MERL Tech DC 2019

In this session, we will be thinking about the following questions:
What are the benefits and risks of using facial biometrics for identification and what are their implications? How do we address gender, racial and other biases when it comes to facial recognition systems and the datasets used for their training? How can we guarantee fairness in a transparent fashion? What are the ways in which we can hold ourselves to the highest possible standards when it comes to privacy and consent? And how do these governing philosophies shape into business policies?

Context:
Digital identification systems that use biometric characteristics are transforming the efficiency and scalability of programs, by being used to verify last-mile delivery, tackle fraud, and vastly increase the power of monitoring and evaluation in development. With the boom of AI, facial recognition systems offer a non-invasive, contactless method of identification, but great efforts need to be taken to ensure that this is done in a fair and ethical manner. Due to the homogeneous, predominantly white-male nature of facial datasets, systems have been shown to struggle with successfully enrolling dark-skinned faces, and have a much higher accuracy when it comes to light-skinned faces. Models that have correctly identified the gender of men have had lower accuracies on women. Such striking racial and gender biases mean that marginalised groups will suffer harmful consequences and inequalities shall amplify unless we take measures now. Identity and human dignity ought to be respected and valued fairly.