Last summer I had the pleasure of speaking with Wade Roush about my Canons album and my path to creating it. Wade is the creator of Soonish, a podcast about “science, culture, curiosity, and the future.”
My story became a part of the latest episode of Soonish – released, July 27, 2018 – which explores the topic of Making Music With Machines. Other voices in the episode include the leader of Google’s project Magenta, which applies machine learning to the creation of new music and visual art; a DJ and EDM label manager; and a team of commercial composers. (See also my full interview with Wade.)
Among the musicians interviewed, I offered perhaps the most conservative perspective on the question of how technology might fit into a musician’s creative process. Of course, I use technology all the time in my musical work; I’m no reactionary. I use notation software to record my ideas and I use software playback in conjunction with high-quality sample libraries (or “virtual instruments”) to hear how it sounds. In my recent work with electric guitar, I’ve been using the latest amplifier profiling tools that allow me to replicate the sounds of beautiful vintage amps without actually owning them. Some of the company and product names that fall into my workflow as of 2018 are Finale, FL Studio, Reaper, UAD, Waves, Kemper, Kontakt, Soniccouture, Spitfire, and so on.
What’s conservative in my standpoint is that I see technology as a tool for realizing my musical intentions rather than as source of material. My music so far has not been about technology or inspired by it, and I haven’t looked to software to help me generate ideas, or to fill in gaps in my work. I’ve used software as my notepad, and sometimes as my orchestra-for-hire, but not as my muse or creative collaborator.
So what do I think about about the prospect that software could someway write new music that people would actually want to hear? What do I think about the possibility that algorithms could generate raw material for a composer like me to experiment with, suggesting melodies or rhythmic patterns based on an analysis of the composer’s previous output or stated preferences? What would I think about software searching one of my own scores for simple musical mistakes? If I’d be OK with the musical equivalent of a “spellcheck,” what would I think about an automated musical grammar or style check? And what would I think about software taking a score-in-progress and generating a possible conclusion for me?
In the Soonish podcast, I probably stand out as the musician least likely to embrace these possibilities. As I said in conversation with Wade, I don’t consider music as a scarce resource that we need software to supply us with. I don’t even see great music as a particularly scarce. In my own life, whenever I’ve felt bored with my daily listening, it’s never taken me long to find something new and wonderful – already written and recorded by humans – to surprise my ear. There’s a problem of too-muchness in the world of music: it’s daunting to contemplate how much great music is “out there” that I will never have the time to focus on and truly connect with.
Likewise, I don’t see musical ideas as a scarce resource. My challenge as a composer hasn’t really been coming up with ideas but with narrowing them down and then manifesting the discipline to bring them to fruition. It’s also true that musicians aren’t really a scarce resource; there’s a surplus of passionate and skilled human musicians in the world looking for opportunities to create and be heard. For these reasons, I contend that we don’t need machines to write music for us; there is no great shortage or crisis in the world of musical creativity that necessitates machine intervention.
Still, we might want machines to write music for us, or to write it together with us, because something interesting might come of the experiment. Every time we attempt to automate a process, we learn a lot about how the process really works, and that knowledge is valuable even if the attempt at automation is unsuccessful. But maybe it will succeed. Looking at the history of AI, we have to observe that lots things were bad until they were good. We have to take seriously the possibility that software-generated compositions might someday be ridiculously good even if the state-of-the-art doesn’t wow us right now.
Although there’s no shortage of music, there’s also no apparent limit on how much music the world can accommodate. If machines could someday write music that would delight us, why would we deny ourselves the pleasure of hearing and perhaps learning from it? Although many composers do not experience a scarcity of ideas, software might still be able to suggest options that a composer would never otherwise conceive, and if the composer’s goal is to make good music, why would the composer shut out a valuable machine-generated suggestion, insisting that every idea in a score be self-originated?
If it’s inevitable that scientists and hackers are going to keep working on algorithmic creativity and that someday the music-writing machines are going to be better than we can imagine right now, maybe we should make peace with the idea and begin taking advantage of the possibilities this field opens up?
So where do I stand on all of this? I’m still thinking about it so this post is to be continued…