A long-winded response to the death of music

Again, not an article, rather a too-long-winded response to Jordi Linares comment on Sven Welt’s article:
https://medium.com/counterarts/ai-will-kill-music-but-the-real-question-is-do-we-care-f04eef48caee

I agree with many of the ideas expressed here (leaving aside the nihilism), except for the proposition that “AI will far surpass us”. Has human music ever proceeded by surpassing itself? Does new music surpass old music? Is Beyonce intrinsically better than Ellington, Mahler, or Bach? No, because music simply doesn’t relate to other music in that way. It never has and I don’t believe it ever will. I mean, you could use something obscene and irrelevant like album sales or streams, but no serious individual would do that, imho.

Concretely speaking, my general position is that all music is (still) human music, since AI is just sampling from a probability distribution defined by human music. Yes, it can interpolate in latent space to find new blends of human styles, but I’m not at all convinced that it’s capable of deeply transformational creativity, the way a human creator is. It simply doesn’t experience the blends of fallibility and skewed excellence that lead to many/most creative discoveries. “Temperature” is in no way similar to having a shitty right pinky finger, or being distracted by a honking garbage truck while trying to recall the start of a melody you’ve been working on. “Beam search” is in no way similar to sifting through fragments of misremembered melodies that are naggingly close to what you want, but nevertheless a mile away. One thing I can say with great confidence is that without being explicitly fine-tuned on The Shaggs, the likelihood of AI ever arriving at such an exquisite balance of imperfections is so close to zero that it is vanishingly unlikely to ever by sampled. Yet a fond look back at The Shaggs by artists in the 90s opened the flood gates for a whole wave of new, human musical creativity. These new artists could never reproduce The Shaggs, but they were sufficiently inspired by this phenomenally and gloriously “wrong” music to make something of their own, and to fail at being The Shaggs in their own special ways. The same could be said of Le Sacre du Printemps.

AI also currently has no critical faculty. Yes, this is likely to be “solved”, in some sense, but on what terms? What terms does a fallible human creator use to critically evaluate something they’ve done? And is it perhaps their inability (or refusal) to ever really “fix” those problems that leads them to their unique musical voices? Personally, as a composer and musician of 30+ years and an AI researcher (in music) of around 15 years, I think this is precisely the case. Human music (aka “music”) isn’t the way it is stricly because of the neural network of our brains, but because of our entire physically, temporally, and culturally embedded being.

Okay, so perhaps you give AI an “intrinsic motivation” (a la Schmidhuber) to create music, complete with the ability to define its own critical aesthetics. What then? In my opinion, it’s highly likely the AI will rapidly move into “musical” terrains that are simply unrecognizable to humans as music. After all, why would a system operating at far greater speeds, with immensely more “working memory”, and no physical limitations on beat induction or perceived pitch range (or even pitch perception) limit itself to the extremely narrow range of perceptible and intelligible human musical pitch and tempo?

In fact, I think things are likely to “go” in the opposite direction. I personally think purely generative music is boring and will be recognized as such in time. It’s a validated fact of music psychology that, broadly speaking, you “like what you know”. People don’t want endless variety in music. Hitting a button and getting a perfect new Beatles song is as boring as playing a video game you can never lose. People want to engage with music; emotionally, psychologically, culturally, and socially. And sampling new Beatles songs, ad nauseam, isn’t going to provide that.