Future Tense

The Curse of “You May Also Like”

Algorithms and “big data” are good at figuring out what we like—and that may kill creativity.

A patron works on his laptop during the Tech Crunch Disrupt conference in San Francisco, California, September 11, 2012.
Will an increasing reliance on algorithms stall creativity and innovation?

Photo by Beck Diefenbach/Reuters

Of all the startups that launched last year, Fuzz is certainly one of the most intriguing and the most overlooked. Describing itself as a “people-powered radio” that is completely “robot-free,” Fuzz bucks the trend toward ever greater reliance on algorithms in discovering new music. Fuzz celebrates the role played by human DJs—regular users who are invited to upload their own music to the site in order to create and share their own “radio stations.”

The idea—or, perhaps, hope—behind Fuzz is that human curators can still deliver something that algorithms cannot; it aspires to be the opposite of Pandora, in which the algorithms do all the heavy lifting. As its founder, Jeff Yasuda, told Bloomberg News last September, “there’s a big need for a curated type of experience and just getting back to the belief that the most compelling recommendations come from a human being.”

But while Fuzz’s launch attracted little attention, the growing role of algorithms in all stages of artistic production is becoming impossible to ignore. Most recently, this role was highlighted by Andrew Leonard, the technology critic for Salon, in an intriguing article about House of Cards, Netflix’s first foray into original programming. The series’ origin myth is by now well-known: Having studied its user logs, Netflix discovered that a remake of the British series of the same name could be a huge hit, especially if it also featured Kevin Spacey and was directed by David Fincher.

“Can the auteur survive in an age when computer algorithms are the ultimate focus group?” asked Leonard. He wondered how the massive amounts of data that Netflix has gathered while users were streaming the first season of the series—how many times did they click the pause button?—would affect future episodes.

Many other industries are facing similar questions. For example, Amazon, through its Kindle e-reader, collects vast troves of information about reading habits of its users: what books they finish and what books they don’t; what sections they tend to skip and which they read most diligently; how often they look up certain words in the dictionary and underline passages. (Amazon is hardly alone here: Other e-book players are as guilty.)

Based on all these data, Amazon can predict the ingredients that will make you keep clicking to the very end of the book. Perhaps Amazon could even give you alternate endings—just to make you happier. As a recent paper on the future of entertainment puts it, ours is a world where “stories can become adaptive algorithms, creating a more engaging and interactive future.”

Just as Netflix has figured out that, given all their data, it would be stupid not to enter the filmmaking business, so has Amazon discovered that it would be stupid not to enter the publishing business. Amazon’s knowledge, however, goes deeper than Netflix’s: Since it also runs a site where we buy books, it knows everything that there’s to know about our buying behavior and the prices that we are willing to pay. Today Amazon runs half a dozen publishing imprints and plans to add more.

The music industry embraced similar methods several years ago, drawing on vast databases of previous hits and misses to predict whether any new music sent their way is likely to hit it big. The advantage here is obvious: One doesn’t need to have connections—once a prerequisite to success—to get a contract. Instead, you just need a song that, based on past trends, is likely to be a hit.

But the disadvantage is also obvious: We might end up with very bland songs that all sound alike. As Christopher Steiner wrote in his 2012 book, Automate This, such techniques “may bring us new artists, but because they build their judgment on what was popular in the past, we will likely end up with some of the same kind of forgettable pop we already have. It’s a clear foible of the technology that all these years of so-so music are included in its analysis.”

Watson—IBM’s supercomputer—is about to start wading through thousands of legal and medical documents to make assessments that no lawyers or academics can (not with so many data to look through, anyway). If the goal is to analyze what has sold in the past and try to predict what, based on all these data points, is likely to sell in the future, Watson could easily expand into music, film, and books.

Alas, such expansion, while benefiting sales, might stall cultural innovation. Would Watson be able to predict—if it were around back then—the rise of the impressionist painting or of the futurist poetry or the new wave cinema? Would it have approved of Stravinsky? Big Data would have probably missed the Dada.

To understand the limits and opportunities of algorithms in the context of artistic creation, we need to understand that the latter usually consists of three elements: discovery, production, and recommendation. Startups like Fuzz target the last element—recommendation—hoping that some would rather be guided by humans than algorithms.

FiveBooks, a smart site for book-lovers, applies a Fuzz-like model to books; here, too, the logic is that humans can outperform the algorithms. Amazon has many great recommendations, but FiveBooks, with its picks from Paul Krugman, Harold Bloom, and Ian McEwan, is in a different league. Recommendation is where human-powered and algorithm-powered recommendations can probably coexist, at least for the foreseeable future, as users find the right balance between the two.

But when it comes to discovery of new talent and the subsequent production of their work, things look much gloomier. After all, recommendation matters only if there’s great art to recommend. If that art is selected based on how likely it is to match the success of previous selections and if it’s produced based on immediate feedback from the audience, sales might increase, but will anything truly radical emerge out of all this salesmanship?

The early signs are not very encouraging. Last December, the Global Times, China’s English-language tabloid, ran a story on the local punk band Bear Warrior, which found an ingenious way to measure the audience response to their songs. Its lead singer is a graduate student majoring in precision instruments at a university in Beijing, so he designed a device—“POGO Thermometer”—that measures the intensity of the audience’s dancing through a series of sensors embedded in the floor carpet in the music hall. The signals are then transmitted to a central computer where they are closely analyzed in order to improve future performances.

According to the Global Times, the band found that fans “started moving their bodies when the drums kicked in, and they danced the most energetically when he sang higher notes.” As its lead singer put it, “the data helps us understand how we can improve our performance to make the audience respond to our music like we intend.”

Perhaps, it would help improve their performance, but when did punk music become so nice? Making the audience happy is something for management consultants—not punk musicians!—to obsess about. The Sex Pistols would have only one use for that carpet and, rest assured, it wouldn’t involve sensors of any kind. But the Sex Pistols, oblivious to feedback, launched a revolution, while Bear Warrior, at best, would launch a career.

This article arises from Future Tense, a collaboration among Arizona State University, the New America Foundation, and Slate. Future Tense explores the ways emerging technologies affect society, policy, and culture. To read more, visit the Future Tense blog and the Future Tense home page. You can also follow us on Twitter.