Uber’s facial recognition software has come under fire after a driver claimed his account was terminated because it could not recognise him due to his dark skin.
The plaintiff, William Fambrough, claims that the software repeatedly failed to recognise him when trying to verify his identification. Fambrough said he even attempted to lighten his images to fix the problem, which then led the company to deactivate his account.
As a safety precaution, the ride hailing service randomly asks drivers to verify their identity to ensure that the same person that they have background checked and cleared to drive is the person using the app.
The feature, which was introduced in September 2016, does this by asking the driver to take a selfie on their phone, which is then checked using Microsoft’s Face API software.
Fambourgh, who worked late-night shifts for the company, said the software was unable to identify him in “pitch darkness”. In response, he began editing his images by lightening his skin tone.
Uber then contacted him and said his images were fraudulent, and terminated his account earlier this month. In documents filed with the court in Missouri, USA, and seen by the Daily Telegraph, Fambrough claims Uber’s actions left him with just $0.08 to his name and caused his migraines to worsen rapidly.
“Uber has not presented any evidence that anyone else other than me has used my account. My skin colour is the only answer for my deactivation and this suit is my only recourse for Uber’s adverse actions,” said Fambourgh.
The complainant did concede that he used an older phone with a low-pixel camera to drive for Uber, which may have contributed to the software’s inability to recognise him. Fambourgh is asking for $227,000 in compensation, and for his account to be reinstated.
Lecturer in computer science at Stanford University, Cynthia Lee, said: “It shouldn’t be surprising to anyone who is familiar with face recognition software that a black man would encounter an issue like this, and it’s unfortunate that Uber apparently did not design their procedures to account for this entirely foreseeable situation.”
A number of facial recognition systems, including Microsoft’s Face API, have come under criticism for having a much higher failure rate with black people and women due to a dearth of images of non-white people being used to train the software.
According to a 2018 study, Microsoft Face API has a 12.9% error rate for dark skinned faces and when predicting a person’s gender, compared to 0.7% for light skinned faces.
This is not the first time Uber’s Face ID system has proved controversial. Last year, it was under the spotlight after it suspended transgender drivers accounts when it failed to match their pictures to the images held on file.