Could Daft Punk Actually Be Robots? How Machine Learning Could Redefine Musical Creativity at Next Big Sound

In the realm of music, Next Big Sound employs a data-centric approach to artist growth and strategy. Where future value lies for the company is in applying a similar model to the underlying product: music itself.

A portrait of Edmond Belamy, a fictional man, was created entirely by Artificial Intelligence and sold at auction in October 2018 for $432,5001. The sale raises interesting questions: what distinguishes art from data? Can the richness of creativity be emulated by the contents of a spreadsheet? In the realm of music, Next Big Sound, a recent aquiree of internet radio giant Pandora, employs a data-centric approach to artist growth and strategy. Where future value lies for the company is in applying a similar model to the underlying product: music itself.

Next Big Sound (NBS) has a clear value proposition for its users: leveraging data analytics and predictive algorithms to create actionable insights for music artists and labels. For Artists and Repertoire (A&R) executives at major labels, whose primary responsibility is to sign undiscovered talent to record deals, there is value in knowing that “musicians who gain 20,000 to 50,000 Facebook fans in one month are four times more likely to eventually reach 1 million” fans2. By analyzing past growth patterns of current successful artists, NBS develops predictive algorithms to spot early prospects. Armed with NBS’s projections, labels can gain competitive advantage by improving their scouting processes and signing promising artists earlier on in their careers.

From the artist perspective, the sheer number of explanatory variables on which to base career-driving decisions is staggering: Twitter mentions, SoundCloud streams, Spotify playlist adds, blog coverage, radio spins, iTunes downloads and more. By aggregating this data NBS can offer invaluable insights to artists: from optimal album release dates to expected tour ticket sales by demographic. Most of the data harvested by NBS is publicly available, and competitors could replicate these predictive metrics. Where NBS can differentiate its service (and has to some extent already) is in its recommendation algorithms.

Predictions help capture existing value, recommendations create new sources of value. Forecasting ticket sales among students in Boulder, Colorado on a Thursday in November can be useful. But beyond city-specific predictions, NBS can also recommend potentially untapped markets that artists should be targeting. In an era where artist earnings from touring often exceed those from streaming royalties and sales, a well-timed, optimally routed tour can make an artist’s career3.

Serving strategic recommendations to artists is a competency that could fundamentally reshape the music industry and elevate NBS to juggernaut status. NBS simply must expand its offerings from logistical suggestions to creative directions. To maximize enterprise value, NBS should leverage its data to develop three core technologies: a feedback mechanism rooted in their “predictive success” model4, a creative direction algorithm using target demographics as inputs, and a music-producing computer system.

NBS can recommend which songs artists should select as their promotional “singles” based on past performance of similar songs5. Functionally, NBS’s algorithms are accurately recognizing if a piece of music will be popular2. Consider this capability as a feedback service: artists upload an unreleased project (e.g. a partially finished musical demo) to the service, and NBS predicts whether the music will perform well commercially. Further, the algorithm could extract successful traits from popular songs and suggest what changes or additions artists should make to their works to improve commercial viability.

Taken a step further, NBS should develop a creative direction algorithm for artists by taking Pandora’s listener recommendation engine and reversing it. Pandora offers ideal listening experiences to its users by “algorithmically curat[ing playlists] based on an analysis of data from sensors in users’ mobile devices, the users’ previous music listening behaviour, users’ relationships with other humans via social media, and acoustic characteristics of millions of songs available in the service’s music library”6. The same formula, applied in reverse, could be an invaluable service for artists: plug in their target demographic (e.g. ages 24-30, Swedish, college educated, female) and the algorithm dictates specific musical choices artists should make in their creation process to appeal to that demographic: instrument combinations, beats per minute, song length, rhythm syncopations, lyrical content and more. With these insights, artists could better cater music to their existing fan base, or specifically target new profitable demographics to spur growth.

At the extreme, NBS could challenge the very existence of musical artists. If the company’s data analytics can predict what type of music will be enjoyed by any demographic and which elements of this music stand to make a particular song successful: could it not harness these technologies to create its own music entirely with AI? These days, some major artists perform anonymously (Daft Punk) or as virtual bands (Gorillaz)7. If NBS were to successfully produce hit songs using only their prediction and recommendation algorithms, attributing the music to an anonymous or virtual act, would anyone know? If the next chart topper resulted from machine learning, would anyone be incensed? Perhaps Edmond Belamy has the answers.

(Word Count 792)

 

Citations

[1] “Is Artificial Intelligence Set to Become Art’s Next Medium?” Christie’s, 16 Oct. 2018, www.christies.com/features/A-collaboration-between-two-artists-one-human-one-a-machine-9332-1.aspx.

