Scientists from the Ruhr-Universität Bochum Institute for Neural Computing (RUB) in Bochum have designed an artificial intelligence (AI) algorithm that can successfully estimate people’s age and ethnicity, Techxplore reported.
Neuroinformatics engineers, however, do not yet know what features AI interprets, although they are fairly specific.
The system can make evaluations.
“We don’t know exactly what features the algorithm is looking for,” said Laurenz Wiskott, professor at Techxplore, at the Institute for Neural Computation.
The system is an eleven-level hierarchical neural network that has been fueled by several thousand photos of faces of different ages. Usually, images are the input required and the correct age is the target introduced into the system. Then the algorithm attempts to optimize the intermediate steps to assess the required age.
However, RUB researchers have tried a different approach. They captured the photos of faces sorted by age, which prompted the AI to ignore the different features from one image to another and to focus only on those that changed slowly.
“Think of it as a film compiled from thousands of face photos,” said Wiskott.
The AI intelligence uses all the features that constantly change from face to face, such as eye color, mouth size, nose length. Rather, it focuses on features that change slowly on all faces.
The system also recognizes ethnic origins.
The principle of slowness also allowed the system to reliably identify the ethnic origin. The photos are inserted into the system sorted by age but also by ethnicity.
As a result, the characteristics of an ethnic group did not change rapidly from one image to another; rather, they changed slowly, but by leaps and bounds.
Indeed, scientists have also provided data on ethnicity to the system and, as with the age factor, they have done so slowly. The AI algorithm can estimate ethnicity with a probability of 99%.