UC Berkeley AI Model Sparks New Theory in Exoplanet Detection
Astronomers at UC Berkeley use machine learning to analyze gravitational microlensing events, inadvertently discovering a new unified theory for observing far-off worlds. The AI unexpectedly identifies lensing anomalies that challenge previously held binary explanations.
Machine learning models continue to transform scientific research by accelerating repetitive tasks and offering unexpected systemic insights. Astronomers at UC Berkeley experience this firsthand when they apply artificial intelligence to gravitational microlensing events, a phenomenon where light from distant stars bends around a nearer object to provide a distorted but brighter view of the far-off world.
Researchers train a machine learning model on known gravity microlensing data to quickly categorize these celestial events as sky surveys grow more detailed. Traditionally, scientists classify lensing anomalies, known as degeneracies, into two main categories depending on whether the background star passes close to the foreground star or its planet. However, the AI model surprises the team by calculating events that fit into these established types while simultaneously identifying hundreds of anomalies that match neither theory.
This unexpected discovery leads the research team to propose a new unified theory for gravitational microlensing that accounts for situations where the background star does not pass close to either the foreground star or planet. By revealing patterns that previously eluded human observation, the AI system actively reshapes the fundamental understanding of how astronomers detect and quantify distant exoplanets.