Artists and scientists harness AI to create songs worthy of Eurovision

No festivals, no concerts, no parties. Measures to contain the spread of COVID-19 have resulted in the cancellation of hundreds of music events around the world, including the world’s longest running television show– the Eurovision Song Contest.

For the first time since it premiered in 1956, no song will be crowned Europe’s favorite. Undeterred, a group of musicians, artists, scientists and programmers got together to create the ultimate Eurovision song, using artificial intelligence. 

Eurovision, for those unfamiliar with the phenomenon, is a yearly song contest created after World War 2 to unite a divided Europe. The first edition itself was a technological experiment, as it joined several European countries in a live broadcast using a terrestrial microwave network. Satellite television hadn’t yet been invented. 

The show has grown from just seven participating countries to a more global competition that allows 50 countries to take part, including Israel and Australia. (As early fans of the show, these two countries broadcast it for years; resulting in their inclusion.)

There’s no accounting for taste…

Eurovision is not Eurovision without friends, families and colleagues arguing passionately about who should claim the title. One person may love a disco-influenced beat, while another may prefer country-inspired ballad. All we can say is that there’s no accounting for taste! Or is there?

To try to crack the code of a Eurovision hit song, 13 teams from Australia, Sweden, Belgium, the UK, France, Germany, Switzerland and the Netherlands competed in the very first AI Song Contest. This was an online project sponsored by the Dutch public broadcaster NPO (VPRO), NPO Innovation and radio station 3FM.

Each team, which included both artists and scientists, fed AI models with huge amounts of musical data taken from existing hit songs. Rhythms, baselines, guitar riffs and lyrics were used to train the AI algorithms. The teams were also allowed to add new sounds or soundbites. Based on this data, the algorithm would predict the best sequence of notes and popular words, creating both new melodies and new lyrics. 

However, attempting to record the first winning AI Eurovision song was not the sole goal of this competition; it also aimed to conduct some valuable research on the creative skills of artificial intelligence and the impact it can have on artists and the music industry. 

The outcome of the competition were 13 very different songs, none of which would be out of place on the radio today. The jury, made up of music and AI experts, was surprised by the diversity, originality and innovation produced by the AI model. 

Humans are still required

However, although the sound machines created new and unique sound combinations, the jury also had to conclude that the creativity was mostly dependent on humans.

People decide what data to use when training the AI algorithms, and it is people who decide on what outputs to use. But although human creativity is required in producing music, AI can recognize well-liked patterns in previous successful songs. For example, people like to hear songs which have certain repetitive motives that they can recognize. Moreover, songs that have a distinctive vocal element are more likely to become a hit. 

On the other hand, people like to hear something new and authentic: a song should not sound forced. So, the best way to create a hit song is to take something recognizable and add a new layer to it. This is where AI shines. Where it would take a human a lifetime to try all possible combinations, a machine can do this is no time. Some songs may sound good, others may sound strange, but is up to the artist to make the actual choices. In the end, it is a question of taste and current trends. Often, a song does not become a hit when it is first released but becomes popular at a much later stage for no real explicable reasons. 

Kraftwerk, a German band, gained worldwide acclaim for their use of homemade electronic machines in their music making. Perhaps AI will become the next musical instrument that helps artists develop new musical genres.

In answering the question about AI’s creative ability, it seems that although AI can have a definite impact on the music industry, it is the harmonious combination of people’s intuition and machine prowess that leads to true musical innovation.

And the secret to writing the next Eurovision hit song? It’s finding the right balance between formulas, magic and, of course, a stroke of luck.

You can listen to the very first winner of the AI Eurovision Contest here: