The app is capable of analyzing faces from video streams, cameras or existing databases and match encode face descriptions with a set of standard personality traits they are likely to possess. The company has created 15 different classifiers, which are capable of evaluating certain traits with 80 percent accuracy and now they are pushing the technology forward with machine learning.
The Art and Science of Facial Recognition
Faception took the scientific claim that our genes affect our personality. A study conducted in Edinburg University discovered that identical twins were twice as likely to have the same personality traits as non-identical twins, meaning their DNA had a higher impact than upbringing. Scientists have already discovered that there are 5 main genes that shape the person’s face. Further research proved that internal facial features are valid signs of personality traits and health. Faception does not take race, age or gender into account as those are not the defining factors of committing a crime – a positive shift from stigmatized racial assumptions. Based on the scientific data available, the company developed the specific classifiers for the scanned personas. For instance, a professional poker player is described as “Endowed with a high concentration ability, perseverance, and patience. Goal-oriented, analytical, with a dry sense of humor. Silent, devoid of emotion and emotional expression, strict and sharp minded, with high critical perception”. Among the early success was the app’s ability to classify 9 out of 11 Paris terrorists with no prior knowledge on the profiled suspects and only three of those proved to have some previous records.
The Challenges for Facial Recognition
The idea of judging the book by the cover isn’t as new as it seems. Physiognomy – “the science” of judging people’s personality based on their facial features was largely rejected and discredited back in the 19th century. Apart from that machine learning and image recognition technologies are still not at their best state. Even the best AI systems today can be as good as the examples they are trained on. Once the computer is exposed to a narrow or outdated data sample, the conclusions will be false. As an example, Domingos mentions a case one of his colleagues conducted. He trained a system to differentiate dogs from wolves and the tests proven to be almost 100% accurate. Later, it turned out that the machine was so successful because it learned to look for snow on the background pictures with wolves. All the wolves pictures had it in the background, unlike the dog pictures. Faception may need to work further into their classificatory and focus on expanding their database to reach the absolute level of accuracy.
