A copyright Snafu in the making?

Are A&R scouts in the music industry next in the growing list of humans whose jobs are shortly to be appropriated by machine learning? Snafu Records seems to think so.

Snafu, which is backed by various music industry bigwigs, claims to have developed an algorithm that finds new music which is off the beaten track, and which will sell.

Sounds great. How does it work?

As would be expected in the case of a proprietary algorithm, public details are scant. We are told that Snafu’s search software scours the far corners of the Internet (on YouTube, SoundCloud, etc.) for around 150,000 tracks per week (i.e. far more than a team of humans could). The tracks are ranked by an algorithm according to listener engagement (taking into account factors such as user comments and listener growth) and the quality of the music itself (to which we will return). A weekly shortlist of 15 to 20 songs is then reviewed by (human) record executives, and Snafu then aims to sign the best artists to contracts. The fledgling artists who are contracted receive marketing support in exchange for a share of their streaming revenues. 

“But what does this have to do with IP?!” I hear you ask, five paragraphs into an IPKat post without so much as a Community Plant Variety in sight. Well, aside from copyright etc. in the software, Snafu’s model got this GuestKat thinking about the risk of inadvertently acquiring music which infringes third party copyright. As shown by many of my fellow GuestKat Hayleigh’s previous posts, the music industry is highly litigious – as the (overused) saying goes, “where there’s a hit, there’s a writ“.

Snafu’s algorithm is, seemingly, trained using the current Spotify top 200 songs. Snafu explicitly says that it looks for songs that are 70-75% similar to those that are currently popular. Surely, then, there is a heightened risk that Snafu’s algorithm will pick out songs that contain musical elements (e.g. a particular hook) that are shared with other songs that are already popular. From there, bearing in mind the relatively low threshold for originality under EU law, a third party copyright infringement claim may not be far away.

Sadly, we cannot look inside the Snafu “black box” to analyse this apparent risk in more detail. It is possible that the 70-75% criterion is (i) marketing puff and/or (ii) measured by reference to an amalgam of the existing top 200, which presumably would reduce the risk of taking a substantial part of / own intellectual creation in a single existing song. [Personally, this GuestKat is sceptical of (ii) given that this would presumably create a Frankensteinian reference point that would have no real value in determining which song might be the next huge choon.]

It is probably only a matter of time before an artist from the Snafu stable (16 and counting) finds success and a third party brings a copyright infringement claim. In that case, the Snafu algorithm and training records (which presumably may be subject to disclosure/discovery on a confidential basis) may hold some interesting evidence to support an infringement claim. Obviously, any case will turn on its own facts (in particular when and how each relevant song was created – i.e. the possibility of independent creation) but the involvement of a similarity-finding algorithm may at least create a presumption of copying and call into question the additional benefit of evidence from expert musicologists.

(Of course, this could work the other way around; perhaps Snafu can turn its algorithm into an additional revenue stream to find popular songs that possibly infringe Snafu artists’ earlier, less well-known, songs… The algorithm will not, however, help with a frequent stumbling block: whether the writer of a later song was exposed to the earlier song, and the related issue of unconscious copying).