The Music Department at Otago University celebrated 100 years of existence recently, launching a book edited by Ian Chapman called Ka Muri, Ka Mua, referring to the famous Māori whakatauki which translates as “I walk backwards into my future”.
The celebrations revolved as much around the future as it did around the past.
There was an interesting panel discussion featuring staff and postgraduate students on the brave new world of AI in music. Attendees were shown how easy it is to create a song online via an AI site.
Following a few simple prompts in a matter of seconds the programme was able to produce a song about spoonbills in the Ska style. Despite the ridiculous content it sounded convincing enough as an imitation of that style.
To do this, AI has “scraped” millions of human-produced songs from around the world and stored them in vast data banks. They are then used as raw material to produce a derivative of the real thing, and also converted into "training data" where even the smallest musical elements (tempo, keys, timbre) may be used in future AI-generated music.
In many cases there is no acknowledgement of the human origins and no royalty has been paid. The human mahi has essentially been taken and re-hashed into a marketable product. Copyright organisations and others are trying to catch up with all this and create laws that offer the human artists protection, but really the horse has bolted.
Earlier this year I collaborated with my wife on a song for a choir in Auckland. Sandy wrote the lyrics and I wrote the music.
We were telling someone about this recently and their first question was “Did you use AI?”. My friend and artist, Paul Wheeler, has had similar reactions to his imaginative, semi‑realistic digital art.
The advent of AI is making people suspicious of the provenance of human art. I am thinking of writing this on my title pages: “Made by AR not AI.”
Musicians are now in a similar situation to other producers whose products are imitated with impunity. Think fake Rolex watches, or fake Cartier handbags. Think fake Madonna, fake pop artists - or fake Mozart.
There are many layers to the fakery, as postgraduate student Grace Shaw recently demonstrated in one of her presentations on AI in popular music, showing how many AI versions of the famous song My Way exist.
Some people don’t see the harm in having AI generated imitations; if you like the sound of it where is the problem? After all, we already have plenty of mass-produced music in our lives, in elevators, supermarkets, malls and so on.
Would it really matter if that music was AI generated? Perhaps not, but many musicians have made a living from creating this type of music, as well as film and game music; they are losing their incomes.
A reaction is already setting in, with many seeking authentic musical experiences rather than machine-made ones.
It is, however, possible to take a more sympathetic or rational approach to AI in music and simply regard it as another new technology.
Think how the invention of the piano transformed music, or the invention of recording. It might be that AI will spawn new types of purely electronic music that humans couldn’t possibly dream up.
Perhaps, in the future, we might have to divide music into two or three types according to what made it: human-made, AI made and Cyborg-made.
Classical music, my speciality, has been slow to catch up with all this but is not immune to this challenge.
For me, the saving grace is live performance. People will still want to go to concerts in future and see humans perform because, essentially, we are a communal species who like communal experiences.
There will, of course, be robotic performers coming our way soon but I’m hopeful they will be regarded somewhat differently to humans.
As to whether composers will maintain their status as original creators compared with AI imitators remains to be seen.
In an interesting and provocative dissertation by postgraduate student Neil Pickering he argued that we cannot prove that AI is not creative. In other words, it is possible that AI produces music that is novel and valuable and creates it in a similar way to humans, drawing on music and musical knowledge that has preceded it; therefore it could be capable of being ``creative’’.
Arguments against this concerning the human process, inspiration, authenticity etc don’t alter the fact that AI creates music that, if you were not told it was by AI you might think was human-made.
My doubts about AI creativity focus mainly on originality and the development of a unique musical ``voice’’ or style. The composition process involves plenty of intuitive and even random decisions as well as more rational ones. These often happen during improvisation.
Composers are not always very logical but they do have the ability to throw up combinations of new ideas that, over a long period of time, can lead to a new style of music.
Take, for example, Douglas Lilburn, arguably New Zealand’s leading composer in the 20th century. As a young man he was influenced by Sibelius; you can hear Sibelius-like motifs transformed in his work.
Sibelius himself was strongly influenced by Tchaikovsky but, again, these influences were transformed into what we think of as Sibelius’ original style. Tchaikovsky was influenced by Schumann, who was influenced by Beethoven, and so on.
Now, compare Beethoven and Lilburn, over 100 years apart. They are chalk
and cheese.
Through countless human innovations we get from one style to a completely different one over time.
I would like to think AI won’t be able to evolve musical styles like this, but maybe I’m wrong.
Let’s see in a 100 years time.
- Anthony Ritchie is a composer and emeritus Professor of Music, University of Otago.