[2] Greenburg, Zack O’Malley. “Moneyball For Music: The Rise of Next Big Sound.” Forbes, Forbes Magazine, 13 Feb. 2013, www.forbes.com/sites/zackomalleygreenburg/2013/02/13/moneyball-for-music-the-rise-of-next-big-sound/.

[3] Titlow, John Paul. “Inside Pandora’s Plan To Reinvent Itself-And Beat Back Apple And Spotify.” Fast Company, Fast Company, 14 Apr. 2017, www.fastcompany.com/3058719/inside-pandoras-plan-to-reinvent-itself-and-beat-back-apple-and-sp.

[4] U.S. Patent Application No. 14/302/200, Publication No. US 2015/0032673 A1 (filed Jun. 11, 2014)(Victor HU, Alex WHITE, applicants).

[5] Bonazzo, John. “Next Big Industry to Embrace Moneyball: The Music Business.” Observer, Observer, 4 Jan. 2017, observer.com/2017/01/pandora-next-big-sound-moneyball-music/.

[6] Wikstrom, Patrik (2015)
Will algorithmic playlist curation be the end of music stardom? Journal of Business Anthropology, 4(2), pp. 278-284.

[7] Wired Staff. “Keeping It (Un)Real.” Wired, Conde Nast, 27 July 2018, www.wired.com/2005/07/gorillaz-2/.

Previous:

Image recognition at Facebook: How machine learning is helping computers — and people who are blind — ’see’ digital photos

Next:

Rent the Runway Wants You to Clean Out Your Closet…Permanently

4 thoughts on “Could Daft Punk Actually Be Robots? How Machine Learning Could Redefine Musical Creativity at Next Big Sound

  1. Extremely interesting to see how AI as a value-adding technology for the artist might at some point become a threat for the very same stakeholder.

    Already today, startups work on AI technology that develops new musical content, mostly used by e.g., movie directors, game studios, or advertisement agencies. An example would be Aiva Technologies, a startup that developed Aiva – the first AI that has the official status as Composer, with copyrights to its own name.

    The question here would be – can this technology ever fully replace artists? And if yes, what would be the long-term limitations? Can AI rethink music, create new genres, be disruptive? Or would be only cater to mainstream tastes, led by more dominant data input vs. more creative, alternative, “underground” art?

  2. The key point to answer the question if AI can actually replace an Artist is how disruptive new musics have to be in order to be successful.

    Since AI will work only on past data to create new content, one could argue that AI will hardly create something disruptive in the music industry. On the other hand, as the article mentioned, since there are several different controllable variables when creating music content (rhythm syncopation, beats per minute, etc), different combination of this attributes might create something that could be seen as disruptive content in the Industry. However, in this case, there won’t be any historical data that will prove the success of this new combination, making AI less useful. Moreover, there are several other factors besides music attributes in an Artist that are probably powerful indicators
    of an artist’s success likelihood (lyrics, looks, etc). These can also be hard to create using exclusively AI. Thus, I see AI more as an ally of music creation process than as a substitute.

  3. Reading the Christie’s article about the portrait of Edmond Belamy, it is interesting that an artificial intelligence has been able to create what humans would think of as “Art”, but I note how the machine only created the portrait after being fed numerous other pictures. I am skeptical that a machine will be able to truly create art using past data analysis. While chaotic on a larger scale, human thoughts and feelings can in some way be anticipated and expected (as studied by psychologists, sociologists, cognitive scientists, etc) and a sufficiently advanced learning algorithm may be able to analyze these trends to determine whether or not a song/piece of art will have mainstream appeal. I do think, however, that this would probably only work on a shorter time scale. Emergent memes and cultural phenomena seemingly change at random, what is viral today may be forgotten by tomorrow, a machine algorithm based on past data will almost certainly miss the most current or upcoming trends and will probably respond much slower than a human mind looking at that same cultural movement. With enough data on a large enough scale, New Big Sound may have found the solution but I will remain doubtful for now.

  4. Absolutely fascinating. I think you need to go sell the idea of building a popularity analysis tool to spotify and pandora. I can see a few different ways to build that product. Firstly – you need consumption data around a large bank of music (spotify and pandora), secondly you would require the master tracks to any given song or at least musical notation so that melody and rhythm can be analyzed. AI would be able to recognize patterns in popular songs on a variety of components, this should probably be segmented by genre. Spotify has the added machine learning component of being able to measure the popularity of songs analyzed by it, then deployed on the spotify application.

Leave a comment