Gab Ibis

The human eyes can reveal a lot about a person’s personality, including their demeanor, emotions, and reactions — authenticity of a person, if one can say.

Now, it seems our eyes may also hold the key to unmasking digital deception.


A groundbreaking technique from the Royal Astronomical Society’s (RAS) National Astronomy Meeting revealed that light reflection in eyes can detect whether an image is artificial intelligence (AI)-generated or not.

Led by University of Hull student Adejumoke Owolabi and astrophysics professor Kevin Pimbblet, they adapted tools used by astronomers in galaxies study to examine the consistency of light reflections in eyeballs.

While AI has made remarkable strides, it often falls short when it comes to replicating the intricate details of human physiology, including the precise nuances of eye reflections.

Hence, ease in detecting falsely attributed images and AI-generated ones.

In a post with RAS, Pimbblet said that the technique used automatically detected eyeball reflections and compared the reflections’ shapes between the left and right eyes. 

"To measure the shapes of galaxies, we analyze whether they're centrally compact, whether they're symmetric, and how smooth they are. We analyze the light distribution,” Pimbblet said.

The pair also used astronomy methods to measure and compare eyeball reflections. They applied the Gini coefficient, which is usually used to study light distribution in images of distant space objects, to evaluate the evenness of reflections across eye pixels.

A Gini value close to 0 means the light is evenly spread, while a value near 1 indicates the light is concentrated in one spot.

A similar method of concentration, asymmetry, and smoothness (CAS) was aso used by the pair to detect AI-generated images. However, it was not as successful in detecting fake eyes.

“It's important to note that this is not a silver bullet for detecting fake images,” Pimbblet added.

The method shows promise, but there are other factors considered in detecting fake images, including image complexity, detail, consistency, and other variables that makes an image “real.”

"There are false positives and false negatives; it's not going to get everything. But this method provides us with a basis, a plan of attack, in the arms race to detect deepfakes,” Pimbblet said.