planets, but half may be fake. What is the biggest hurdle to proving a candidate is a real world?
Most new candidates are uninhabitable 'Hot Jupiters'. Why is this discovery still a revolutionary step in the search for alien worlds?
How will AI and new telescopes like the Roman change our fundamental understanding of how solar systems like our own are formed?
Unveiling 11,554 New Worlds: How T16 and Machine Learning Revolutionize Exoplanet Discovery
Overview
In April 2026, the T16 Planet Hunt team led by Joshua T. Roth tripled the number of known exoplanet candidates by identifying 11,554 new worlds using machine learning to analyze light from over 83 million stars, including those 16 times dimmer than before. This breakthrough enabled large-scale studies of planetary populations and revealed unusual planets with ultra-short orbits and those in the Neptunian Desert, challenging existing theories and suggesting Earth-like habitable planets may be rarer. Advanced AI tools like RAVEN and ExoMiner++ play a crucial role in validating these candidates, while future telescopes will characterize their atmospheres, paving the way for a deeper understanding of planetary system formation across the galaxy.