Rock & roll changed everything. Not least, the music industry.
Before the mid‑’50s, sheet music outsold records. But no‑one needed a printed score to learn a 12‑bar blues, and no score could capture what made Chuck Berry recordings compelling. Record sales overtook those of printed music, and musicians learned songs by ear, not by reading.
Then, in the mid‑’80s, MIDI took music in another new direction. Sequencing lent itself to repetition and precision rather than more traditional forms of musical expression, while sampled instruments captured one‑dimensional snapshots of the real thing.
Which brings us to today’s technological revolution. Sound On Sound and Sonarworks have carried out a major survey of musicians’ attitudes to artificial intelligence. We’ve taken a deep dive into the results in this issue (read it here), and they make for interesting reading. Unsurprisingly, AI is viewed as a threat to creativity; but there’s also a perception that this threat is genre‑specific. It’s widely felt that styles such as jazz and blues, which resist easy MIDI‑fication, will also be least susceptible to AI.
I’m not so sure. To my ears, the ways in which generative AI currently falls short of the mark mostly relate to audio quality. I don’t hear it as robotic or mechanical, or lacking expressivity. And I think that’s because AI is not reductive in the same way that sheet music or MIDI or sampling are.
Like rock & roll fans of the ’50s, AI has been trained on music in all its richness, not a dehydrated version of it. Instead of condensing music to its essentials, AI analyses the recordings themselves, identifying patterns on a very deep level. Train it on genres where subtle tempo variations are crucial, and it’ll faithfully replicate that. If a singer has a particular inflection at the end of a long vowel, AI will notice. