Monday 1 April 2019

Warner Music signs distribution deal with AI generated music app Endel

By Hayleigh Bosher and first published on the IPKat 
AI and copyright is one of the hottest topics of the time in IP at the moment. This is hardlysurprising since it raises so many interesting questions about creativity and 
ownership. As Prof. Bernt Hugenholtz argued at the March Alicante Congress 
on AI and IP [reported here] that copyright protection – if any - for AI-generated 
works needs to be balanced against the overarching goals that such 
protection aims to achieve.

However, this has not stopped Warner Music signing a bundle of code to 
create 20 new albums this year! Warner Music is an American multinational 
entertainment and record label, the third largest in the global music industry, 
with artists from Ed Sheeran, Coldplay, Madonna to Led Zeppelin. 
The latest to be "signed" by the multi-billion dollar company, is an algorithm. 

Endel, developed by a start-up based in Berlin, creates tailor-made custom 
sound frequencies based on personal user inputs such as weather, time of day, 
location, and biometric details such as heart rate. 

Evidently, Endel is not signatory 
to the contract with Warner, as such. 
The company has agreed a 50/50 
distribution deal, covering a total of 
20 albums that will be released 
throughout 2019 with Warner Music. 
(Cat SamplingImage: Andy Miccone

This is not the first time AI-generated 
music has created distribution deals. 
Aiva Music is a composition 
algorithm that famously became the 
first AI to register with a collecting 
society (SACEM) and 
recently partnered with Believe Distribution (owned by Song Records) to release 
its latest album. Sony also has its  Flow Machines project, which involves 
a algorithmic composition tool that is formally credited as a songwriter, 
producer, instrumentalist and/or vocalist in all of the tracks’ liner notes for its 
debut album 'Hello World'. However, the creators also include a list of 
human contributors who provided  songwriting, instrumentation, 
mixing and mastering support.

The Endel-Warner deal is a step forward in that there are no human collaborators
 in the generation of the new sounds. Nevertheless, a human - aside from 
obviously creating the AI - also had to, input sounds and data into Endel. 
Interestingly Stavisky describes the work as being "generated based 
on different combinations of inputs" rather than created. 

Some of these inputs, or instrumental stems, were created by Endel's 
co-founder and sound designer Dmitry Evgrafov. Each sound is then 
allocated metadata according to certain parameters which the app 
can read and use to generate a soundscape. So, whilst it might seem that 
the sounds are created with a click of a button, Stavisky explained that it 
took "1.5 years of work developing our algorithm and creating and 
tagging the stems.” 

In terms of copyright ownership, on a theoretical level some argue that 
the creator of the AI might be the owners of the outputs, others suggest that
 it could be the AI system itself. Other potential owners could be the creator 
of the "inputs" on the basis that this is the personality being expressed. 
Or, perhaps, the investor of the AI project on an economic justification 
of remuneration as encouragement.

Endel: "Personalized sounds to help you focus and relax"
In any event, for Warner and Endel, 
it was practical issue. Stavitsky said 
that  when Warner asked for the 
songwriter information in order to 
register the  copyright of the music, 
they decided to list the co-founders 
and  software engineers, saying 
“I am now credited as a songwriter 
even though I have no idea 
how to write a song."

So, whilst we are still discussing what we think the outcome should be in theory, 
in practice some are marching ahead on the basis that the copyright holders 
are the company founders and the AI engineers. At this stage, given the 
extensive skill, labour and effort that went into the development of Endel
 it might not be so controversial. But what happens if [when] it is a self-learning 
machine that doesn't require as much human effort?

So many other questions come to mind - is Endel sampling? Do they need 
a licence? Who would be liable if Endel created an infringing piece of music? 
All of the registered copyright holders? The specific engineer who input a 
copyright protected work? But it is only if Endel uses a substantial part of 
that work in a new song that it would be infringing. Are the engineers able 
to programme the system not to take a substantial part? As we well know,
 its not about quantity in which case that might be viable, but since it is 
something decided on "quality", on a case by case basis, it's not so straight 

The future for copyright and AI remains to be seen, but it appears 
that the time is ripe to be discussing such issues!

No comments: