Studies Show Racial and Gender Bias Throughout On-Demand Labor Market
Researchers are finding racial and gender disparities in just about every corner of the on-demand labor market. A new study showed black people received more negative reviews than white workers on online labor marketplaces TaskRabbit and Fiverr and that people who used the services to hire female workers were less likely to leave any feedback at all.
The findings from the academic study, expected to be published in February, follow similar research showing that drivers with Uber and Lyft and Airbnb hosts were discriminating against customers with black-sounding names. TaskRabbit was previously the subject of another study, which showed that workers were less likely to take jobs in low-income neighborhoods.
The problem isn’t that technology companies have designed systems that are explicitly discriminatory but that they enable people to act on their biases, said Christo Wilson, an assistant professor at Northeastern University and an author of the new study. “There are limits to what you can do to make your users act differently,” he said. But the structure of the marketplaces matter.
In the latest study, researchers scraped data from user profiles on TaskRabbit and Fiverr’s websites in December 2015, collecting photographs, reviews and the worker's placement in search rankings. The researchers hired people to look at the photographs and categorize each person based on gender and race. An abridged version of the study was published on Saturday. The MIT first reported on the research last week.
While researchers found evidence of bias on both sites, the study showed TaskRabbit’s platform to be particularly problematic because user feedback influenced where people showed up in the search rankings. White women and black men appeared lower in rankings, while Asian workers and black women were higher. Given that people tend to choose one of the first options presented to them in any online search, this could impact employment prospects.
While TaskRabbit acknowledged that it incorporates reviews into the algorithm it uses to decide search rankings, it said they carry relatively little weight. "We spend a lot of time thinking about this issue and working to minimize discrimination and bias on our platform,” the company said in an e-mailed statement. TaskRabbit also said many jobs are arranged through its “quick assign” feature, which matches workers to tasks automatically, instead of letting users choose who to hire. But it’s not clear whether this eliminates bias because those assignments are generated by algorithms similar to those used to rank workers in search results.
Fiverr also disputed the results of the study. “Demographic information is not required on Fiverr, and our user experience places an emphasis on the Gig, not the person completing it,” it said in a statement. The company added that many people don’t post photographs of themselves.
When presented with similar patterns of racial disparities, Airbnb responded by de-emphasizing user photographs on the site. In September, the company began urging hosts to let guests book units without being screened through a feature called Instant Book and said it would prevent people from listing units on days they had told potential guests the spaces were unavailable. But it declined to remove photographs from its platform altogether, as some critics had pressed it to do.
Wilson said companies should strive to quantify the racial disparities on their platforms and write software that explicitly offsets them. He suggested, for instance, ranking black workers higher in search by default. “The counterintuitive thing is you can’t be colorblind or neutral,” he said.
This is bound to be controversial. Few people are calling on technology platforms to collect more demographic information about their users. Decision-making by computers is being questioned in everything from the criminal justice system to news on social media, and the idea of an algorithmic affirmative action may provoke winces. Wilson acknowledges as much: “Nobody wants to get into this.”