Image Credit: The elephants and the
The UK might have flopped in this year’s Eurovision Song Contest, but there is a new competition on the scene that we might have a better chance of winning. The AI Song Contest is an international competition in which teams across the globe submit four-minute tracks that have been created using AI. The competition which ran for the first time last year was initially created to see if a computer could help to make songs that are on par with Eurovision hits. One of the teams competing this year is The Elephants and the (yes you did read that correctly) – a band who took the brief very seriously and went so far as to use an AI-generator to create their name. We caught up with the University of York’s very own Dr. Tom Collins, founding member of the team, to find out more about AI music production.
In terms of the process, Tom explains that their algorithm works by analysing existing songs and finding similar moments, which in turn generates a “network of musical possibilities”, which is how they get “new sounding but stylistically coherent output”. Working in this way obviously comes with a whole new set of benefits and drawbacks. The main attraction for Tom though is the novelty of AI music production. “Even though I know the algorithms” he says, “I still don’t know what I’m going to get for a new set of input or training data. Sometimes you’re pleasantly surprised (or unpleasantly surprised!)”.
One benefit of using AI is that it creates many different possibilities. However, Tom did have to generate thirty melodies, thirty bass lines and thirty drum tracks for this project, and only around half of those, he says, sounded as though they had something promising in them. Despite this, Tom describes the process as one that is very exciting to be a part of; “it’s basically like working with a human musician but one that you can keep refreshing with different musical inputs.”
In the Eurovision Song Contest, the aim is to appeal to voters, often across language barriers. One question surrounding AI music then, is whether it is ever really possible to create music that taps into emotions as well as human expression can. Tom is very clear about the limitations of AI in this area; in his honest opinion, “it’s someway short of what I would call listenable music, or music that has human expression in it.” He explains that there is human input with choosing the instruments, “but at the end of the day we’re not synthesising the human voice from the lyrics or a melody line yet, and even if we did, I think that would have its own limitations in terms of imitating human capability.” One aspect of the project that does have a human element though is the saxophone part, which has “more dynamic and timing expression” than some of the other elements of the track.
Given these limitations, it seems a little far-fetched and black-mirror-esque to suggest that AI could ever ‘take over’ music production. But I did ask Tom whether he thought that AI could ever become a dominant form of music production or even put conventional musicians out of work. On this topic, he brings up a book called Robot-Proof, which is about how to ensure that higher education produces students who cannot be replaced in the world of work. In a musical vein, he suggests that in “areas that use music in a throw-away fashion- adverts where there isn’t much of a budget or YouTube videos where someone needs copyright-free music”, these gaps could definitely be filled by using algorithms. AI may play a role in automating something that a person was doing before, but Tom does not seem to think that it would ever replace humans entirely-”it would just be that the human focused on other aspects of the music production”.
On the band’s website, there is an option to remix an AI saxophone riff yourself. This got me wondering whether anyone could try their hand at larger-scale AI music production. In terms of his predictions for the future, Tom suggests that the music industry will move to a more two-way street in terms of creators and consumers being blended, and he’s confident that AI will come into that. Whether AI music production will produce successful outputs is obviously a product of chance and skill or expertise, but Tom does encourage others to try their hand at it. “Composition as a word just means putting things together, so why shouldn’t anyone have a go?”.
To listen to all thirty-eight entries from around the world and vote for Dr Tom Collins and his team’s track ‘Circus’, visit aisongcontest.com .