BY: SAMANTHA TAPP
Imagine a criminal system that wasn’t based upon the assumption of innocent until proven guilty, but rather innocent until a computer algorithm analyzes your face and identifies you as a law-breaker. No, this isn’t the next feature being implemented into our law system, but researchers out of Shanghai Jiao Tong University have found that their program can distinguish between criminals and non-criminals just by their facial features.
In their controversial study, Xiaolin Wu and Xi Zhang, set out to discover whether a computer could detect if a human was a law-abiding citizen or not just by interpreting natural features. In their paper, ‘Automated Inference on Criminality using Face Images,’ the researchers concluded they were successful in determining who was a criminal, and even stated they found a new law governing ‘the normality for faces of non-criminals.’
For the study, they used ID photos of 1,856 Chinese men between the ages of 18 and 55, half of which are criminals. The men had no facial hair, scars, or other defining markings. None of the ID photos were mugshots and out of the 730 criminal photos, 235 had committed violent crimes including kidnap, rape, assault, murder and robbery.
The next step involved the men putting the images into a machine learning algorithm to test if the neural network could identify who was a criminal and who wasn’t. The results are alarming, with the computer having an accuracy of 89.5 per cent.
“By extensive experiments and vigorous cross validations,” the researchers wrote, “we have demonstrated that via supervised machine learning, data-driven face classifiers are able to make reliable inference on criminality.”
The men noted that there are three specific facial features that the computer uses to be precise in its findings: the curvature of the upper lip (23 per cent larger for criminals); the distance between the two inner corners of the eyes (6 per cent shorter for criminals); and the angle between two lines drawn from the tip of the nose to the corners of the mouth (20 per cent smaller for criminals).
Top row- criminal faces; bottom row- non-criminal faces.
More so, Xiaolin and Xi found that criminal faces have much more variety than those of non-criminals. To put it more simply, they said that criminal and non-criminal faces can be categorized into two different “manifolds.” Three different facial structures were found belonging to non-criminals, while four belonged to the criminal manifold.
“In other words, the faces of the general law-biding public have a greater degree of resemblance compared with the faces of criminals, or criminals have a higher degree of dissimilarity in facial appearance than normal people,” the men wrote.
According to Motherboard, the researchers said the computer did present some false positives by identifying non-criminals as criminals, and false negatives by identifying criminals as non-criminals.
Unsurprisingly, the study has stirred up controversy online; some even going as far as to call the men “irresponsible socially.”
In the paper the men admit that they are “not qualified to discuss or to debate on societal stereotypes.” But as The Intercept points out, computers and softwares are designed by people and in this case, the men created the software to depict criminality from facial features, therefore the program is not free from inherent bias.
It goes without saying that this method is not going to be implemented into criminal systems. But the amount of controversy that has been stirred up by this study does imply that more research will be conducted on this field of study, perhaps with varying sexes and ethnicities.