Opinion: results to date show that artificial intelligence still has a very long way to go to emulate the work of human musicians

Google recently released their first ever AI-powered "doodle" in honour of Johann Sebastian Bach on the composer’s birthday. The program allows users to input their own melodies and the algorithm will attempt to harmonise them in the style of Bach, based on a dataset of 306 of the composer’s works. The results were underwhelming, at times even outright humorous, with the only certain takeaway being that AI still has a very long way to go in the field of music.

Although we have not yet arrived at a point where AI’s musical capabilities are worthy of discussion, it is worth determining whether or not there is a precedent for algorithmic approaches in music composition, not least in the music of Bach. Bach is regarded as the greatest master of Baroque counterpoint, a style of composition which involves highly complex, intricate polyphonic textures (the presence of and interaction between two or more simultaneous melodies).

Bach’s music, and indeed the counterpoint of the Baroque and Renaissance eras in general, does have an undeniably mathematical aspect to it from an analytical perspective. Accordingly, music students are taught the art of counterpoint as a series of steadfast rules, do’s and don’ts according to the context.

From RTÉ Lyric FM's Culture File, a day in the life of the industrious JS Bach

But to declare this music algorithmic would be highly inaccurate and the aforementioned rules are mostly the product of musical theorists of later eras. Indeed, for almost every such rule attributed to Bach’s style, there are instances to be found of him breaking it in his works. Ultimately, the music of Bach and of all great contrapuntal composers was as much a product of intuition and raw, inspired creativity as it was theory and craft.

It is arguable that the first instances of what could be considered overtly algorithmic composition occurred in the "twelve-tone" works of Josef Matthias Hauer and, of considerably more fame, Arnold Schoenberg and his disciples Anton Webern and Alban Berg. The goal of this music was to eradicate all traces of tonality, that is music written according to harmonic hierarchy around a central tone.  

This approach, also known as serialism, reached its logical conclusion in the later "total serialism" of Pierre Boulez and others, in which all the elements of a composition are determined by distinctly algorithmic means. The approach was undeniably cerebral and the results are correlatively palatable only to an extremely limited audience.

From RTÉ Radio 1's Arena, music critic and author Paul Griffiths remembers the maverick French classical composer  Pierre Boulez on the occasion of his death in January 2016

What Google have demonstrated with their Bach doodle is AI’s as of yet extremely limited capability to imitate an established style through machine learning analytical techniques. However, it would be a mistake to assume that what defined Bach’s music, and ultimately what made it great, was solely his study and consummate mastery of musical techniques. 

Bach began his musical studies at a very early age, having been born into an extremely musical family. He suffered tragedy early in life, both his parents having died by the time he was ten. He subsequently lived with his brother Johann Christoph Bach, who was an accomplished musician capable of continuing Johann Sebastian’s musical education.

Always eager to enhance his knowledge, Bach would secretly copy his brother’s prized collection of musical scores by moonlight, having been forbidden access to them. That Bach was not easily deterred by obstacles to progress is further evidenced by the report that he once travelled 450 kilometres each way on foot (and without leave from his employers) to meet and hear the renowned composer and organist Buxtehude, whose music he greatly admired. 

READ: Are we outsourcing our emotions to the Spotify algorithm?

A huge portion of Bach’s immensely prolific output was composed for religious purposes, and religion was always a profound source of inspiration for him artistically. "The final aim and reason of all music is nothing other than the glorification of God and the refreshment of the spirit", he said.

While AI will undoubtedly become ever more adept at imitating existing music, it will never be able to emulate the individual life, emotions and personality that created the music. In fact, it is likely that if AI were ever to create what could be deemed genuinely expressive music, it would express a state of existence utterly different to the human condition. The resulting music would probably be equally unrelatable to human ears – not too unlike the codified, algorithmic music of the total serialists.

From RTÉ Radio 1's The Business, more and more movie studios are using artificial intelligence to decide whether or not a movie should be made.

Nonetheless, it is likely that AI will eventually factor into the composition of music to at least some degree, most probably in the fields of commercial pop music and also in music for film, television and advertisements. In this hypothetical scenario, a director may hesitate to hire a composer if they could simply push a button and have an algorithm compose and produce music they are willing to use. 

However, by the same logic, the hypothetical parsimonious director and their entire crew could also find themselves out of the job. Future films and TV shows would be created entirely by algorithms, complete with fully realistic looking (and sounding) computer generated actors and scenery.

Grim as this prognosis may seem, it certainly does not spell the end for artists. If anything, AI generated art works will most likely increase the value of, and likewise heighten the appreciation for ‘human-made’ art, at least for a decent (and discerning) portion of the population. After all, who wants to listen to an algorithm?


The views expressed here are those of the author and do not represent or reflect the views of RTÉ