Computers are worse at recognising women and people of colour than white men. Documentary Coded Bias shows that the problems don’t stop there
12 August 2020
Face-recognition AI could only “see” Joy Buolamwini when she wore a white mask
7th Empire Media
Ongoing film festival screenings
IN HER first semester as a graduate student at the MIT Media Lab, Joy Buolamwini encountered a peculiar problem. Commercial face-recognition software, which detected her light-skinned classmates just fine, couldn’t “see” her face. Until, that is, she donned a white plastic mask in frustration.
Coded Bias is a timely, thought-provoking documentary from director Shalini Kantayya. It follows Buolamwini’s journey to uncover racial and sexist bias in face-recognition software and other artificial intelligence systems. Such technology is increasingly used to make important decisions, but many of the algorithms are a black box.
“I hope this will be a kind of Inconvenient Truth of algorithmic justice, a film that explains the science and ethics around an issue of critical importance to the future of humanity,” Kantayya told New Scientist.
The documentary, which premiered at the Sundance Film Festival earlier this year, sees a band of articulate scientists, scholars and authors do most of the talking. This cast primarily consists of women of colour, which is fitting because studies, including those by Buolamwini, reveal that face-recognition systems have much lower accuracy rates when identifying female and darker-skinned faces compared with white, male faces.
Recently, there has been a backlash against face recognition. IBM, Amazon and Microsoft have all halted or restricted sales of their technology. US cities, notably Boston and San Francisco, have banned government use of face recognition, recognising problems of racial bias.
People seem to have different experiences with the technology. The documentary shows a bemused pedestrian in London being fined for partially covering his face while passing a police surveillance van. On the streets of Hangzhou, China, we meet a skateboarder who says she appreciates face recognition’s convenience as it is used to grant her entry to train stations and her residential complex.
“If an AI suspects you are a gambler, you could be presented with ads for discount fares to Las Vegas”
The film also explores how decision-making algorithms can be susceptible to bias. In 2014, for example, Amazon developed an experimental tool for screening job applications for technology roles. The tool, which wasn’t designed to be sexist, discounted résumés that mentioned women’s colleges or groups, picking up on the gender imbalance in résumés submitted to the company. The tool was never used to evaluate actual job candidates.
AI systems can also build up a picture of people as they browse the internet, as the documentary investigates. They can suss out things we don’t disclose, says Zeynep Tufekci at the University of North Carolina at Chapel Hill in the film. Individuals can then be targeted by online advertisers. For instance, if an AI system suspects you are a compulsive gambler, you could be presented with discount fares to Las Vegas, she says.
In the European Union, the General Data Protection Regulation goes some way to giving people better control over their personal data, but there is no equivalent regulation in the US.
“Data protection is the unfinished work of the civil rights movement,” said Kantayya. The film argues that society should hold the makers of AI software accountable. It advocates a regulatory body to protect the public from its harms and biases.
At the end of the film, Buolamwini testifies in front of the US Congress to press the case for regulation. She wants people to support equity, transparency and accountability in the use of AI that governs our lives. She has now founded a group called the Algorithmic Justice League, which tries to highlight these issues.
Kantayya said she was inspired to make Coded Bias by Buolamwini and other brilliant and badass mathematicians and scientists. It is an eye-opening account of the dangers of invasive surveillance and bias in AI.
More on these topics: