Is your algorithm feminist enough?
Streaming music platforms such as Spotify, Apple Music and iHeartRadio may be giving male artists more airtime, according to a new study that discovered a “widely used” algorithm is more likely to recommend male artists, as opposed to “female and mixed-gender artists.”
An analysis of 330,000 streaming music listeners over a nine-year period showed that only 25% of songs played were led by women. Their observations revealed that “on average,” the platform will start by spinning six tracks by men before choosing a female artist.
“Users had to wait until song seven or eight to hear one by a woman,” wrote Christine Bauer, assistant professor of Human-Centered Computing at Utrecht University in the Netherlands, and Andres Ferraro, Ph.D. candidate at Pompeu Fabra University in Barcelona, Spain. Their work was presented in the Proceedings of the 2021 Conference on Human Information Interaction and Retrieval last month.
In an editorial about their research in the Conversation, the study’s authors point out representation for women and queer artists is already “tremendously low.” Just 23% of artists appearing on the 2019 Billboard Top 100 were not men, according to an unrelated study. Researchers also noted in the article that less than 20% of registered songwriters and composers are women
When the industry promotes more men, listeners’ behavior will reflect that trend. And the algorithm, which aims to please the user, will follow suit.
In a previous report of listeners’ interests, researchers at Spotify and Harvard University found that the disparity between artist genders varied widely by genre: 95% of rap and hip-hop streams were male-only groups and performers. For pop, as much as 40% of spins were female artists or groups that included at least one woman. By contrast, metal fans listened to women wail a mere 0.7% of the time, and nonbinary artists at a rate of 7%.
Bauer and Ferraro suggest that the radio can be retrained to play fewer men over time by creating a positive feedback loop, starting with the streaming service. By retooling the algorithm to forcibly rank men lower in a playlist, a simulation demonstrated that, eventually, the listener would begin choosing more female and nonbinary artists of their own volition, signaling to the machine to integrate more non-male artists. And so on.
The authors hope that this sort of strategy may be considered as a means to correct certain biases in artificial intelligence, particularly in terms of race and ethnicity. Popular dating apps, for example, have been accused of stealing swipes from people of color. For minority groups, the trend can also be a detriment to their health, when clinical software favors white patients over black.
“So far, our simulation could demonstrate the benefits of a simple re-ranking approach. But responsibility is, of course, not with the platform providers alone,” Bauer and Ferraro conclude. “The rest of us need to follow.”