How AI Quietly Fixed The James Webb Space Telescope’s Blurry Vision

From left to right: Images of the galaxy NGC 1068, Jupiter’s moon Io and Wolf-Rayet star 137, before (top) and after (bottom) AI sharpening. (Image credit: Max Charles/University of Sydney)

The James Webb Space Telescope, humanity’s most powerful eye on the cosmos, has been quietly suffering from a case of blurry vision. But instead of astronauts flying a billion-dollar repair mission, a group of Australian researchers fixed it with artificial intelligence – and a few clever lines of code.

The issue cropped up in one of Webb’s smaller instruments, the Aperture Masking Interferometer (API), which rides aboard the Near-InfraRed Imager and Slitless Spectrograph (NIRISS). Designed by Professor Peter Tuthill’s team at the University of Sydney, the API helps Webb hunt for dim exoplanets orbiting bright stars by combining light from different parts of the telescope’s giant mirror. But when astronomers first activated it, the images came back fuzzy – disappointingly out of focus.

It wasn’t a mechanical flaw this time, like the one that famously plagued Hubble in 1990. Back then, NASA had to send astronauts 320 miles above Earth to mount corrective optics on the telescope. Webb, however, sits about 930,000 miles away at the Sun-Earth L2 point – well beyond the reach of any rescue mission. So, when the API started producing distorted data, scientists knew they’d have to solve it remotely.

The culprit turned out to be subtle electrical distortions in the telescope’s infrared camera detector, which warped how light signals were being recorded. To fix it, two former Sydney Ph.D. students, Max Charles and Louis Desdoigts, built a neural network capable of spotting and correcting those flawed pixels automatically. They called their creation AMIGO – short for Aperture Masking Interferometry Generative Observations.

AMIGO works by recognizing patterns of interference caused by the detector’s electronics, then reconstructing what the image should look like. It’s a digital version of cleaning a smudged lens – but on a cosmic scale. “Instead of sending astronauts to bolt on new parts, they managed to fix things with code,” Tuthill explained in a statement from the university.

When tested on real data, the improvement was dramatic. AMIGO brought sharper clarity to images of a faint exoplanet and a cool, low-mass red-brown dwarf 133 light-years away. In other runs, it helped Webb capture the structure of a black hole jet, volcanic activity on Jupiter’s moon Io, and the gusty winds from a variable star.

“This work brings JWST’s vision into even sharper focus,” said Desdoigts, now at Leiden University. “It’s incredibly rewarding to see software extend the telescope’s scientific reach.”

Since becoming operational in 2022, the James Webb Space Telescope has already rewritten textbooks – revealing ancient galaxies, unexpected black hole activity, and the chemistry of distant worlds. Now, with its AI “glasses” in place, Webb’s view of the universe just got even clearer, proving that sometimes the biggest fixes in space don’t come from rockets, but from algorithms.