As mankind’s understanding of this world develops, technology also evolves. Facial recognition AI is one of the most controversial aspects of technology as a whole. While the implementation of it in public spaces has raised some eyebrows, in mobile phones it is a highly revered feature.
Well, the thing is that facial recognition depends on photos that are uploaded online first to identify us. As we grow older the face changes, we get wrinkles and skin gets loose. Thus, AI starts struggling to identify us if our pictures are outdated.
While you might be thinking oh it must take years upon years for the algorithm to start to have trouble recognizing us, the thing is that it only takes 5 years for the struggle to start and after 10 or 20 years the AI has no chance of ever recognizing us if our pictures aren’t updated.
To test this a team working at the Norwegian University of Science and Technology led by Ph.D. candidate Marcel Grimmer conducted a test where they created almost 50,000 faces using AI and then synthetically aged them and tested AI on them.
Because commercial systems do not share how their system works, the team had to use open-source recognition systems as it was their second best (available) option. The research found out that it became harder for a system to recognize the person as age increased. According to the result, if a person had exceeded 20 years of their photo being updated then the chances of them being recognized by the algorithm are almost zero. Unless you are Tom Cruise, cause that man does not age.
The age of a person also had a huge impact as for people who were under the age of 20 or over the age of 60, the system had a very hard time recognizing them. As for the former, when a person is in their late teens or early ones they tend to change a lot physically, so it seems logical that an algorithm does not recognize them. Moving on to the latter, the head shape changes, and a person’s wrinkles become more defined, so the failure rates are high there as well, according to Grimmer.
Read next: Research Shows How Much Free Android Apps Siphon User Data
Well, the thing is that facial recognition depends on photos that are uploaded online first to identify us. As we grow older the face changes, we get wrinkles and skin gets loose. Thus, AI starts struggling to identify us if our pictures are outdated.
While you might be thinking oh it must take years upon years for the algorithm to start to have trouble recognizing us, the thing is that it only takes 5 years for the struggle to start and after 10 or 20 years the AI has no chance of ever recognizing us if our pictures aren’t updated.
To test this a team working at the Norwegian University of Science and Technology led by Ph.D. candidate Marcel Grimmer conducted a test where they created almost 50,000 faces using AI and then synthetically aged them and tested AI on them.
Because commercial systems do not share how their system works, the team had to use open-source recognition systems as it was their second best (available) option. The research found out that it became harder for a system to recognize the person as age increased. According to the result, if a person had exceeded 20 years of their photo being updated then the chances of them being recognized by the algorithm are almost zero. Unless you are Tom Cruise, cause that man does not age.
The age of a person also had a huge impact as for people who were under the age of 20 or over the age of 60, the system had a very hard time recognizing them. As for the former, when a person is in their late teens or early ones they tend to change a lot physically, so it seems logical that an algorithm does not recognize them. Moving on to the latter, the head shape changes, and a person’s wrinkles become more defined, so the failure rates are high there as well, according to Grimmer.
Read next: Research Shows How Much Free Android Apps Siphon User Data
