The Guardian recently published an article describing one Uber driver’s legal test case accusing the company of indirect race discrimination their use of facial recognition software that has misrecognized and locked out him and several other Black drivers.
The driver, who asked not to be named, submitted a photo through the Uber app as part of regular protocol to verify his identity. The software failed to recognize him, locking him out of the app for 24 hours, so he submitted another photo which returned no match either. He then received a message 4 days later saying that his account had been deactivated and saying that “[t]he matter is subject to no further review,” contradicting Uber’s statement that those who had their accounts deactivated could appeal against the removal with human review — similar to the human expert review they claimed is exercised before the deactivation of any driver.
Abiodun Ogunyemi, a Nigerian driver that worked for Uber Eats, suffered a similar situation when he was locked out from the app after failing the facial recognition test several times potentially as a result of his longer hair and beard in comparison to the photo he initially uploaded to Uber. However, his other facial features were clearly visible in the failed attempts, and this inability of the facial recognition software to properly recognize faces of people of color has caused similar situations to happen to several other drivers, many of them Black.
Microsoft, who developed the software in use by Uber, has admitted in the past that it is worse at recognizing people of color, an issue prevalent across many facial recognition softwares. The sources of this issue, which often include a lack of diverse data at the training level of the algorithm, are important to identify and change going forward in the development of more advanced facial recognition softwares. The consequences of the failed facial recognitions were felt at a real level by the many Uber drivers who were removed from the platform, thus taking away an important source of income for many at the hand of faulty software.