We have come a long way in the discussion to what is good music. Usually it is being fought over between generations, typically between parents and their children. My grandmother is still a big fan of classical music. My mum on the other hand has from an early age on shown more interest in the music of artists such as David Bowie or Queen, rather than Mozart. I on the other hand, as great as Queen is, or rather was, am also a big fan of techno music and I can guarantee you that neither my mum or my grandmother think that it is good music.
I will now predict how that discussion might go between myself and my future children. Maybe by the time they are around and start developing their own taste in music, my raving self will have calmed down a bit and will be a big fan of then old-men band Milky Chance. My children will maybe also like them, just as I still am fond of Queen. But their favorite artist will be someone like SKYGGE or YACHT. Sounds familiar? Probably not. SKYGGE is a producer who has gone beyond using synthesizers to create new sounds: his music is written using artificial intelligence (AI). Similarly, YACHT is a dance-pop band who has just released an album that they have written together with AI researchers and Google’s Magenta project (i.e. an open-source AI for generating audio and visual art) (Magenta, 2019). They have done so by using AI for basically their entire process, starting from the melodies and lyrics to the album cover and music videos. AI in this case means a software capable of learning from data and making predictions based upon what it has learnt (Forbes, 2016).
In this article I will focus on the melody creation and try to explain how it works. Essentially, instead of playing random notes the system interpolates from other music. To do this it first has to understand that music. It does so by listening to the songs and creating a model with which it can reproduce them (Magenta, 2019a). Initially this does not seem like a very special thing, as the output will be the same as the input. However, by having created this model the software is then able to create new sounds and melodies which are not simply random but fit in well in the context of the previously learnt music (Magenta, 2019b). This means that feeding such a system with songs by only one artist, it is able to compose new songs which sound as if they had been composed by that same artist. Even though YACHT used the tools to create the melodies they still select the ones they wanted and played around with the interpolation options the software had. Nevertheless, in the end the melodies were not written by a human being, but by a machine.
To me this illustrates another step in the evolution of music: from acoustic to electric and now to computer-created. What do you think? Was YACHT still the artist that created the new album or the AI researchers or maybe the computer itself?
SKYGGE – Magic Man
YACHT’s first AI-created single and video – (Downtown) Dancing
References:
Forbes. (2016). What Is The Difference Between Artificial Intelligence And Machine Learning?. Available at: https://www.forbes.com/sites/bernardmarr/2016/12/06/what-is-the-difference-between-artificial-intelligence-and-machine-learning/#33c7c552742b
Magenta (2019a). YACHT’s new album is powered by ML + Artists. Available at: https://magenta.tensorflow.org/chain-tripping
Magenta (2019b). MusicVAE: Creating a palette for musical scores with machine learning. Available at: https://magenta.tensorflow.org/music-vae
