Pop Goes the Algorithm
By Christopher Steiner
When musicians like Norah Jones and Maroon 5 are “discovered” by a machine, it may be time to listen to the algorithms. But will engineers’ formulas make all music sound formulaic? A tech journalist describes how bots are not just picking the next great musical hits—they’re reaching for the musical stars.
As a musician and writer, Ben Novak drove the car he could afford in 2004: a 1993 Nissan Bluebird. The vehicle propelled him around his hometown of Auckland, New Zealand, just fine. Novak’s main complaint about the car concerned its radio, which could capture only two FM stations out of the dozens broadcasting in the city. As somebody who spent every spare minute imbibing or playing music, Novak found this no minor aggravation. But he didn’t have money for a new car, so he left the dial fixed on the BBC, his only acceptable option.
Being stuck on the BBC had its benefits. Novak was well armed for cocktail conversation on current events, and he could always crack off a new piece of intellectual fodder when chitchat grew stale. More important, he didn’t miss a short BBC report on technology developed in Spain that, the person being profiled claimed, could predict which songs would turn into pop hits.
“I’m driving down the road listening to this and I think, ‘That’s interesting,’” Novak remembers. “I could have just kept driving and forgot about it, frankly, but I was getting off at the next exit.”
Novak hit his exit, drove to his house, and sat down in front of his computer. He brought up the Web site belonging to what was then called Polyphonic HMI.
For $50, its algorithm would analyze any music file Novak uploaded. Potential hits earned high scores, duds got low ones.
“I mean, for $50—it was such a small amount of money when you think about what it could mean in the long run,” Novak says. “So I did it.”
Novak had written a song a few years before called “Turn Your Car Around” that he believed held significant potential. He uploaded the song and sat at his screen, waiting for a result.
Finally, the Web site whirred to life with an answer for Novak. The algorithm behind the site used a number scale to rate songs. Anything more than 6.5 had decent hit potential. Anything past 7 had a hook made for the pop charts. Novak’s song scored a 7.57—as high as the algorithm had scored many of the biggest rock hits of all time, such as “Born to be Wild” by Steppenwolf and “Peaceful Easy Feeling” by the Eagles. “I was really happy, obviously,” says Novak. “But it wasn’t really clear what came next.”
In Spain, where Polyphonic HMI was based, the computer engineers who maintained the algorithm took note of the song for its high score. They pulled it down off the server and played it in the office.
“There was clearly something there,” says Mike McCready, who ran the company. “Our guys played it over and over again.” A musician himself, McCready called some recording label contacts in Europe and got the song circulating.
About two weeks after he submitted his song on McCready’s Web site, Novak’s phone rang. It was a representative for Ash Howes, a music producer in the United Kingdom with a few dozen hits in his pocket. He had a young British pop star, Lee Ryan, who needed more tracks to fill out his album. Howe thought Novak’s song would fit in well. In fact, he thought it could be a single.
Novak quickly agreed to a favorable deal: He would get 50% of all royalties when the song was played on the radio, on TV, or in an advertisement. Novak’s song not only went on Ryan’s album but was also designated the CD’s first single. The song debuted at number 12 on the English pop charts, and for two months in a row it was the most played song in the UK.
At the Crossroads of Music and Technology
Novak isn’t shy about crediting an algorithm. The music world is one in which a hair’s width of luck can make an artist or keep him from being discovered. Algorithms that sniff out talent can change that.
“This whole music thing is just a huge gamble for anybody who goes into it,” he says. “This program, this Web site—it aligned the planets for me.”
The algorithm that changed Novak’s life was devised by a group of engineers in Spain headed by McCready, an American who took an odd path to becoming an authority on the technology that’s changing the future of music.
“I had all these friends who were getting rich with Internet companies—or thought they were getting rich,” McCready says. “It seemed like the thing to do.”
It indeed was the thing to do in 2000, and McCready went to work as the head of marketing for an online music start-up called Deo. The Swedish company fashioned itself as the first open marketplace for music. Musicians and bands could upload their music to Deo, where they could sell it directly to consumers.
Like many companies born in that era, Deo had been infused with boundless optimism and a large pile of money. And just like hundreds of other start-ups, Deo had misjudged its appeal and market. Few people knew what a digital file was, and those who did were likely getting them illegally through sites like Napster. A year later, Deo ran out of cash and folded. The experience proved valuable for McCready as he got to spend a year at the crossroads of music and technology.
