On May 27, 1824, Ludwig van Beethoven conducted the premiere of his Ninth Symphony, concluding with the glorious “Ode to Joy” in the last movement. His 10th Symphony was eagerly anticipated—but he died in 1827, before he could complete it.

A few decades after that tragedy befell his predecessor, the Austro-Bohemian composer Gustav Mahler went out of his way to prevent history from repeating itself. According to his wife, Alma, his grand scheme was to name his ninth symphony “Das Lied von der Erde,” or “The Song of the Earth,” instead of numbering it. That way, his next symphony would become his 10th, but be numbered ninth instead.

But as ingenious as this plan sounds, Mahler still contracted fatal pneumonia after drafting a sketch of his 10th Symphony in 1911.

The “curse of the ninth” is a popular superstition in classical music, cast upon several famous composers who died shortly after writing their ninth symphonies. Beethoven was the first; British composer Ralph Vaughn Williams, Austrian maestro Anton Bruckner, and Czech master Antonín Dvořák are also said to have been struck by the so-called curse.

Mahler and Beethoven left several tantalizing blueprints of their 10th Symphonies behind. Now, computer scientists are developing algorithms for artificial intelligence (AI) to lift the “curse of the ninth” and complete the unfinished works of these classical masters.

AI Composition: An Origin Story

Using AI to do the job of human composers isn’t a new phenomenon: The history of algorithmic composition can be traced back to about 500 BCE. At the time, the Greek philosopher, mathematician, and music theorist Pythagoras noted the relationship between mathematics and music.

From the 11th to the 14th centuries, music theorists like Guido d’Arezzi and Franco of Cologne established rules for music notations, such as time values of single notes, pitches, and rhythms. Such standardization allowed Western composers to develop more sophisticated practices in composition, imbued with characteristics of different historical periods like Baroque, Classical, and Romantic.

Thanks to the tight-knit relationship between mathematics and music, the rules that dictate the pitch, rhythm, and harmonic progression in classical music are also programmable and interpretable to AI. That algorithmic analysis mimics the process of human-composed classical music, which starts with one or a few motifs, or phrases of musical ideas, like the famous “dah-dah-dah-duh” at the opening of Beethoven’s Fifth Symphony. Composers then develop the motifs into more complex melodies and themes, weaving together a cohesive piece of music.

Bach to Basics

AI composition follows a similar workflow, according to Hugo Flores, a Ph.D. student at the Interactive Audio Lab at Northwestern University. His research focuses on the intersection of machine learning, signal processing, and music. Flores gave an example of composing cantatas in Johann Sebastian Bach’s style using AI and deep learning: “I would put all the Bach cantatas into one single format and train the machine learning model using those examples,” he tells Mental Floss.

Like human composers developing a motif, the key to AI composition is to let the AI “predict the next set of notes or the next measure given the previous measures,” he says. In 2019, the Google Magenta and Google PAIR teams designed an AI that creates four-part harmonization in the style of Bach from two measures of melody.

In the same year, Ali Nikrang, a senior researcher and artist at the Ars Electronica Futurelab, in collaboration with Markus Poschner, chief conductor of the Bruckner Orchestra Linz, led the effort to complete Mahler's 10th Symphony for the project “Mahler-Unfinished.” Nikrang’s team implemented MuseNet, a deep neural network that adopts various musical styles to generate four-minute musical compositions, to flesh out the unfinished work.

Nikrang explained that the team started with the first 10 notes of the 10th Symphony—an “unusual and dark theme,” he said in an interview with Ars Electronica—and let MuseNet take over the composition. However, the melody that MuseNet generated “was only playable on the piano, and [the team] had to edit it for the big orchestra by hand.” Their orchestration largely retained the musically relevant content of the master’s blueprint, but in their case, “the master was the AI algorithm.”

Conquering A Grand Challenge

Professor Ahmed Elgammal, director of the Art & AI Lab at Rutgers University, made an even more heroic attempt at AI music composition. He led a team of computer scientists at Playform AI to conquer the grand challenge of completing Beethoven’s unfinished 10th Symphony.

Composing a symphony involves many parts to harmonize and rules to follow. When the Beethoven project began in 2019, “Most AI available at the time couldn’t continue an uncompleted piece of music beyond a few additional seconds,” Elgammal explained in an article for The Conversation. Fortunately, Beethoven left more than 50 sketches behind that alluded to a complete picture of this symphony. Though the sketches can serve as excellent input for the AI, they are fragmentary and almost indecipherable due to his idiosyncratic handwriting. To truly capture the essence of Beethoven’s composition, the team also brought on composers, musicologists, and musical historians, intending to teach the AI “both Beethoven’s entire body of work and his creative process,” Elgammal writes.

Their effort of more than two years, “Beethoven X,” was released on October 9, 2021, with the world premiere performance on the same day in Bonn, Germany. While hints of his Fifth and Ninth Symphonies scatter throughout the AI’s composition, according to Elgammal, audience members who weren’t experts in Beethoven’s composition couldn’t tell where Beethoven’s phrases ended and where the AI extrapolation began.

Should Human Composers Be Worried?

If you’re a composer, there’s no need to fret in the face of innovations in AI composition. “You can try to finish Beethoven’s last symphony, but there’s no way that you can fill in the gap” with AI alone, Flores explains, “because Beethoven wrote based on his daily experience. A neural network would not be able to predict the nuances of life that would have filtered into the piece.

In fact, music composition is full of nuances rooted in the lived experience of the composers. For example, researchers can train an AI to recognize and mimic the blasting cannons in Tchaikovsky’s “1812 Overture.” But the AI would not know those cannon sounds represent the victory of Russians resisting the Napoleonic invasion in 1812. Nor would the AI experience the chill that the cannon sounds send down one’s spine. In other words, nothing can truly replicate and extend the life and emotion of a human composer.

However, thanks to AI music generation, music composition is now more accessible than ever and can help non-musicians unleash their creativity. Platforms such as Amper allow users to create royalty-free music through AI with defined length, genre, and instrumentation.

Though the generated music may not be as trailblazing as Beethoven’s and Mahler’s symphonies, those creative outlets break down the barriers to writing music, sparing novices the intimidation of reading sheet music and learning a musical instrument.

Computer scientists like Flores are also continuously improving the machine learning algorithms so that AI can better recognize different instruments and musical patterns while keeping the artists and music technologists in the loop. “Because, after all, we’re trying to make tools for artists, not to replace the arts and setups,” he says.

What lies beyond the curse of the ninth? Human creativity, empowered by AI.

Related Tags