Music Created by Learning Computer Getting Better

367 14 Loading
David Cope has created Emily Howell, an artificial composer that learns from criticism.

David Cope has created Emily Howell, an artificial composer that learns from criticism.

Like many composers, Emily Howell has her own unique style that she refines constantly. She’ll listen to what critics have to say about her work and then try to write music that is good enough to silence those critiques. Aiming for musical perfection is a very human endeavor, but Emily Howell isn’t a human composer at all. She is a computer program designed by composer, programmer, and author David Cope. A retired professor of music at the University of California Santa Cruz, Cope has spent almost 30 years at the forefront of synthetically composed music. With Emily Howell, he hopes to prove that great musical composition is no longer limited to mankind. Her CD from Centaur Records is set to drop soon.

Cope’s original effort to get a computer to generate new music was called Experiments in Musical Intelligence, or EMI (pronounced ‘emmy’). That program was able to analyze the work of other composers and produce a completely new piece in their style. You can download MP3s of the EMI work here, there are compositions in the style of Beethoven, Joplin, Cope himself, and many others. Emily Howell, in some ways the offspring of EMI, analyzes her own work just like EMI would analyze famous composers. The result is a feedback loop that, when coupled with critiques from Cope, has created a musical learning machine. Self-analysis is a key ingredient in artificial intelligence, and Emily Howell may be one of the precursors to self-aware computer programs.

Who wants to play synthetic music?

Back in 1980, Cope was having some writer’s block trying to compose an opera. He got the idea for a computer program that would write music for him. The first compositions were completely awful, and had to be transcribed from numbers into notes. That’s when Cope wondered if instead of trying to write rules for the program to create music, he couldn’t just have the program look at his compositions and determine the important rules on its own. From that concept, EMI was slowly born, working to first analyze Cope’s music and later to create works based on that analysis. After getting EMI to work for his own music, Cope found that it could analyze other composers just as well. Soon EMI was channeling the genius of Bach, Chopin, Beethoven and many others. It has created thousands of new compositions that are stylistically indistinguishable from the great masters of music.

With EMI, Cope ran into prejudice from the established musical community. Performing a newly discovered Bach score would be wonderful, performing a new Bach piece that was synthetically created would be blasphemy. Cope tried, but never was able to get high-profile musicians to play his pieces publicly.

Which is a shame. Listening to the MP3s (here’s the link again) I’m pretty impressed with the work. Now, full disclosure, I have about as much musical talent as a metal pipe falling down the stairs. Still, I like what I hear and I can identify the styles as belonging to the composers listed. I’ve been told that the EMI compositions sound ‘lifeless’ but that may be because they aren’t being performed by living musicians.

In 2003, David Cope decided to take a new tack. Instead of generating new pieces of old composers, he was going to create an entirely new composer. Hence Emily Howell was born. Her first task was to analyze a database of original EMI pieces and use that to launch her into writing new compositions. Unlike many human composers, Emily’s great at taking critical feedback. The program can receive audio or written critiques from an audience and use them to generate the next composition. Details of how Emily Howell works are provided in Cope’s book Computer Models of Musical Creativity.

Chances are that Emily Howell will not be nearly as popular as EMI. While EMI could seemingly bring old composers back from the Dead, Emily is just one more musician in a sea of hopeful composers. But that, according to Cope, is exactly what makes her so special. She’s an actual competitor to human composers, generating new work in a style all of her own. While many others may be pursuing artificial intelligence, Emily Howell is proof of artificial creativity.

Not being a musician, I have no idea what the long term impacts of EMI and Emily Howell will be on the music community. It’s unclear if AI could ever replace human composers, but if it could, we’re likely to see opposition to the idea just as we see some opposition to the increased use of robotics in manufacturing.

I do know that the concept of artificial creativity is going to be huge, especially when it evolves into creative problem solving. Think of the benefits that could arise from a computer program that could not only perform a task, but could think of new ways to perform that task and find the best of its own ideas. Creative learning machines could be the definitive technology of the 21st century.