During that twelve months, he met a small tech firm in Barcelona that had developed an algorithm for analyzing the underlying structure, patterns, and architecture of popular music. McCready spent time with the company’s engineers and concluded that the technology actually worked. He proposed forming a new company built around the technology that would pitch to musicians and record companies. They called it Polyphonic.
The algorithms behind Polyphonic work a wondrous dissection on the music they’re fed. The particular science behind the company’s algorithms is called advanced spectral deconvolution. The process breaks the songs up mathematically, isolating tunes’ patterns of melody, beat, tempo, rhythm, pitch, chord progression, fullness of sound, sonic brilliance, and cadence.
Polyphonic’s software takes this data and builds three-dimensional models with it. By looking at, instead of listening to, the song’s 3-D structure, the algorithm compares the song to hits of the past in as objective a way as is possible. Putting a just-analyzed song on the screen with number-one tracks of the past shows a kind of cloud structure filled in with dots representing songs. The hits tend to be grouped in clusters that reveal their similar underlying structures. Get close to the middle of one of those hit clusters and, while you’re not guaranteed success, you’re in very good shape.
The Bot that Came Away with Norah
When he was in Barcelona perfecting the algorithm, McCready ran as many to-be-released albums through his bot as possible. It was these test cases that would reveal if the algorithm had any real power. The algorithm rated most of the unreleased CDs as ho-hum. But one, the algorithm said, contained nine likely hits out of 14 total songs. Those are Beatles numbers. McCready could hardly believe it.
Nobody had heard of this artist, which made McCready worry that the bot was wildly wrong. But then the album, Come Away with Me, was released, selling more than 20 million copies and netting its artist, Norah Jones, eight Grammy Awards. Jones had found the clusters.
“Some people describe a hit song as a brain itch,” McCready says. “And you scratch that itch by listening to the song over and over again.”
The clusters were the itchiest spots, and McCready thought his company had struck upon the formula for identifying musical gold. The music industry already tried to pick its own hits, but it was only right 20% of the time. McCready’s tool, if it worked, would be the industry’s holy grail.
The A&R Bot Said, “I Don’t Hear a Single”
Tom Petty’s 1991 song “Into the Great Wide Open” features a well-known verse that refers to the record industry’s equivalent of a baseball talent scout: the A&R man. A&R (artists and repertoire) staff at record labels are the gatekeepers to the recording industry. They can make careers and launch a musician from obscurity to stardom. In the pop music world, A&R people’s jobs depend on finding singles. An artist without a single, as Tom Petty’s sardonic invoking of that cliché suggests, isn’t worth much.
One problem is that A&R is an inherently subjective trade. It’s not baseball; a 100 mph fastball to one scout is a 100 mph fastball to another. There’s no denying that kind of raw talent that can be quantitatively measured so easily. With music, however, talent exists everywhere.
But not all musical talent has the potential to appeal to mass audiences. Some of the most brilliant artists in the world may never become known outside a small circle of fans. Other artists whose musical talent may not be deeper than that of an average 8-year-old piano player can take over the world with one catchy pop lick.
For that reason, music remains a business much like book publishing. The record labels depend on one album out of fifty to keep them profitable and justify the signing bonuses they dole out to new acquisitions. It’s only by casting a wide net that labels assure themselves of scoring the hits they need. A&R people who unearth more diamonds than average are well compensated. Those who consistently pluck stars from obscurity become legends; these rainmakers often ascend to become leading executives at the label.
Even though A&R people are always hunting for promising artists, most of their signings come through personal relationships or direct referrals. The music business remains far from a meritocracy, even if merit is measured by the ability to hook teenage girls’ ears. But what if A&R could be made into a science? Being right just 30% of the time would be a giant improvement on the industry’s historical rate. And artists with a knack for pleasing listeners wouldn’t have to wait for that random connection or recommendation that may never come. Polyphonic’s algorithm, McCready thought, could prove the answer.
Readying the A&R Bot for Prime Time
Ben Novak’s success primed Polyphonic for what seemed like a high ceiling. Mike McCready’s tool also identified Maroon 5 as an act carrying a high probability of success before the public had any idea who the band was.
The software certainly wasn’t right about everything. It’s given high marks to heaps of songs that never gained traction with wide audiences. But there was no denying that McCready had created something that worked, something that could shape the future of the music industry.
