Those following the ‘robots writing songs’ narrative might like to learn that things have progressed yet again.
On Tuesday (May 12), the results of the first ever AI Song Contest were announced, in a competition organized by Dutch broadcaster VPRO in cooperation with radio station NPO 3FM and NPO Innovation.
The AI Song Contest was designed as a playful nod to the Eurovision Song Contest – the reliably kitsch, televized annual music competition that usually (in a non-COVID year) draws a global TV audience of more than 180m people.
We say ‘playful nod’; it’s actually a bit more blatant than that. The pitch of the AI Song Contest literally reads: “The next major Eurovision artist might be a computer. In the AI Song Contest teams of musicians, artists, scientists and developers take on the challenge of creating a new Eurovision-like hit with the help of artificial intelligence.”
The winner of the Contest was partly decided by a panel of highly-qualified experts, including British exec Ed Newton-Rex, whose AI music company Jukedeck was acquired by TikTok owner Bytedance last year.
Newton-Rex’s fellow panellists were Netherlands-based AI researcher and composer Vincent Koops and US-based Anna Huang, formerly an AI Resident on the Magenta project at Google Brain, where she worked on generative models for music.
These three judges voted on a selection of AI-generated tracks, after which their scores were added together with the ‘public vote’ from an international online audience in order to crown a winner.
In the end, thanks largely to a chunky share of that audience vote, this track – Beautiful The World by Uncanny Valley – emerged victorious.
(Uncanny Valley, pictured, are actually from Australia, rather than Europe, but Australia are also participants in the real Eurovision Song Contest these days, so fair enough.)
According to VPRO, the track was “largely written using artificial intelligence”.
The team behind the song fed audio samples of animals native to Oz – including koalas, kookaburras and Tasmanian devils – to make it as reflective as possible of their homeland, and then they let AI do its work.
VPRO adds: “The audience’s enthusiasm for the Australian entry is a big boost for AI as a creative technique for the future. It shows that a computer can write a feel-good song with a catchy hook.”
“This shows that a computer can write a feel-good song with a catchy hook.”
AI Song Contest organizer, VPRO
Beautiful The World, though, wasn’t actually the favored track of the AI Song Contest’s judges.
The Critics’ Choice prize, as it were, went to the runner-up of the actual competition, the brilliantly-titled I’ll Marry You, Punk Come.
It was created, well, by Artificial Intelligence, working in tandem with German team Dadabots x Portrait XO.
After Australia (No.1) and Germany (No.2), at No.3, came Abbus by Dutch team Can AI Kick It?
Not going to lie, this is MBW’s personal favorite. (Ask yourself honestly, if it turned up in the lower reaches of the Billboard Hot 100, would you bat an eyelid?)
Described as a ‘duet’ between AI and artist Willie Wartaal, the track’s lyric was largely generated by machine, and ends with the controversial, machine-created line: “We want revolution, kill the government, kill the system!”
The Dutch team used Arran Lyon’s AI system to create hundreds of melodies and bass lines, which they then fed into another AI filter – Ashley Burgoyne’s Eurovision Hit Predictor – to surface the catchiest options.
The three AI Song Contest judges said in a statement: “We were amazed by the teams’ wide range of innovative approaches to using AI in their creative process in creating AI Song Contest songs. Every song felt very personal in different ways, and this reflects how the artistic vision of each of the teams drove how they collaborated with AI.
“Composing a song with AI is hard, because not only do you have all the creative challenges that come with writing a song, but you also have to juggle with getting the machine learning right. Working with machine learning is this constant push and pull, where you try your best to direct it in a certain direction with the data you have, priming it with your own tune, and then it steers you in another direction.
“Overall, we were delighted by the diversity and collaboration within teams, whose members… gave the audience a look into the exciting future of human-AI co-creativity in music.”
AI Song Contest Judges inc Bytedance’s Ed Newton-Rex
“We see teams embracing this unpredictableness, listening up close and finding inspiration that in some cases fueled the whole story of the song. We see teams leveraging the firehose of ideas that machine learning generates, which often requires you to listen to a massive number of samples before you find your gem. Some teams approached this problem by creating another model that evaluates and ranks the machine learning output for them, while also being sensitive to the ethical implications of using AI in songwriting.
“We are encouraged to see teams who might not have as much musical experience before find their musical voice through the use of AI. Overall, we were delighted by the diversity and collaboration within teams, whose members not only pushed the boundaries of their personal creativity, but also gave the audience a look into the exciting future of human-AI co-creativity in music.”
