Could artificial intelligence (Artificial Intelligence or AI) identify a criminal just by his facial features? With the appearance of photography, some 19th century scientists believed they could develop methods to identify criminals by their facial features. Their assumptions ultimately proved unreliable, but a new artificial intelligence (AI) technology claims they were right after all.
Xiaolin Wu and Xi Zhang from Shanghai Jiao Tong University of China have developed a neural network that is supposed to recognize criminals by their facial features.
To achieve this, researchers have used vision-engineing algorithms to look at photographs by contrasting photographs of criminals and non-criminals. The original goal was to find out if a neural network can be trusted.
In this process, the scientists fed the neural networks a total of 1856 ID photos of men (without beards) ages from 18 to 56. Half of them had a criminal record.
Researchers used only 90% of photos to train AI to recognize the differences between the two groups, and used the remaining 10 percent for their testing.
The result was impressive. The neural network could find out who had a criminal record with the astonishing accuracy of 89,5 percent.
“Αυτά τα εξαιρετικά συνεπή αποτελέσματα είναι ενδείξεις για την εγκυρότητα της αυτοματοποιημένης εξtreatmentof conclusions on the faces of criminals, despite the historical controversy surrounding the subject", report the researchers.
The MIT Technology Review, explains that there are three determinant features of the person counted from the neural network to make its classifications:
The curvature of the upper lip, which is an average of 23 percent greater for criminals.
The distance between the two inner corners of the eyes, which is 6 percent less to the criminals,
and the angle between the two lines coming from the tip of the nose to the corners of the mouth, which is 20 percent smaller.
What is extremely interesting is that compared to non-criminals, criminals tended to have much more variation in their personality characteristics.
"In other words, the faces of people who are not ebluelawmen have a greater degree of similarity than criminals' faces, or criminals have a higher degree of dissimilarity in facial appearance than normal people,” Xiaolin and Xi cite as an additional observation.
Of course, there are too many questions about the parameters of this study and how reliable the proposed method is.
The photographic sample given to the neural network was extremely limited. The fact that artificial intelligence could one day have the capacity to perform various identification work of the person can not be denied.
Now how threatening these methods can be for people who are law-abiding citizens is a matter that scientists should look at with caution.
If you are interested in the topic you can find more details in PDF with full academic work.