Despite the promise of the technology, A&R personnel weren’t too keen on giving credence to a tool that, if it lived up to its claims, would threaten their jobs. Many A&R people and recording executives laughed at the notion that a machine had any place in their world. When told about the work and ideas behind Polyphonic, Lorraine Barry, the global marketing manager at Virgin Records, scoffed.
“The modern-day A&R man—a machine, a computer program? A bit of a frightening thought,” Barry said. “I think it’s a marketing ploy. It’s pretending that it can be a science.”
Whether the software has made the A&R game into a science will remain debatable. But there’s no denying that McCready’s crew wasn’t welcome within the industry. The music business isn’t renowned for being open to change. “I think it finishes just ahead of the Amish in that respect,” McCready says.
The business model behind Polyphonic depended on the music industry utilizing it as a new A&R instrument. That bet proved cheeky. Polyphonic, for all its wizardry, wasn’t able to make any money.
A&R people were loath to use a method that could hasten their own demise. Without their cooperation, Polyphonic floundered. McCready laid off staff and thought about what he’d do next after going from Nebraska farm boy to watch mogul to pop star and now tech founder.
Combing the Musical Data for Future Stars
With little to lose, McCready changed his model. In 2008, he moved to New York and became friendlier with the music industry. He recapitalized with new investors and dubbed his company Music Xray. Just as before, he invited artists to upload their work to his site and databases, but now he also allowed A&R men and producers to post veritable help-wanted signs when they might be looking for a new tune or artist.
Music labels, advertising firms, marketers, and music producers are often looking for a certain kind of sound. For instance, a music label may be on a mission to find the next Radiohead, or a marketing firm may think the Rolling Stones’ “Brown Sugar” is the perfect song for their television spot, but they can’t afford to pay the kind of money the Stones command.
Unsurprisingly, when a legitimate talent scout at a major label issues a query looking for new artists, there’s usually an avalanche of responses. The same goes for when a movie producer posts a request for an original score or a particular kind of song for a soundtrack. As people in the music business will tell you, they’re very busy. Wading through thousands of submissions from random musicians—many of them mediocre or worse—isn’t something that will often crack their daily agenda.
This is why established artists tend to get the lion’s share of new work. Finding new musicians takes too much time. This is where McCready’s algorithm comes in. It can quickly sort the right sounds from the wrong ones, allowing a music industry insider to find the closest match to their original query. In the instance of “Brown Sugar,” the algorithm would comb its databases of submitted music for the tracks that best imitate the riffs, beat, rhythm, style, and overall sound the Rolling Stones struck in that song.
Or A&R people can simply look for the highest-scoring songs within different genres from the artists who have uploaded to Music Xray. McCready’s warehouse of data grows larger each week. It’s quickly becoming an encyclopedia of world musical talent, a heartening development for musicians out there, such as Ben Novak, who, as hard as they try and as talented as they may be, fear their work and sound may never make it out of their garage.
It’s more than possible that many of our future music stars will be produced by Music Xray’s algorithm. It’s already happening, in fact. Since 2010, McCready has landed more than 5,000 artists opportunities with music labels and other commercial outlets.
“I’m finally getting love letters from record labels,” McCready says. There will always be human decision makers at some level, he thinks, but his bot and its feel for the clusters of popularity will eventually change who the public ultimately hears.
The efficiencies and the new breadth of artists that McCready’s model opens up to the music industry are such that it’s only a matter of time until the major labels—all labels, really—come to rely on an algorithm to pick the musicians they sign and the songs they market. It’s akin to when word processors first hit the market. At first, most people kept banging on their typewriters, as only the early adopters could see past the processors’ small screens, funky printouts, and the scary idea of keeping all of one’s work on a five-inch floppy disk rather than on paper that could be seen and held. But eventually, screens got bigger, the software got better, and the idea of using anything else became nonsensical. That day is coming to the music world.
In the House of the Rising Bot
It’s likely that one day we’ll see garage bands jamming out a track and then scrambling over to a laptop screen to see how that version fared in the 3-D world of hit clusters. Such quick affirmation in a creative field is rare. But it also begs the question: Rather than an explosion of variety, will algorithms lead to a music world of forced homogenization?
It’s already true that a large chunk of the hits that populate the Top 40 were written by the same group of people. Martin Sandberg, for one, a Swedish songwriter who goes professionally by the name of Max Martin, got his start in the 1990s when he wrote a series of number-one hits for Bon Jovi, the Backstreet Boys, and Britney Spears. Since 2008, he’s written more than 10 number-one hits and more than 20 Top 10 singles, including “DJ Got Us Fallin’ in Love” by Usher and “I Kissed a Girl” by Katy Perry.
Knowing that particular humans are gifted at writing hooks for the masses—and knowing what Mike McCready’s algorithm already knows about the general characteristics of hit songs—it’s easy to speculate that popular music could soon be ruled by bots. It’s a certainty that record labels will serve up whatever the tastes of the day happen to be—and little could be better suited for such a task than an algorithm tuned to spin out saccharine hits.
But as bots move into the business of music creation, the door will be left open for disruptors. And whereas disruption usually comes from technology, it’s likely that pop charts, with what will certainly be a backbone of algorithmically conceived songs, will be left vulnerable to indie artists who create something truly different.
About the Author
Christopher Steiner is the co-founder and co-CEO of Aisle50, a start-up offering online grocery deals. He was previously a technology journalist at Forbes.
Reprinted from AUTOMATE THIS: How Algorithms Came to Rule Our World by Christopher Steiner with permission of Portfolio Books, a member of The Penguin Group, (USA) Inc. Copyright © 2012, Christopher Steiner.
- About WFS
- Contact Us
- Frequently Asked Questions
- History of WFS
- Board and Council
- Press Room
- Futurist Gear
- Are You the Next CEO of the World Future Society?
- Book a WFS / Futurist Magazine Speaker
Free Email Newsletter
To sign up for Futurist Update, our free monthly email newsletter, enter your email in the box below and click Save.
It’s September, 2099. Time Magazine is bringing out a 21st Century Retrospective issue which will hit the newsagents and book stores worldwide in December. The senior news editor has commissioned a leading historian, Dr Joe White, head of the Stanford department of history, to write a political overview for the news magazine’s special edition.
'Look for weak signals of the future, especially when they impose a fear on us' - Interview with Jacintha Sheerder
How do you see the future in fifty years?
Some very interesting and promising technologies are being developed right now, which can influence life in the future drastically. Technologies are often seen as a means instead of a goal in providing a solution for some of the great global challenges, like climate change, healthcare for a growing population, and scarcity of resources. When talking about energy scarcity and for instance the demand for energy, currently it is equally likely that we end up in a world of scarcity or in abundance. The question with many technologies is whether we can implement them in the same pace as these challenges are increasing exponentially. There is still a lack of confidence in how technologies would affect us as human beings. Especially technologies that are imposed on the human body give rise to resistance. How far do we want to go? There could become a tipping point in society. People could say, this is the limit as they no longer have grip of what technology does to their bodies and brains.
Look for weak signals of the future, especially when they impose a fear on us - Interview with Jacintha Sheerder
How do you see the future in fifty years?
March 14, 2014 - Turning the CO2 we pump into the air into a valuable product sounds like a pretty good idea. It certainly would make it attractive for emitters to ensure they captured the gas because there would be a market for it.
March 14, 2014 - If you live in London in the United Kingdom as a pedestrian you may soon find it a lot safer to cross a street. New smart sensors and cameras are being tested that can detect a build up of pedestrians at a corner and adjust traffic signals to the volume.
In the post-industrial age, inefficiencies in the world are finally breaking down: interview with Stephan de Spiegeleire
Interview with Stephan de Spiegeleire, Senior scientist the The Hague Centre for Strategic Studies
How do you see the future in fifty years?
I personally expect a lot of positive changes to happen in the coming few decades. There remain enormous (self-imposed) inefficiencies in the world which we are finally starting to break down. The fact that so many opportunities for international trade arbitrage – most of them not a reflection of underlying endowments but of glaring political and regulatory inefficiencies – continue to exist is a perfectly example of this. Technology and globalization are starting to alter that – slowly, painfully but still quite decisively. We can in my opinion also expect major breakthroughs in both physical AND social technologies that will lead to quantum leaps in fields like education, health, and in many other spheres of private and public life, including even the (hopefully) ‘smarter’ use of armed force, something that we at HCSS are doing a lot of work on.
Classified data does not stay classified forever. Accordingly, two books written by former CIA employees provide a glimpse into the CIA’s inner workings and its role for the future in national security.