Respondus, a vendor of online proctoring software, has been granted a patent for their “systems and methods for assessing data collected by automated proctoring.” The patent shows that their example method for calculating a risk score is adjusted on the basis of people’s skin colour.
The “Adjustment to Final Tally” for dark skin is described as follows in the granted patent:
For dark complexion persons, a racial detection feature may be provided so that a downward adjustment can be made to the final risk tally. This feature may be implemented when high contrast (e.g., white balance) issues are detected. Although significant improvements with respect to facial “recognition” have recently occurred, facial recognition improvements don’t necessarily equate to facial “detection” improvements. For example, if there are significant white balance issues in the video, current systems may not be able to achieve the basic facial “detection” that is requisite for achieving facial recognition.
This is one more example – in a long list of examples – showing that software using an algorithm to detect faces can’t treat people with different skin colours equally. It is quite shocking to learn that vendors of this type of software have actually known this for quite a while now (the application was filed in mid 2019). It further strengthens Robin Pocornie’s case against her university: her experience of Proctorio (another vendor of online proctoring) was different from her fellow students with a lighter skin colour.
Hat tip to Ian Linkletter for finding this patent – and the relevant passage – and for sharing it with us. Linkletter was sued by Proctorio for tweeting links to their YouTube videos (similar to videos that this patent filing links to, and which have been taken offline by Proctorio). Do read more about his case.
See Systems and methods for assessing data collected by automated proctoring on Google Patents, or download the PDF.
Image from the patent application.