Hey Mister DJ

Robots vs. pop stars: Who is better at curating your music?

Stay on Top of Enterprise Technology Trends

Get updates impacting your industry from our GigaOm Research Community
Join the Community!

The Apple rumor mill is getting its groove on these days as new details appear about a revamped streaming service slated to be launched in the coming months. 9to5Mac reported last week that Apple is working on a new music service that uses some of Beats Music’s technology, but is going to be deeply integrated into iTunes and iOS.

Business Insider followed up with another report Monday, suggesting that the project will feature curated streams from well-known musicians. [company]Apple[/company] also recently hired BBC Radio DJ Zane Lowe, and is looking for music journalists who could be writing copy for the new service. All of this suggests that the company is looking to keep Beats Music’s focus on human curation and build a more radio-like experience, possibly with help from many music celebrities.

The question is: Do music fans really want this? Do musicians make for good DJs, and do well-known names help to unlock the 30-plus million song catalog of a music streaming service?

Or would algorithms simply do a better job?

Musicologists and trillions of data points

The debate over human versus automated curation is almost as old as online music itself. [company]Pandora[/company] was one of the first services to embrace the idea of human curation in a personalized streaming environment when it built its Music Genome Project back in 1999.

The idea at the time was to not simply play songs because algorithms deemed them as a logical choice based on the behavior of other users, but actually figure out how each song sounds, which instruments it features and which tempo it uses. Pandora hired dozens of curators to catalog more than one million songs based on up to 450 musical criteria, and its website describes these curators like this:

[blockquote person=”” attribution=””]“The typical music analyst working on the Music Genome Project has a four-year degree in music theory, composition or performance, has passed through a selective screening process and has completed intensive training in the Music Genome’s rigorous and precise methodology. To qualify for the work, analysts must have a firm grounding in music theory, including familiarity with a wide range of styles and sounds.”[/blockquote]

Today, Pandora still relies on the Music Genome Project, but it is also using algorithms and data to make its playlists work.

Others took a different approach and ditched the human expert altogether, instead relying on the wisdom of the crowds and big data analysis to generate that perfect playlist. [company]The Echo Nest[/company], for example, which was acquired by Spotify a year ago, is using close to 1.2 trillion data points on more than 36 million songs to automatically generate playlists for Spotify and other services. The Echo Nest co-founder Brian Whitman will be at our Structure Data conference in New York next month to tell us how he wants to use all that data to reinvent the music industry.

Park rangers, not gatekeepers

Lately, the pendulum has swung back to human curation, with Beats putting a heavy emphasis on its expert curators, and Slacker building a radio-like experience around YouTube stars and other personalities. The reports about Apple’s plans now seem to suggest that the company wants to go further down that road, embracing stars to become both brand ambassadors and actual curators of your music.

However, not everyone is convinced that this is a good idea. Online music industry veteran Tim Quirk, who used to head music programming for pioneering streaming service Rhapsody and then did the same thing for Google Play Music, took to Twitter today to object to the idea that musicians make good curators. Here are some highlights of his arguments:

[pullquote person=”” attribution=”” id=”917022″]Will the future of music look like Sirius XM or like Netflix?[/pullquote]

Of course, many will argue that there is value to expertise, and point to great radio DJs, so of which even are musicians. That’s why I asked Quirk what it takes to bring this kind of personality-driven curation to streaming services. His answer:

“Subtract the personalities. Seriously. They need curation that doesn’t brag about itself.”

In the end, this may all come down to the question what music services want to be, and how they plan to appeal to millions of consumers who have thus far shied away from music subscriptions. Do they want to be more like traditional radio and guide listeners through a catalog of millions of songs? Or do they want to be the celestial jukebox that brings millions of songs to your fingertips, ready for you to go on your own adventure?

In other words: Will the future of music look like Sirius XM or like Netflix? The first company to find a compelling answer to that question may be able to really take music subscriptions mainstream — with or without celebrity DJs.

5 Responses to “Robots vs. pop stars: Who is better at curating your music?”

  1. The accuracy of algorithm-based playlist compilation is dependent on the quality and quantity of the metadata associated with each track in a music library. At the moment song metadata is limited and not particularly accurate, but as it improves, algorithms will be able to create extremely relevant playlists based on an individual’s sex, age, location, existing music library, historic listening habits, activity, … even heart rate and weather. No human, even the listener themselves, will be able to create better playlists which continually evolve based on context and environment. Unfortunately we are not quite there yet so we may have to put up with the musings of Bieber, Bon Jovi and Barry White in the meantime!

  2. Richard P

    LAUNCHcast (subsequently acquired by Yahoo!, becoming Yahoo! Music) was using algorithms quite effectively in the 1990s. The Listener would be asked to rate songs according to his/her personal preference. The algorithm would deliver “new” as-yet unheard music to the Listener, based on the ratings of other people whose ratings profile was similar to that of the Listener. As a long-time user of that service, I thought the algorithm did an excellent job of discovering new music for me. I was sorry when Yahoo! dumbed-down the product and then sold it to Rhapsody.

  3. Personally I have not seen examples of good music curation by robots. Doubt pop-stars are better curators though. i think music listeners would be better of with those that allready curate music: bloggers, recordstore clerks, etc.
    I used to be an avid music buyer from the mid 90ies until ’07 (and then I became a dad and no longer had the funds… or time…) but there were are couple of record store clercks that i trusted 100% with being able to hand me a stack of records across various genres that were all mostly my taste. Same goes for blogs. I frequented a handful of blogs that consistently put me on to new (and old) music that i loved. I guess for every genre there are a couple of those curators. So if they sign up Antal from RushHour or Serge/Klen from Clone as curators or Alwin, I’ll make use of the service.

  4. So, it comes down to two entirely different things for me and my needs can change on a dime. One minute I am really interested in discovering new stuff or not knowing what song follows (the radio effect) and the next minute I want to hear stuff that is familiar and in my comfort zone and many times I do not even know what I want until I hear it. There is a third aspect, feeling of community by following and enjoying other’s musical tastes that isn’t really a factor for me.

    I think the latter is what drives human curated playlists, and would be very difficult for an algorithm to reproduce because it requires a listener to establish a relationship with the curator. Of course, people follow virtual pop-stars, so it may be more possible than occurs to me.

    The problem is this is a very personal situation and no person or algorithm is going to completely figure that out. Maybe when health sensors are much more accurate and can infer a person’s emotional state and understand what responses work best for that individual, then an algorithm might get it mostly right, but a human curator is only going to go on what fits their personal needs, so I don’t think personalization will work in those instances.

    I’m betting it will be a blending of data analysis and human actors to front the plan…

    …which is how modern radio works. Funny how that works out.