Swiss AI identifies 44 star systems hiding potentially habitable exoplanets with 99% accuracy

Scientists from the University of Bern (Switzerland) using artificial intelligence have discovered 44 star systems that may harbor Earth-like exoplanets. The findings are based on a machine learning algorithm that analyzes data from synthetic planetary systems. While the model did not confirm the direct existence of these objects, it narrowed the search for astronomers by pointing to the most promising targets.

The main difficulty in detecting exoplanets is due to their small size and faint emission. The known 5800 confirmed objects provide insufficient data to train the AI, so the Swiss team used synthetic systems created by the Bern Model of Planet Formation and Evolution. This tool simulates the evolution of exoplanets starting from the protoplanetary disk and takes into account dozens of parameters, from gravitational interaction to chemical composition.

Illustration by Leonardo

“The Bern model is one of the few that reproduces interconnected processes with such detail. This allowed us to train the algorithm on realistic but hypothetical systems,” explained Dr. Jan Alibert of the Space Habitability Center.

The AI revealed that key indicators of hidden Earth-like planets are the mass and orbital period of the innermost known planet in the system. Applying these criteria to 1600 systems with stars of classes G, K and M (Sun analogs, orange and red dwarfs), the algorithm identified 44 targets with an accuracy of 0.99. This means that in 99% of cases there may indeed be at least one exoplanet in the habitable zone.

However, the model has limitations. For example, it does not reproduce the frequent neighborhood of “super-Earths” (planets with masses larger than Earth but smaller than gas giants) with “cold Jupiters” (gas giants in distant orbits) in real systems. In addition, the synthetic planets were found to be closer to their stars than their observed counterparts.

“Perfection is not required here,” Alibert noted. – Even partially accurate predictions save years of observations. We now have 44 priority areas for the search for life.”

The next step is to verify the results with next-generation telescopes such as Europe’s extremely large ELT telescope and the Plato Space Observatory, scheduled for launch in 2026.