Spotify is acquiring Kinzen, a startup that specializes in using machine learning to analyze content and report potentially harmful statements to human moderators. In a press release, Spotify says the acquisition is meant to help it “deliver a safe, enjoyable experience on our platform around the world,” even as it ramps up its focus on user-generated podcasts and audiobooks, which could potentially contain misinformation or other violations of Spotify’s policies.
Spotify has already been working with Kinzen, claiming that it’s been partnered with the company since 2020 and that the startup’s tech has been “critical to enhancing our approach to platform safety.” According to Kinzen’s site, its tech is capable of analyzing audio content in several languages, and it uses data from the internet and human experts to figure out if certain claims are harmful. (It even claims to be able to spot dog whistles, seemingly innocuous phrases that actually refer to something with a darker meaning.)
Spotify has mountains of user-generated audio, and it’s tough to catch all the rule-breakers
Dustee Jenkins, Spotify’s global head of public affairs who’s quoted in its press release, says that bringing Kinzen in-house will improve the company’s ability to “detect and address harmful content, and importantly, in a way that better considers local context.”
As my colleague Ariel Shapiro has noted, the company has spent the past few years consulting with experts and launching advisory councils in an attempt to figure out content moderation, especially after it purchased do-it-yourself podcasting platform Anchor in 2019. Spotify says that there were over a million podcasts added to its service by “independent creators” last year — a monstrous amount of audio content that it really doesn’t have a lot of oversight on.
Spotify hasn’t been able to avoid controversy even with content that it has a direct connection to — the public debate over what people like Joe Rogan say on Spotify-exclusive podcasts is a perfect example of that. And if there are issues that crop up in that proportionally tiny amount of content, imagine what could be lurking unnoticed in some of those million-plus podcasts.
As a recent report from the Anti-Defamation League points out, content that’s clearly in violation of Spotify’s policies can still escape enforcement action. The report looked at explicitly white supremacist music, but Kinzen seems to be more focused on finding problematic spoken-word content and bringing it to the attention of moderators; Spotify’s press release says it’s “particularly suited for podcasting and audio formats.” The algorithm’s going to have to work overtime if it wants to listen to the thousands of podcasts added to the platform each day.