Does AI Have The Potential To Create a Song That Tops The Charts?
While developments are happening gradually so that AI is able to mimic the inner workings of the human brain, and even start to eloquently express itself in the form of writing, success has been limited since we only have so much of an understanding of what goes on inside our heads. But songs are a completely different challenge, where the words do not matter as much as the rhythm and beat.
Recently a team of Dutch academics utilised an artificial intelligence algorithm to write a song. The team, known as ‘Can Ai Kick It’ used data from the previously popular songs in the Eurovision contest. Much to their surprise, the outcome was a completely unique genre, songs with a robotic voice that idolised rebellion and “fighting the system”, creating its own genre: Technofear.
What goes beyond the simple idea of creating a song is the effect that it can have on society. The themes of anarchy and chaos that form the lyrics of the song display just how much of an influence the product of an algorithm can be. However, like the Tay chatbot developed by Microsoft in 2016 that started using racist and sexist comments after being trained on Twitter, the fault lay with the comments made by users on twitter used as data, essentially the shortcomings of our society, not the algorithms.
Yet there is still a huge interest in AI for the music industry, as artists look towards algorithms to produce unique sounds and possibly inspire new genres of music. The Dutch Broadcaster VPRO holds an annual AI song contest to present the commercial value of using these algorithms in a competitive industry. This year's entries consisted of songs about anarchy from the Dutch team, a chart topping dance hit from Australia with sound bites from koalas, kookaburras and Tasmanian Devils and death metal vocals from Germany, a slightly strange but surprisingly rhythmic range of songs.
The Dutch team also did analysis using AI on the types of music that are generally winners at Eurovision contests and the elements that create it. Their findings suggested that melodies with ranges of three to seven notes and songs with simple rhythmic patterns were the most popular, while also showing a preference for high tonality, where it is hard for the ear to identify the key. It also showed that a certain level of atonality – where it is hard for the ear to identify the key – was crucial to Eurovision success. A key part of training the algorithm that academics stressed was the use of negative information in their machine learning algorithms, in order to successfully train the system and improve accuracy. This type of analysis can then be used to train the songwriting algorithm to produce a song that will be liked by the majority of the mass population.
The hope for this contest is to not provide competition to an already large industry of songwriters, singers and music technicians, but to instead highlight the capability of AI to create unique sounds - there are reports that AI has already begun to utilise sophisticated drum machines, arpeggiators and master the software needed for composition. But songwriting isn’t like creating a scientific model or determining statistics, there is no right and wrong, at the end of the day, although AI may be able to produce sounds never heard before, it will be long before they are able to mimic the creative that some of the leading artists in the music industry have. Songwriter and artists that fear the growth of AI songs and the future of music should instead utilise the technology to its full potential, creating a future of new, unique and appealing songs for a very long time.