[photo credit: David Cope]

Discussion — 14 Responses

  • gideon October 9, 2009 on 9:02 pm

    When I listen to music I like sharing the experience that the song writer felt when they created the piece, thats what a song writer does, they take some feeling in their gut and in their heart and translate it into sound. Which should be impossible if you think about it but it can be done and because of that the experience is magical. Its almost mystical when done by a master. That is why the concept of machine generated music is so disgusting to performers because by removing a person from the conception of a song it also removes the humanity from it, the shared emotion. When you hear something beautiful you think ‘wow what was going on in the composer/songwriter’s mind while they were creating it?’. Its something that exists for the very reason that we relate to it. A machine generated song doesn’t know how you feel, it doesn’t know your life’s experiences, just how to shift numbers around until it stops receiving negative feedback from its programmer. Imagine if your parents or someone you loved sent all their correspondence (phone calls, email) to you via a computer generated system without any direct input, just a machine generated message. At the end the machine says ‘I love you’. Would you really feel loved? By your loved one for not even speaking to you directly in their own words, who instead relies on a machine to transfer some automated message? I doubt it. How about the machine, would you you feel loved by that? I doubt that too. That is exactly why computer generated songs disgust people who create or perform music. Its not about being replaced (that happens every generation to every musician), its literally about taking the soul out of music, and that soul is the very reason for music’s existence. A song is a conversation from a persons heart, and a machine that generates songs from someone else’s style is like you taking a conversation from someone else, changing some of the details, rewording it, and then pretending you came up with a thought even though you had nothing to say. You could say that people do that too with music themselves, but you’d miss the fact that those people aren’t the ones who are good at songwriting. I’d add that a sentient program that really thinks on the level of a person might be able to move people but it would be because it could reason like us, not because it could imitate us.

    • Fernando Torre gideon February 21, 2013 on 5:28 am

      Call me what you will, but I can already see a few places where this technology would be received with open arms:
      1. places where the music playing doesn’t matter. I.E: places where it’s basically background noise (like those video games that need constant background music, all in a certain style, but most of the time without any specific meaning attached)
      2. struggling composers E.G.: How would I start this next movement if I weren’t so tired? Well, I can either go through all my pieces looking at what I’ve done before, by which time I’ll surely be way more tired, or I could let a computer do it, and then modify it’s output (like artists who use photoshop to create “noise” textures and later fine-tune them)
      3. to produce scratch music

      As far as eliciting emotion, the computer is trying to see what notes are associated to what emotion, in much the same way a human would: trial and error.

  • gideon October 9, 2009 on 5:02 pm

    When I listen to music I like sharing the experience that the song writer felt when they created the piece, thats what a song writer does, they take some feeling in their gut and in their heart and translate it into sound. Which should be impossible if you think about it but it can be done and because of that the experience is magical. Its almost mystical when done by a master. That is why the concept of machine generated music is so disgusting to performers because by removing a person from the conception of a song it also removes the humanity from it, the shared emotion. When you hear something beautiful you think ‘wow what was going on in the composer/songwriter’s mind while they were creating it?’. Its something that exists for the very reason that we relate to it. A machine generated song doesn’t know how you feel, it doesn’t know your life’s experiences, just how to shift numbers around until it stops receiving negative feedback from its programmer. Imagine if your parents or someone you loved sent all their correspondence (phone calls, email) to you via a computer generated system without any direct input, just a machine generated message. At the end the machine says ‘I love you’. Would you really feel loved? By your loved one for not even speaking to you directly in their own words, who instead relies on a machine to transfer some automated message? I doubt it. How about the machine, would you you feel loved by that? I doubt that too. That is exactly why computer generated songs disgust people who create or perform music. Its not about being replaced (that happens every generation to every musician), its literally about taking the soul out of music, and that soul is the very reason for music’s existence. A song is a conversation from a persons heart, and a machine that generates songs from someone else’s style is like you taking a conversation from someone else, changing some of the details, rewording it, and then pretending you came up with a thought even though you had nothing to say. You could say that people do that too with music themselves, but you’d miss the fact that those people aren’t the ones who are good at songwriting. I’d add that a sentient program that really thinks on the level of a person might be able to move people but it would be because it could reason like us, not because it could imitate us.

  • larry October 10, 2009 on 5:32 pm

    ok so who owns the copyright

  • larry October 10, 2009 on 1:32 pm

    ok so who owns the copyright

  • Tom C October 12, 2009 on 5:32 am

    To gideon: I think you misunderstand how the human mind works. I’m a composer (not very good), and I’ve studied music for many years with composers and musicians who ARE very good. Universally, they have said that their first — and only — goal when writing their music is to make it “good.” Meaning that the idea that a song is driven by, say, the loss of a child, and every note is chosen to express that grief, is rubbish. Music is driven by rules.

    Now, true, we humans can map ideas and emotions onto music after the fact. In fact, that’s what movies do — masterfully — by placing a song in a minor key over video footage of something tragic, they “double” the sadness quotient. But that’s not in the music at all, and it’s very, VERY unlikely that ANYTHING like what’s going on in the movie is what was going on in the composer’s mind at the time of composition.

    This doesn’t tell us anything about music; rather, it tells us how adaptable the human mind is at assimilating new pieces of information… at finding patterns in the world around us (even when they aren’t there).

    I think that this story is amazing, and look forward to future creations.

  • Tom C October 12, 2009 on 1:32 am

    To gideon: I think you misunderstand how the human mind works. I’m a composer (not very good), and I’ve studied music for many years with composers and musicians who ARE very good. Universally, they have said that their first — and only — goal when writing their music is to make it “good.” Meaning that the idea that a song is driven by, say, the loss of a child, and every note is chosen to express that grief, is rubbish. Music is driven by rules.

    Now, true, we humans can map ideas and emotions onto music after the fact. In fact, that’s what movies do — masterfully — by placing a song in a minor key over video footage of something tragic, they “double” the sadness quotient. But that’s not in the music at all, and it’s very, VERY unlikely that ANYTHING like what’s going on in the movie is what was going on in the composer’s mind at the time of composition.

    This doesn’t tell us anything about music; rather, it tells us how adaptable the human mind is at assimilating new pieces of information… at finding patterns in the world around us (even when they aren’t there).

    I think that this story is amazing, and look forward to future creations.

  • Geoff W October 12, 2009 on 8:21 am

    I’m really excited for the advances in technology that have progressed us to this point. As a musician that struggles to come up with my own original pieces I can appreciate the difficulty of creating an original work of music that people, in general, would enjoy listening to. Just because a computer can do it better than me doesn’t threaten me at all. There are lots of things that computers do better than me. Recognizing and processing patterns is one of them, and really all that music is when broken down.

    No matter the source, it is music. It is art. Different people will analyze art and have different opinions, different implications, and be impressed upon by the art differently. The soul of the music is in the ear of the beholder and how they connect with the stimuli presented. The sound of a brook in a forest can be described as nature’s music. Calming, soothing, and derived from objects with no self awarenes and no person behind them. Emily Howell is not so different except that she/it can improve upon the existing output based on “critisism” instead of a river only able to follow the path of least resistance.

  • Geoff W October 12, 2009 on 4:21 am

    I’m really excited for the advances in technology that have progressed us to this point. As a musician that struggles to come up with my own original pieces I can appreciate the difficulty of creating an original work of music that people, in general, would enjoy listening to. Just because a computer can do it better than me doesn’t threaten me at all. There are lots of things that computers do better than me. Recognizing and processing patterns is one of them, and really all that music is when broken down.

    No matter the source, it is music. It is art. Different people will analyze art and have different opinions, different implications, and be impressed upon by the art differently. The soul of the music is in the ear of the beholder and how they connect with the stimuli presented. The sound of a brook in a forest can be described as nature’s music. Calming, soothing, and derived from objects with no self awarenes and no person behind them. Emily Howell is not so different except that she/it can improve upon the existing output based on “critisism” instead of a river only able to follow the path of least resistance.

  • Jax October 12, 2009 on 6:36 pm

    People reject it because they are scared, we are not as special as we think. I find it pretty awesome that we can create a neural network to do this kind of thing. The only thing it can’t do on its own is write a piece of music for a purpose (e.g. for a cause or such like).
    Certainly could go a long way in terms of replacing the mainstream though. :)

  • Jax October 12, 2009 on 2:36 pm

    People reject it because they are scared, we are not as special as we think. I find it pretty awesome that we can create a neural network to do this kind of thing. The only thing it can’t do on its own is write a piece of music for a purpose (e.g. for a cause or such like).
    Certainly could go a long way in terms of replacing the mainstream though. :)

  • ricemine October 21, 2009 on 3:11 pm

    i’m just a fan of the fact that it responds to criticism…isnt that what pop music is? not necessarily art though.

  • ricemine October 21, 2009 on 11:11 am

    i’m just a fan of the fact that it responds to criticism…isnt that what pop music is? not necessarily art though.

  • Anonmicron February 25, 2010 on 3:32 pm

    I wonder what would happen if you fed Emmy some Ozzy