An international research team has opened a way to clearly observe celestial events such as black hole mergers and neutron star collisions using artificial intelligence (AI).
An international research team including Google DeepMind and the California Institute of Technology (Caltech) has dramatically improved the performance of gravitational wave detectors using AI-based control technology, it was announced in the international journal Science on the 5th.
Gravitational waves refer to the gravitational energy spreading like waves when massive events occur in the universe, such as star explosions or black hole formations. By enhancing the signal sensitivity in the 10 to 30 Hz range, it is possible to detect mid-mass black hole mergers or neutron star collisions earlier.
However, gravitational wave signals can easily be interfered with by earthquakes, winds, temperatures, or equipment. Notably, the Laser Interferometer Gravitational-Wave Observatory (LIGO), which detected gravitational waves for the first time in history, precisely controls the angle of mirrors to detect signals, but noise from the control system interfered with signal detection. LIGO installs mirrors at both ends of a vacuum tunnel several kilometers apart and fires lasers. If a gravitational wave passes while the laser is traveling, the mirror moves slightly.
To solve this, the research team introduced an AI technique based on reinforcement learning called 'Deep loop shaping.' Reinforcement learning is a method where AI interacts with surrounding elements to learn what is effective. It is similar to training a puppy by praising or rewarding specific behaviors rather than merely teaching them a certain behavior.
In this study, the AI learned how to control LIGO's mirrors and minimized noise in the 10 to 30 Hz band. As a result of applying AI techniques to LIGO, the noise generated during the control process was reduced to one-thirtieth to one-hundredth of previous levels.
The research team noted, 'If sensitivity in the low-frequency range improves, we can predict neutron star collisions twice as quickly, giving astronomers the chance to observe cosmic explosions in real time alongside light and neutrinos,' adding, 'This also makes it possible to explore unknown events such as mid-mass black hole mergers.'
References
Science (2025), DOI: https://doi.org/10.1126/science.adw1291