Something Old to Something New: From Bach to AI

April 6, 2017 § 5 Comments

Composers can spend their entire lives studying the music of deceased composers before them in order to find inspiration for their own compositions. They can then spend years drafting manuscript after manuscript before they finish their work. However, the process doesn’t end there. After music is composed, the composer still has to find musicians willing to play the piece and people willing to listen.

In Cloud Atlas, it takes Frobisher, a struggling composer, quite a while before he even finds inspiration through Arys for the Cloud Atlas Sextet. What if in the time that it took Frobisher to compose The Cloud Atlas Sextet, a computer program could immediately compose five thousand different pieces of music based on the composition style of Frobisher? What if Frobisher’s musical thoughts and ideas could be extracted from the source and recombined into an entirely new composition that still retained every musical nuance of Frobisher as a composer? In this case, Arys wouldn’t have even needed Frobisher; he could have just used the computer program to compose his music. What if I told you that this computer program exists and has composed thousands of different compositions in the style of other composers, and it has even combined the influences of multiple composers into its own unique voice?

In the 1990s, David Cope, Professor of Music Theory and Composition at the University of California-Santa Cruz, created the program Experiments in Musical Intelligence, or EMI for short. EMI is able to take the notes in existing music and convert it into data. The data consists of numbers corresponding to the frequency (pitch), duration (rhythm), speed (tempo), volume (dynamics), and articulation of each note, converting the entire composition into numbers. EMI then analyzes the all of the data, looking for patterns (i.e. “Note 50 tends to be followed by note 56 frequently,”) before being able to compose an entirely new composition based on the characteristics of the source music.

In addition these impressive feats, the compositions of the program are musically indiscernible from the compositions of its source composers when played by human musicians. A four part chorale for voice composed by EMI in the style of Bach has all of the characteristics and nuances of Bach’s chorales, from harmonic structure to melodic development. What’s even more impressive was even able to use Mahler’s music to compose an entire opera in the style of Mahler. Mahler didn’t even write operas. EMI eventually developed into the program now known as Emily Howell. Emily Howell has all the capabilities and the entire database of EMI, its prototype, but it additionally has an interface that allows David Cope to musically and linguistically give Emily positive and negative feedback on its compositions, essentially making Cope the teacher and Emily the student, except Cope is able to take all of the credit for the work. Sound familiar?

In order to further understand EMI/Emily Howell as both a program and a composer, it helps to understand music from an algorithmic perspective through a brief explanation of algorithmic composition. Since the beginning of written music and even perhaps before then, music composition has historically had different sets of rules to be followed. At the most basic level, a note may sound better when played with a specific note as opposed to another, and one sequence of notes may sound better than another sequence of notes. Because of this, a formal set of rules was eventually developed. The success and genius of prolific composer Johann Sebastian Bach posthumously spawned the foundation of tonality for almost all of Western Music. This algorithm dictated rules such as which chord progressions to use (harmonic function) and how the notes within those chords move to the next note. This “Tonality Algorithm” is in fact so established that it is a common exercise in music theory to harmonize a melody in the style of Bach. Moreover, this algorithmic perspective can be applied to musical improvisation as well as composition. In Chord-Scale Theory, specific scales are to be used in correspondence with specific chords (i.e. the notes in a major scale are conventionally played over a major chord).

It takes Emily/EMI mere moments of analysis to develop algorithms for every composer that it studies, while it took humans hundreds of years to analyze Bach and develop an algorithm for the foundation of tonal harmony. Because of this, EMI/Emily can develop its own musical voice by using the influence of multiple composers to in a sense, recombine the musical DNA of the composers until finding its own unique composition style subject to David Cope’s approval and modification.

The invention of Emily/EMI raises questions pertaining to the authenticity and legitimacy of music. If EMI composes pieces that pass Turing Tests such that a listener cannot distinguish whether or not the piece was composed by Stravinsky or EMI, does that challenge the notion that music requires a human element to be viewed with merit and virtuosity? Does Cope still retain the title of composer when he teaches EMI to compose a symphony? Is Cope a composer, a programmer, or both?

“In twenty years of working in artificial intelligence, I have run across nothing more thought-provoking than David Cope’s Experiments in Musical Intelligence. What is the essence of musical style, indeed of music itself? Can great new music emerge from the extraction and recombination of patterns in earlier music? Are the deepest of human emotions triggerable by computer patterns of notes?

“Despite the fact that Cope’s vision of human creativity is radically different from my own, I admire enormously what he has achieved. Indeed, this lovingly written book about a deeply held vision of musical creativity should, I think, earn its place as one of the most significant adventures of the late twentieth century.”

–Douglas Hofstadter, author of “Godel, Escher, Bach” and “Fluid Concepts and Creative Analogies”

A friend of mine told me that the human essence in music is absolutely essential to a composition’s legitimacy. However, when that “human essence” is able to be broken down into numbers and modified into an art that is completely new and unique in its own way, does that not undermine the “human essence” that is so precious to us?

As a musician, this invention and breakthrough in computer music is absolutely terrifying to me. If anything I do as a musician or composer can be done exponentially faster by a computer and be just as much of a “Paolo Dumancas Composition” as my actual compositions, could my passion for music, such a significant part of my identity, be inevitably undermined and rendered obsolete?

-Paolo Dumancas

Sources:

http://artsites.ucsc.edu/faculty/cope/

http://www.radiolab.org/story/91515-musical-dna/

Cope, David “Computer Models of Musical Creativity” (2005), MIT Press.

