‘Peak algorithm’: Why some people are pushing back against machine curation

This story is from The Pulse, a weekly health and science podcast. Subscribe on Apple Podcasts, Spotify, or wherever you get your podcasts.

Find our full episode about the 20th anniversary of YouTube here.

In 2006, during  the early days of YouTube, Mia Quagliarello’s job was to hire and lead a team called “coolhunters.” 

They looked for interesting videos about music, news, film, comedy, or beauty to feature on the YouTube homepage. 

Whenever they found one, it was like “a diamond in the rough,” Quagliarello said.

These videos were not immediately popular, but the coolhunters thought they deserved a larger audience.

There was a parody techno music video for a song called My Hands Are Bananas, and a 79-year-old man in England who had lived through World War II and talked about his life. 

Quagliarello said the coolhunters did not want to pick winners and losers on YouTube, so they were also part of the communities they watched. For example, the person looking out for comedy videos to feature made them himself and new other comedians.

Then in November 2006, Google bought YouTube. 

Some time after that, Quagliarello said, executives started pushing the site to feature videos that would keep people on the platform longer, rather than looking for different or creative videos to feature. 

“I always felt like a gazelle on the Serengeti being hunted by lions who were the suits, the engineers, and algorithms (who) were always salivating over what little real estate we had on the homepage and really wanted to optimize the site,” she said. “I felt like my team’s expertise and purpose was not entirely appreciated once we got acquired by Google, and it would be a matter of time before we would be deemed obsolete.” 

She said the coolhunters were proud of the relationships they built within various communities, but it was not always easy to prove to Google that they were making money for YouTube. 

The company disbanded the coolhunters in 2010, and she left the year after. 

Today, algorithms decide what people see not just on YouTube but other major platforms like Facebook, Spotify, and Netflix. But this means something gets lost, said Miguel Gomez, who opened a video store outside Philadelphia in 2012 called Viva Video.

He said while most people thought he was foolish, his team developed a loyal customer base who wanted to talk to other film lovers and watch movies they might not have found otherwise. 

For instance, he recalled one customer who usually rents art house movies, but one day he recommended she try Rubin & Ed, a 1991 comedy starring Crispin Glover as a man who goes on a road trip across a desert to bury his cat. She loved it. 

“It’s not the sort of thing that’s going to show up on your algorithm,” he said. 

Some clerks were  more eclectic with their picks. Gomez said one employee he used to work with would recommend an experimental 1980s movie with no plot, no dialogue, and no characters called Koyaanisqatsi to almost anyone who walked through the door. The movie consists of slow motion and time lapse videos like rocket launches, bats flying, clouds floating, people walking, set to music by famous American composer Phillip Glass.

“If you were coming in with your kid for a sleepover, he would tell you to take Koyaanisqatsi. To some degree … that’s a little insane, but it’s also awesome.” 

Despite those connections, the video store closed in 2021 when their landlord got a much better offer on the property. 

The heyday of video stores may be long over, but more people are pushing back against algorithm-driven culture in different ways, said Kyle Chayka, a staff writer at the New Yorker and author of the recent book Filterworld: How Algorithms Flattened Culture. 

“We’ve gone through this period in the 2010s of reaching peak algorithm. And we now kind of understand why algorithmic recommendations are sometimes bad, or we shouldn’t rely on them completely.

As examples, he pointed to the rise of fashion newsletters where people pay to support someone to give clothing recommendations, and DJs who have enough of a following to make money from subscriptions. 

He acknowledged that it can be hard to tell where taste starts and bias begins, but machine curation is also flawed in that it only works by engagement — it makes what is already popular more popular. 

“What I find really powerful about human curators is their ability to surface the weird culture, the obscure stuff, really niche finds that they like, and then pass it on to more people. And that’s kind of how culture gets developed and innovates at all.” 

Quagliarello, who used to lead the coolhunters at YouTube, is now head of creator community at Flipboard, an app and platform that curates online content like news stories, social media posts, and videos into feeds that people can follow and customize. Algorithms pick what goes in the feeds, but high-profile ones like news, politics, and tech get human attention. 

“We often talk about curation at Flipboard as a layer cake, and you don’t always know what the ingredients of the cake are,” she said. “There’s a lot of wizard-behind-the-curtain tweaking that goes on to make sure that what’s being surfaced is high quality.”

Chayka cited smaller, more specialized platforms like NTS and the Criterion Channel as platforms that have a good balance of human and algorithm input. He said that while these platforms are not as big as giants like Spotify and Netflix, he sees that as a feature — not a bug. 

“Through the social media era of the 2010s, we learned that infinite scale is bad … I don’t think there should be a platform like Facebook or TikTok that works for a billion people. It just seems to have too many negative consequences and too much of a homogenizing force for what gets broadcast across it. So I think I would prefer to return to the smaller scales of human curation and of individual voices rather than recapitulate this giant global scale social network model.”