Australian artificial intelligence company Appen has been called for including a question about prospective workers’ complexion during recruitment.
Texas-based social media strategist Charné Graham posted on Twitter about what she called a “paper bag test” in a job application for the company – referencing an old discriminatory practice where African Americans had their skin tone compared with the colour of a paper bag.
A screencap in Graham’s tweet shows a drop down menu for ‘complexion’ under which are six options for types of skin colour ranging from “light – pale white” to “very dark brown to black”.
Speaking to Nine newspapers, Graham said the job she was applying for involved “remotely fulfilling random tasks at a very low pay rate”.
“None of these tasks/projects required my complexion,” she said.
“I’m aware that Appen is an artificial intelligence company but as a Black woman the question is very off putting and triggering with no clear explanation as to why you would need that information.”
In a blog post, Kerri Reynolds, Appen’s senior VP of HR and crowdsourcing, explained how the company requires data “from diverse populations of people” for its core business of selling the data sets used to train AI.
“We ask for demographic data from our contributors to understand the diversity of our crowd and provide diverse datasets to our customers,” she said.
Has anyone ever seen this on a job application? pic.twitter.com/OVqwNbTioR— Charné Graham (@CharneGraham) May 10, 2021
Appen boasts having a remote workforce of over one million people – its ‘crowd’ – who annotate specialised data for use in machine learning and artificial intelligence.
Algorithmic bias is an important concern among ethicists and human rights activists who have called for greater scrutiny over the design of AI systems.
Research has shown, for example, that facial recognition systems carry higher error rates when matching faces of darker skinned people.
This has real effects – like when US man Nijeer Parks was arrested because a facial recognition scan returned a false positive result.
To its credit, Appen appears aware of the potential bias in training data and deliberately seeks to account for it in its annotation workforce.
Graham described the company’s decision to ask for her skin tone as “a failed attempt” to create diversity.
“Diversity, equity and inclusion are vital in every company,” she told Nine.
“But since the tech industry has been historically known to exclude women and Black people, this question was not the best way to test out ‘machine learning’.”
Reynolds said Appen has since removed the complexion question.
“We acknowledge that it required an explanation,” Reynolds said.
“We are working to improve the ways we communicate with our contributors and provide our customers with diverse data to mitigate bias.”
Appen is listed on the Australian Stock Exchange (ASX) and currently has a market capitalisation of $1.39 billion.
It made just shy of $600 million in revenue last year but had its stock trading halted last week due to a sudden burst of volume and rapid price drop.