Advertisements

Tagged: , , , , , , , , ,

§ 5 Responses to Something Old to Something New: From Bach to AI

  • bkcallander says:

    I think you could expand upon the moment when you say that “human essence is essential to music”. Because I believe this is the critical part. Not in the wishy-washy humans add more soul to their music or computers can never conceive true inspiration sort of way, but in that humans are the ones who enjoy music. There is much more to the culture of music than the simple conception of songs and new styles. Throughout history, music often becomes synonymous with culture and cultural movements. This is the part that algorithms will not be a part of. I guess the question raised is not what is music, but what does music represent?

    Like

  • Paolo, your post reminds me of an article I just read about a similar program that generates the actual music, while a human arranges the songs and writes the lyrics.

    Here’s the 2016 article from Quartz describing the program, FlowMachines, from Sony Computer Science Laboratories in Paris: https://qz.com/790523/daddys-car-the-first-song-ever-written-by-artificial-intelligence-is-actually-pretty-good/

    Instead of writing music based on the style of the composer, FlowMachines produces songs based on genres of music, like “pop” or “jazz.” Sony is planning to release a full album later this year.

    No one is claiming that this music is at all “good” or emotionally evocative, not because it’s missing some kind of “human essence” to which your friend adheres, but because AI-produced music is all experimental at this point. Perhaps in the future we’ll see music that is actually indistinguishable from “pure” human-produced music — perhaps even some Billboard number-one singles.

    Such a development wouldn’t threaten me, as someone who is not a musician, but I can guarantee you that the market for “authentic human” music will always persist. People are nostalgic, especially when it comes to music.

    Like

  • scifivandyguy says:

    Hey Paolo,

    I thought this was some solid work here, and I really found your discussion of EMI and digital music writing technology. I have previously engaged with softwares like SmartMusic that have incredible abilities — including providing feedback on whether or not the human musician accurately met required pitches, tempos, etc. The technology you describe, however, transcends the capabilities of SmartMusic clearly, and the compositional power of most people without significant musical background. I echo your sentiment of anxiety surrounding this technology. Perhaps to console a little bit, I think that ultimately — there is a lot more soul and fire that goes into a musical composition aside from just the pattern of it… Listen to the music of Phillip Glass or Schoenberg. I think that technology like this will become increasingly useful perhaps to save humans the effort of creating musical teaching manuals, primers, etc that contain basic sheets for learning, but I firmly believe that we do not have to worry about this tech usurping the need for human musicians. For one, I feel that part of the enjoyment of visiting the symphony or listening to a piece of classical music (or any music) is knowing that you are engaging with the feelings and lived experience of another person in that moment. It wouldn’t be the same, for me at least, to listen to the music designed by a program like EMI or Emily Howell.

    Like

  • I’m not sure if replicating Bach means that we never needed Bach to begin with. Once someone has an established body of work, it may not be hard to write soundalikes, but the body of work had to be created in the first place. And creating that work didn’t happen in isolation. Beyond the notes on the page, composition is inherently an art form defined by interaction: interaction between the notes and the musicians, between the musicians and the audience, and between the score and the social context in which it is written. While a computer program may be able to replicate part of this process, it seems unlikely that these complicated social interactions could be performed by a computer in the near future.

    Like

  • zachgospe says:

    Your discussion about artificially created music closely parallels what we have already discussed in terms of literary work. As both a writer and a songwriter, I originally harbored the same feeling of anxiety and opposition to A.I. as you express in this post; it is certainly a warranted response to a technology that could reduce the social importance of your individual talent. It threatens our conceptions of art as a distinctly human, creative endeavor whereby an individual, through innovation and practice, can make a lasting mark. However, after years of grappling with this feeling and researching more into A.I., I believe that musicianship will still be a revered and valuable quality far into the future.

    While theoretical A.I. are capable of producing a massive quantity of high quality work in the fraction of a second, they are limited by the same vetting process that human composers must face. It doesn’t matter how much a composer writes, or how innovative their work is; their work must first be heard to be of value. As A.I. will produce in the course of a day more work than any human could listen to in a lifetime, I can imagine there would always need to be a liaison between a human audience and an artificial intelligence in order to mediate what pieces are worth the time of publication and performance. This role would undoubtedly fall to composers or music critics who are accomplished in their own right, and they would likely be given the liberty to alter any A.I. produced compositions, placing their name as at least a co-author. Further, because these roles will no doubt exist alongside the human musicians and conductors who will perform pieces, the human mediators will likely have a bias towards human creations, allowing an greater percentage of human compositions to be recorded and performed.

    Already, the market for art is oversaturated. Anyone can create and publish their work on the internet, and as such, it is increasingly hard to distinguish yourself as an artist from the rest. This will continue to be an uphill battle, especially once A.I. produces in vaster and vaster quantities. A.I. will be able to produce better pieces with greater efficiency, but there has always been competition in art. Human artists will not fall out of value, but rather be faced with a great challenge, and one that will likely lead to incredibly unique and powerful pieces. Photography was once thought to be the end to painting. After all, it creates perfect “paintings” in a fraction of the time. Yet, painting still remains and a thriving and innovative industry. I would argue that a similar process will occur in the musical world once A.I. becomes more proficient. And, if all else fails, we will continue to consume art as an audience, as hobbyists, and as masters within our own localities.

    Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

What’s this?

You are currently reading Something Old to Something New: From Bach to AI at Science/Fiction.

meta

%d bloggers like this: