Comedian Marc Maron has a theory: All of us, when growing up, had That One Guy who introduced us to the really cool stuff in popular culture — the best music, the coolest films, the good books, and comedy records. It could have been an older brother or sister — That One Guy is a gender-neutral term. We usually encountered him or her in our teenage or college years. It might have been the punk rock girlfriend who knew about The Clash, the older kid down the street with Monty Python on VHS, or the friend who turned you on to Lizzo before she got big. Without That One Guy, our lives might have gone in another direction entirely.
But today, That One Guy is at risk of being supplanted by That One Machine: the social media algorithm, the personalized playlist, the recommended movie queue. Netflix, Spotify, TikTok and YouTube, even Apple News and Google News: their algorithms all track what we like, then give us what they think we want.
In news, that leads us to the filter-bubble problem: Once people only get information that aligns with their political perspective, they can cultivate their own reality. But even in the lower-stakes realm of entertainment and popular culture, the filter bubble nudges us away from wandering into fresh territory, subtly and relentlessly whittling down our options, offering a path of least resistance.
Often, no human beings are directly involved in these recommendations. Humans might curate the categorical bins the system draws from — music streaming services are famous for hiring music nerds — but the actual tour guide is a data-driven algorithm powered by a blunt form of artificial intelligence. It’s a bunch of lines of code, trained to identify your online behavior patterns and generate a corresponding action. You like true crime movies? Here are five more true crime movies! You like old-school rap? Here are five more old-school rap playlists!
As our machines try endlessly to please us, we risk a dull, even slightly scary future. By skipping over vast swaths of things we don’t like, we might miss out on something we’d love.
But if AI has created the problem, AI might have a fix. We just need smarter AI.
“To All the Boys I’ve Loved Before.” “Irreplaceable You.” “Always Be My Maybe.” “Falling Inn Love.” If you think you’ve seen a lot of original romantic comedies scrolling across your Netflix homepage, you’re not wrong. Low-budget rom-coms are booming on the streaming service, which discovered that fans often click on recommended titles after the credits roll, happy to stay put and splash around in the genre.
It’s one way that personalized feeds and recommendation engines have affected the movie business. David Singh, a veteran studio executive who’s worked for Disney, Sony Pictures, and Twentieth Century Fox, says the cyberbalkanization of audiences has made it harder to reach a mass number of people all at once. But it has empowered studios to greenlight films that seem to fit a rabid but narrower audience.
Pinned down by our tenacious data profiles, we’ll be forever looking into our own reflections, prisoners of our own tastes.
“The idea of ‘broadcasting’ is gone, outside of giant events like the Super Bowl or the Academy Awards,” Singh says. “We used to make a television spot thinking it would have to appeal to a lot of people. Now we know that it might be seen by a more defined group. We have to work a little harder to reach a lot of people, but you can target specific people a little easier.”
That’s potentially good news for studios and filmmakers trying to get their movies to a particular audience. But it’s much harder to get people to branch out, hop genres, and find stuff they might not expect to like. Pinned down by our tenacious data profiles, we’ll be forever looking into our own reflections, prisoners of our own tastes.
The same goes for the music industry, which has been transformed by the rise of streaming services such as Spotify, Pandora, and Apple Music. Mark Richardson, music writer for the Wall Street Journal and former editor-in-chief of the online magazine Pitchfork, says that personalized feeds and recommendation engines are changing listening habits.
“Personalized playlists have replaced radio listening for most listeners,” Richardson says. “I think it’s made the biggest impact for passive consumers, which is the majority of listeners,” Richardson says.
For several decades now, radio has been mining data to inform its playlists within relatively broad genre designations — Top 40, New Country, Classic Rock. Streaming services have accelerated the trend, mixing in hundreds of hyper-specific playlist categories sorted by era (1990s R&B) or subgenre (Outlaw Country) or mood (Totally Stress Free).
Music services differentiate their subscribers into two distinctive groups, Richardson says: “lean in” and “lean back” listeners. “Those who are more inclined to find new artists and dig deep into specific sounds and scenes are still doing that,” he says. Those listeners tend to actively mine the streaming service and go after what they want, just as they might’ve raided a specific record bin in a music store 30 years ago.
But those who just want music to play in the background — the majority of listeners — prefer streaming playlists. And that reality is influencing certain artists: those who prize streaming success above all.
“There are consulting firms, however shady, that try to coach artists to sound a certain way to increase their chance of being on a certain playlist,” says Richardson. This has generated a kind of machine within the machine, a data-driven approach to hitmaking that has been called the moneyball of music.
Richardson expects these trends to continue, thanks in part to the rise of voice assistants and smart speakers. “The appeal of ‘just play me music I like’ is quite strong,” he says. “I’d probably be most worried if I was in radio.”
But the changes can hurt musicians whose work defies categories. “There’s a lot of great and interesting music that doesn’t lend itself to streaming algorithms,” Richardson says, “and as more revenue goes to these other places, these artists have fewer resources.”
Streaming services try to recreate some of the That One Guy effect by hiring music professionals to curate specific playlists. And often, people find those recommendations on their own — particularly when the glut of choices starts to feel overwhelming.
“Netflix has recommended tons of shows to me — like, say, that new Toni Collette show, ‘Unbelievable.’ But I didn’t care at all until a mom on my son’s lacrosse team told me that she totally loves it,” Singh says. “The algorithmic recommendations may be getting smarter, but it’s going to take more than that for me to give up hours of my time. I still need the human involved.”
That’s because recommendation robots aren’t as smart as humans. Usually, their decisions are pretty simple. If the system sees you’re watching a lot of comedies, or listening to a lot of hard rock, you’re going to get more comedies and more hard rock. But when they sift the massive tangle that is popular culture, machines still miss out on the nuances.
“I’m a Grateful Dead man,” says Mike McGuire, an online media and digital distribution analyst with the research company Gartner. “Amazon knows this and it’s telling me people who like the Grateful Dead really like Jefferson Airplane, Jefferson Starship. Uh, no. I very much don’t like Jefferson Starship. That’s terrible. How could you make that link?”
McGuire’s conjecture: The Amazon algorithms are likely looking at the first-thought, surface-level associations that your square uncle might make at Thanksgiving: “Summer of Love! San Francisco!” he says. “That’s as far as it’s thinking it through.”
A sophisticated recommendation engine — a digital version of That One Guy — would know that almost all Grateful Dead fans have eclectic taste in music. The Dead folded all kinds of weird and wonderful components into their sound, McGuire says — folk, jazz, bluegrass, country, psychedelic rock.
“If you’re building a recommendation engine for Dead fans, we’re not into any one genre or one band,” McGuire says. “What you really need is a ‘Surprise Me’ button.”
As McGuire describes it, a Surprise Me button could fit any part of the culture. In news or politics, imagine telling the system to assemble a package of articles on a given topic that includes multiple viewpoints. Or you could imagine different versions of “Surprise Me”: On your music playlist page, for instance, a button could provide a sampler of different genre approaches on a theme.
McGuire says technology could help us by doing the grunt work of pulling out representative movies or songs from the online realm’s churning ocean of noise. In the more consequential realm of breaking news, analysis, and politics, such AI assistants could help us by finding viewpoints that are very much outside of our usual information diet.
“It ultimately comes back to us as consumers of information,” McGuire says. “Are we willing to say to the system, ‘Expose me to things that don’t fit my profile’?”