Buzz
On May 27, 1824, Ludwig van Beethoven conducted the first performance of his Ninth Symphony, ending with the majestic 'Ode to Joy' in the final movement. His 10th Symphony was highly awaited—but he passed away in 1827, before he could finish it.
A few decades later, the Austro-Bohemian composer Gustav Mahler made a deliberate effort to avoid a similar fate. According to his wife, Alma, his plan was to call his ninth symphony 'Das Lied von der Erde,' or 'The Song of the Earth,' to avoid the 'curse.' This would make his next symphony the 10th, but it would be numbered as the ninth.
However, despite the cleverness of his strategy, Mahler still succumbed to fatal pneumonia after completing only a sketch of his 10th Symphony in 1911.
The 'curse of the ninth' is a well-known superstition in classical music, affecting several renowned composers who passed away shortly after completing their ninth symphonies. Beethoven was the first victim, followed by British composer Ralph Vaughn Williams, Austrian conductor Anton Bruckner, and Czech composer Antonín Dvořák, all of whom are said to have been struck by this so-called curse.
Both Mahler and Beethoven left behind intriguing sketches of their 10th Symphonies. Now, computer scientists are developing AI algorithms to break the 'curse of the ninth' and complete the unfinished compositions of these classical geniuses.
AI Composition: The Beginning of the Journey
Using AI to compose music isn’t a modern idea: The history of algorithmic composition can be traced back to around 500 BCE. The Greek philosopher, mathematician, and music theorist Pythagoras recognized the connection between mathematics and music.
Between the 11th and 14th centuries, music theorists like Guido d’Arezzi and Franco of Cologne created systems for music notation, such as defining the time values of notes, pitches, and rhythms. This standardization allowed Western composers to create more complex compositions, influenced by various historical periods such as Baroque, Classical, and Romantic.
Due to the close connection between mathematics and music, the principles that govern pitch, rhythm, and harmonic progression in classical music are also programmable and interpretable by AI. This algorithmic analysis mirrors the process of human-composed classical music, which often starts with a motif or a few musical phrases, like the iconic 'dah-dah-dah-duh' from Beethoven's Fifth Symphony. Composers then expand these motifs into more intricate melodies and themes, creating a unified piece of music.
Back to the Fundamentals
According to Hugo Flores, a Ph.D. student at the Interactive Audio Lab at Northwestern University, AI composition follows a similar workflow. Flores, whose research explores the intersection of machine learning, signal processing, and music, provided an example of using AI and deep learning to compose in the style of Johann Sebastian Bach: 'I would format all the Bach cantatas into a single style and then train the machine learning model using those examples,' he explained to Mytour.
Much like human composers who develop motifs, AI composition involves allowing the AI to 'predict the next set of notes or the next measure based on the previous measures,' says Flores. In 2019, the Google Magenta and Google PAIR teams created an AI capable of creating four-part harmonization in the style of Bach using just two measures of melody.
That same year, Ali Nikrang, a senior researcher and artist at Ars Electronica Futurelab, along with Markus Poschner, chief conductor of the Bruckner Orchestra Linz, spearheaded the project 'Mahler-Unfinished' to complete Mahler's 10th Symphony. Nikrang’s team used MuseNet, a deep neural network that can adopt various musical styles, to generate four-minute musical compositions and bring Mahler’s unfinished work to life.
Nikrang shared that the team began with the first 10 notes of Mahler’s 10th Symphony—described as an 'unusual and dark theme,' in his words during an interview with Ars Electronica—and then allowed MuseNet to take over the composition. However, the melody MuseNet produced 'was only playable on the piano, and [the team] had to manually adjust it for the full orchestra.' The orchestration maintained the essential musical elements of the original composition, but in this case, 'the master was the AI algorithm.'
Tackling a Monumental Challenge
Professor Ahmed Elgammal, the director of the Art & AI Lab at Rutgers University, made a remarkable attempt at AI-driven music composition. He led a team of computer scientists at Playform AI to undertake the grand task of completing Beethoven's unfinished 10th Symphony.
Composing a symphony involves harmonizing multiple components and adhering to established rules. When the Beethoven project started in 2019, 'Most AI systems available at the time couldn’t extend an incomplete piece of music beyond a few seconds,' Elgammal explained in an article for The Conversation. Fortunately, Beethoven had left over 50 sketches that offered a glimpse into the full scope of the symphony. While these sketches provided valuable input for the AI, they were fragmented and nearly illegible due to his unique handwriting. To truly grasp the essence of Beethoven’s work, the team enlisted composers, musicologists, and music historians, aiming to teach the AI 'both Beethoven’s complete body of work and his creative process,' Elgammal writes.
After more than two years of effort, the project 'Beethoven X' was unveiled on October 9, 2021, with its world premiere performance on the same day in Bonn, Germany. Although the AI's composition contained echoes of Beethoven’s Fifth and Ninth Symphonies, Elgammal noted that audience members who weren’t experts in Beethoven's music couldn’t distinguish where Beethoven’s phrases ended and where the AI’s extrapolation began.
Should Human Composers Be Concerned?
If you're a composer, there's no reason to panic about the rise of AI in music composition. 'You can try to finish Beethoven’s final symphony, but there’s no way AI alone can fill in the gaps,' Flores explains. 'Beethoven wrote based on his daily life experiences.' A neural network wouldn't be able to capture the subtle details of life that would have shaped the composition.
In reality, music composition is deeply tied to the personal experiences of the composers. For example, researchers can teach an AI to replicate the blasting cannons in Tchaikovsky’s '1812 Overture.' But the AI would lack the understanding that these cannon sounds symbolize the Russian victory over Napoleon in 1812. Moreover, it wouldn't feel the spine-chilling effect the sound has on listeners. Ultimately, AI cannot fully replicate the emotion and life experiences embedded in a human composer’s work.
That being said, AI-generated music has made composition more accessible than ever before and can help non-musicians unlock their creative potential. Platforms like Amper let users create royalty-free music with specified length, genre, and instrumentation.
While the music produced may not be as groundbreaking as Beethoven's or Mahler’s symphonies, these creative tools eliminate the barriers to composing music, allowing beginners to bypass the complexities of reading sheet music or learning an instrument.
Computer scientists like Flores are constantly refining machine learning algorithms so that AI can better identify various instruments and musical patterns, all while ensuring that artists and music technologists remain involved in the process. 'After all, we’re creating tools for artists, not aiming to replace art or the creative environments,' he explains.
What lies beyond the curse of the ninth? Human creativity, augmented by AI.
