Guest essay
“The Tinkering Method: How Computer Learning Informs Music Pedagogy,” by Jeff Lundy
Music and statistics are among my most passionate intellectual dabblings. Recently, I struck upon the crazy idea of seeing what people have done at their intersection. To my luck, I discovered that the statistical analysis of music recently experienced a small resurgence among music analysts.
There are many interesting avenues pursued by this new research. In the interest of being brief, however, let me just say that this work is still in the “proof of concept” stage. From what I’ve gathered, it’s going to be a long time before a piece of software robustly dissects the style of a work, or before it creates a custom-tailored piece of music. But of course, academic research is rarely about finding practical, ready-to-use products. Just striving to get a computer to understand music is in itself quite informative.
For instance, consider the opinion of British computer scientist, Darrell Conklin. In a recent review piece he explains the best process for getting software to create stylistically-competent compositions. Essentially, his message is this: software does the best job of imitating a musical style when you give it a piece of music from which to start, and then let it modify it thoroughly. This is infinitely better than the rather brittle process of letting the computer try to create its own melodies from scratch. Why is modification better than creation? Because software that starts from scratch lacks a long-term vision of where it’s going, and it frequently can’t find a good “next note” to put in a melodic sequence. By giving the computer an existing piece of music from which to work, you give it a guide-wire to get past localized problems.
What does this have to tell us? Well, I believe there is a pedagogical insight to be gathered here. Computers are fairly dumb when it comes to understanding and creating music. Being dumb at understanding music, computers are not unlike novice musicians. Thus, what we learn about teaching computers to understand music may have something to tell us about teaching human beings to understand music.
In particular, the modification method outlined above supports an opinion I’ve held for a while; namely, that conventional music pedagogy suffers from a flaw. When presented with a novice composer, many teachers try to fill a student’s head with the “fundamentals” of Western music. Admirably, these teachers hope to get the student started with a complete foundation. However, I believe this approach falters because it misses the cognitive gap between analytically understanding a language’s grammar, and experientially understanding how to manipulate that grammar in a useful way. This misunderstanding is akin to giving a child her first bike, and then sending her off to ride it, with no further guidance than the assembly instructions. Hearing how the various components of the bike are assembled – “Place screw A into the pre-drilled hole labeled D, and turn with a Philips-head screwdriver until tightened” – is not going to help poor Sally learn to ride a bicycle. Similarly, novice composers have trouble grasping how fundamental musical elements translate into full-blown musical phrases.
In opposition to conventional teaching, I would argue that novice composers need to follow a process similar to the learning process of statistical software. For the sake of convenience, let me call the following approach the “tinkering method.”
At first, students following the tinkering method would listen to lots of music in a certain style, and even try to absorb as much music theory as they possibly can (this inductive approach is similar to how predictive software “learns” about new music). With their limited processing skills, however, novice composers would benefit from having a guide-wire, much like their computer counterparts. So instead of trying to expect complete originality from students at the beginning of their journey, it seems preferable to let students lean on the compositions of others. Perhaps the best approach is to instruct students to rearrange phrases in an extant work; or (following the process outlined by Conklin) to have students randomly select notes in an extant work, which they replace with alternate notes believed to be stylistically appropriate.
This tinkering method will give novice composers the experience of creating something new, right from the outset (increasing their engagement with the process). Moreover, it’s likely that after a while, students will begin to connect their small bits of superficial understanding into a larger whole. By tinkering around with already existing works, it should become apparent how various momentary choices shift the direction of a piece into various long-term directions.
Eventually, students following the tinkering method would advance to a little more freedom. Specifically, I envision them creating compositions with the understanding that they are permitted to steal whole phrases or devices from other composers (warning them of the dangers of plagiarism, of course). Basically, students would be advised that they are free to take as much from other composers as they wish, but that they are encouraged to use as little existing materials as possible. Given time, I believe that students would require less and less stealing from other composers; until eventually the students would no longer need to lean on an existing piece of music at all.
In conclusion, I would say a word to those concerned about using computers as a model of the human brain. I recognize this is a valid concern. Prominent cognitive philosophers have argued that the computer’s serial processing is a terrible analog of the brain’s parallel processing. Nonetheless, I believe in this instance I have drawn an appropriate conclusion from an example of computer learning. The idea for the tinkering method comes not from a direct analogy to the computer (i.e. students should listen to lots of MIDI and then draw correlation tables). Rather, it draws from the spirit of trying to teach a new trick to a poorly-equipped information processing device. In fact, if there is a fault in following the computer analogy too closely, I would say that it lies in the conventional pedagogical approach. It is this view of the student that seems over-enamored with the ideal of programming basic commands into a student’s head.
Jeff Lundy is a PhD candidate in Sociology at the University of California, San Diego. His dissertation is a statistical analysis of how and why American households overspend (i.e. spend more money than they make).
February 13, 2010, 9:58am
