YouTube’s Recommendation System is Criticized as Harmful. Mozilla Wants to Research It

Posted on the 17 September 2020 by Thiruvenkatam Chinnagounder @tipsclear

YouTube's video recommendation system has been repeatedly accused by critics Sending people into rabbit holes of disinformation and extremism. Now Mozilla, the non-profit organization that creates the Firefox browser, wants YouTube users to help them investigate how the controversial algorithms work.

Mozilla announced on Thursday a project asking users to download a software tool that will give Mozilla researchers information about what video recommendations users will get on Google's own platform.

For more of that

Subscribe to the CNET Now newsletter to receive the most important stories of the day from our editors.

YouTube's algorithms recommend videos in the What's next column? On the right side of the screen, in the video player after the end of the content or on the homepage of the website. Each recommendation is tailored to the person watching and takes into account things like the history of the observation, the list of channel subscriptions or the location. The recommendations can be innocuous, like any other live performance by the band you're watching. However, critics say YouTube's recommendations can also lead viewers to marginal content such as medical misinformation or conspiracy theories.

Mozilla's project comes from the fact that YouTube, which has more than 2 billion monthly users, is already struggling with viral toxic content. Earlier this year, the company struggled to hold shares in the planemia, a video spreading false information about COVID-19. YouTube and other platforms have also taken a blow to help spread the media QAnon conspiracy theory, who baselessly claims that a group of "Deep State" actors, including cannibals and pedophiles, are trying to bring down President Donald Trump. The stakes will continue to rise in the coming weeks as Americans search for information online ahead of the US presidential election.

"Despite the dire consequences, YouTube's recommendation algorithm is completely mysterious to its users," said Ashley Boyd, vice president of advocacy and engagement at Mozilla, in a blog post. "What does YouTube recommend that users in the US watch in the last few days before the election? Or in the following days when the election results may not be clear?"

In order to participate in Mozilla's project, users have to install an "extension", a kind of software tool for Firefox or Chrome, the Google browser. Using the tool called Regret Reporter, users can flag videos they think are harmful and send the information to Mozilla's researchers.

Users can also add a written report detailing what recommendations led to the video and anything else they think is relevant. The extension also automatically collects data on how much time a person spends on YouTube. Mozilla hopes to gain insight into which behavior patterns lead to problematic recommendations.

When asked about the project, YouTube stated that it had problems with the Mozilla methodology.

"We're always keen to see the research on these systems and explore ways to work even more closely with researchers," Farshad Shadloo, a YouTube spokesperson, said in a statement. "However, it is difficult to draw full conclusions from anecdotal examples and we continuously update our recommendation systems to improve the user experience."

He said YouTube made more than 30 policy and enforcement updates to the recommendation system in the past year. The company has also cracked down on medical misinformation and conspiracy content.

Mozilla has taken a close look at YouTube's algorithms in the past. Last year, the organization awarded a scholarship to Guillaume Chaslot, a former YouTube engineer and outspoken critic of the company, to support his research on the platform's artificial intelligence systems. In July, Mozilla launched a project it sponsored, called "TheirTube," which allows users to see what YouTube's recommendations for people with different ideological views might look like.