Reading Response 1: MusicLM



The use of MusicLM and AI for music production give a lot of food for thought, some auspicious, some worrying. It is a forage into a realm yet to be seen, and hence, a lot of caution, intentionality, and critical thinking would be required to harness the full benefits of the intersection between AI and music while not compromising human creativity in the ways it has been exercised traditionally.

My immediate reaction to the OpenAI jukebox as well as the work by Agostinelli et al. was that of fascination. How do works by Matisse get translated into music? Is it a bit like the Schubert Wanderer Fanstasie’s album cover always being the famous painting of a wanderer staring into the sea of fog? The pure fact that machine learning ventured into this unknown territory filled my heart with a certain sense of adventure. I felt like the boy protagonist of sci-fi movies that just stepped into the portal that takes him to 50 years into the future.

But also, I was worried. The examples given were so different from the way I would have imagined things. The famous Munch Scream painting was very different from the way that I had imagined things. I often make certain associations between music and paintings, prose, and even colours. For example, Rachmaninoff’s Second Sonata is always a dark navy. Ravel’s Alborado Del Gracioso is always Seurat’s neo-impressionist painting of a circus for me. I draw parallels between Chopin and a South Korean poet Yoon Dong-ju, who wrote about his yearning for homeland in face of colonialism with the most exquisite and beautiful language. Would I feel less secure about these associations when artificial intelligence tells me what Yoon Dong-ju’s paintings sound like according to data?

Is this just something in development, a shortfall of which extensive datasets and more training would solve? Right now, it only was able to train on 5.5K pieces of data linking text to music. Like Chat GPT that effortlessly spits out tragic sonnets about astrophysics, stories about a lonely sorcerer and a fairy with translucent wings with perfect rhymes, will it produce songs that make us laugh and cry in a few years? Would I, if I learn a song that made me cry was produced by OpenAI jukebox with the command ‘Write a lugubrious song with strings and piano’, feel betrayed, ashamed, or hopeless?

I recently spoke to a good friend of mine about his friend who studies neuroscience. Apparently, there are scientists who believe that our qualia (subjective experience), our worldview, and our emotions are all nothing more than chemical reactions. We talked for hours about whether that reduces the value of our life and our world into nothing more than the perceived beauty of our stories. To me, music is a unique element of our life that touches upon what is very personal and emotional. I am still unsure if I feel that the machine's venture into this territory is an unwanted encroachment.