Google's DeepMind is using AI to Hear Deeper Into the Universe
Imagine LIGO as the universe’s most sensitive "listening device." Now, picture that device quieting its own background noise to hear the deepest, most elusive cosmic signals.
That’s exactly what Google DeepMind, LIGO, and GSSI have achieved with their Deep Loop Shaping method. By using AI to clean up LIGO’s control system, they've made the detector 30–100 times more sensitive in a critical frequency range of 10–30 Hz, where signals from heavier black hole mergers and the early moments of neutron star collisions begin.
The Big Idea: Quieter Control, Better Data
LIGO is designed to detect the tiniest ripples in spacetime, but every time it tries to listen for these signals, even the smallest vibrations can throw off its measurements. These unwanted vibrations, known as control noise, originate from its own feedback system, which is supposed to keep the mirrors perfectly still. This issue is similar to a waterbed effect: when one part of the system dampens a vibration, another part inadvertently boosts it, making it harder to get accurate readings.
This is where Deep Loop Shaping comes in. Instead of just throwing more feedback into the system, DeepMind’s AI has learned how to stabilize the mirrors without adding any noise. In tests, it reduced that control noise by up to 100 times in the 10–30 Hz range—the sweet spot where scientists look for massive black hole mergers and early neutron star collisions.
Studying the universe using gravity instead of light, is like listening instead of looking. This work allows us to tune into the bass.
Rana Adhikari, Professor of Physics, Caltech (2025)
Why This Matters: More Data, Better Science
The impact of the AI is significant. By minimizing internal noise, LIGO can now detect more events, such as intermediate-mass black holes, that were too faint to observe before. With less distortion and clearer data, astronomers are now able to explore areas of the universe that were previously beyond their reach.
This breakthrough could help answer some of the most profound questions about how galaxies form, how black holes grow, and what happens in the extreme environments of space.
The Technology Behind It: Reinforcement Learning in Action
The AI uses a technique called reinforcement learning. Basically, it’s taught how to adjust LIGO’s control system so that the mirrors remain steady, without causing any noise in the frequency bands where LIGO listens. It does this by responding to frequency-domain rewards (basically, a way of measuring how much noise is reduced).
The AI was initially trained in simulated environments before being tested on the actual LIGO system in Livingston, Louisiana. The outcome? Not only did the AI perform well, but it also surpassed all expectations, managing to reduce noise by as much as 100 times in some of the most challenging feedback loops that LIGO faces.
What’s Next?
This technology goes beyond just detecting gravitational waves. The Deep Loop Shaping method has exciting possibilities in various fields where managing vibrations and reducing noise are essential. Consider areas like aerospace, robotics, or structural engineering. In any situation that demands exceptional stability in constantly changing environments, this method could truly make a significant impact.
Further reading: DeepMind blog (project + numbers), Science paper (methods/DOI), Physics World (low-frequency stakes).
Y. Anush Reddy is a contributor to this blog.
