The eyes will help us understand if we are looking at a person who does not really exist!
See also: Top 10 jobs in technology for 2022
Nowhere is this phenomenon more troubling than in the strange world of "deepfakes" and other computer-generated faces that do not really exist.
Scientists have now developed a technique that could help us detect if the people we are looking at are really real people, as opposed to the "ghosts" created by artificial intelligence (AI).
According to a new study led by the first author and computer scientist Hui Guo from the State University of New York, the secret is in the eyes - specifically in the shapes of the daughter.
In the magnification of the artificial eyes of fake faces created by a machine learning system called generative adversarial network (GAN), the researchers noticed something funny in the pupils.
Unlike real daughters, many of the fake ones were not actually round.
"Daughters have almost circular shapes," the team explains in the study.
"Compared to real persons, we observe that visible artifacts and inconsistencies can be observed in the eye areas of persons created by GAN."
See also: Can AI score your next test?
According to the researchers, this strange phenomenon is due to the fact that GAN models lack the understanding of the anatomy of the human eye, especially in terms of the geometric shapes of normal pupils.
To investigate the prevalence of this phenomenon, the researchers developed a detection tool that automatically extracts the contours of the pupils from the eyes in the photographs and then evaluates them to check if they have elliptical shapes.
In an experiment that ran the tool in a database with 2.000 images (1.000 were real people and 1.000 were fake), the system worked reliably to distinguish the two groups.
According to the team, such technology could one day help tackle the malicious use of fake realistic faces used to deceive people on platforms. social media, including.
At least until someone can teach artificial intelligence the right shape of an eye pupil, of course.
Source of information: sciencealert.com