These natural telescopes let astronomers study objects too faint or distant to see otherwise. They can magnify galaxies from the early universe, map the distribution of invisible dark matter, sharpen measurements of the universe's expansion rate, and even help detect exoplanets.

How did AI narrow the field?

The map's sheer size made visual searching impractical. The team turned to a residual neural network — a type of machine learning designed to spot subtle patterns in images — to identify the distinctive signatures of lensing. Earlier work with these techniques had produced a catalog of roughly 3,500 potential lenses; the latest AI-powered search uncovered 76 new candidates.

"The DESI Legacy Imaging Surveys' deep and wide-field images have provided an unprecedented foundation for discovering new gravitational lenses," says Aleksandar Cikota, an associate scientist at NSF NOIRLab, study co-author and principal investigator of the follow-up observing program. "What we're building now is a carefully confirmed sample that researchers can use for years to come."

The searches also depended on advanced computing infrastructure from the National Energy Research Scientific Computing Center (NERSC), a DOE Office of Science user facility — a reminder that modern astronomy runs on powerful computation as well as telescopes.

How were the lenses confirmed?

Finding candidates was only the first step. To prove an object is a genuine gravitational lens, astronomers must measure the distances to both the lensing object and the background source. That requires spectroscopy — splitting light into its component wavelengths to reveal an object's physical properties and distance.

MUSE proved especially well suited to the task. It uses integral field spectroscopy, which records a spectrum at every point across an image rather than from a narrow region, producing a three-dimensional data set. This allowed the researchers to study the lensing object, the background object and nearby companions in a single observation, while its broad wavelength coverage revealed faint spectral features. Of the 76 candidates, 70 proved to be true gravitational lenses.

The follow-up study was led by Emerald Lin, an undergraduate student at UC Berkeley, and Ivonne Toro Bertolla, a research assistant at NSF NOIRLab and science operations assistant at Las Campanas Observatory.

Why do these lenses matter?

The newly confirmed lenses have joined the DESI Strong Lens Foundry Project, one of the largest confirmed-lens collections to date. Combined with observations from the Hubble Space Telescope, the James Webb Space Telescope, the recently launched Nancy Grace Roman Space Telescope, and large surveys such as the Vera C. Rubin Observatory's Legacy Survey of Space and Time, the sample will help researchers probe dark matter's distribution, study distant galaxies and test models of cosmic evolution.

"The combination of large surveys and AI-powered searches is transforming how we discover rare objects in the sky," says Cikota. "These tools allow us to systematically search for systems that, previously, only a few years ago, we might have discovered only serendipitously."

The project also demonstrates the lasting value of the DESI Legacy Imaging Surveys, which began as a campaign to build a 3D map of the universe. "Each confirmed lens adds to a resource that will benefit the entire astronomical community," Cikota says. "Together, we are building a powerful foundation for future discoveries."