Typical facial recognition algorithms have an embarrassingly hard time identifying people of color. Obviously, that is a problem for a technology that’s meant to serve everyone. It is even more of an issue considering the various uses of facial recognition—such as by law enforcement.
An innovative software engineer from Ghana knows that it doesn’t have to be this way. Charlette N’Guessan has developed new facial recognition software that’s specifically designed to identify African faces. Her vision of what facial recognition technology should look like is an example for companies of all shapes and sizes to follow.
She also won the Africa Prize for Engineering Innovation, making N’Guessan the first-ever female recipient.
Pushing the Envelope
N’Guessan is the daughter of a high school mathematics teacher. From a young age, she knew that her strengths were in the STEM field. As such, she moved to Ghana to start her own business in 2018.
Rebecca Enonchong, a judge from the Africa Prize committee, said of N’Guessan’s work, “It is essential to have technologies like facial recognition based on African communities, and we are confident their innovative technology will have far-reaching benefits for the continent.”
The tool that N’Guessan and her team developed is called BACE API. It uses a combination of facial recognition and artificial intelligence (AI) to verify people’s identities using smartphone cameras. The tool is designed to be built into existing products, such as a banking app.
In the age of COVID-19, facial recognition has become even more popular as people are warier of things like fingerprint sensors.
“In Africa we are more familiar with fingerprints, but we believe that facial recognition is more like a natural biometric technology,” N’Guessan says.
She and her team will receive $32,000 along with the Africa Prize. N’Guessan plans to reinvest it to help the technology expand out of West Africa.
There have been numerous instances where facial recognition tools were proven to be racially biased against people of color. The issue can be attributed to the fact that the majority of facial recognition software is created by white programmers and is trained on datasets mainly comprised of white faces. When those algorithms need to identify a person of color they often make mistakes and are significantly less accurate.
Over the summer, Big Tech companies like IBM, Microsoft, and Amazon pulled away from facial recognition in response to the killing of George Floyd. Even so, the industry remains largely unchanged.
N’Guessan’s tool shows that there is a better way. The solution is incredibly simple—use more diverse datasets. She says, “The more you train your model, the more your model is able to identify more faces. We’re happy that our system is able to give a high level of accuracy.”
N’Guessan, like many others, believes there is no reason that a facial recognition tool can’t identify all different faces equally. This is something that the industry as a whole needs to address as soon as possible.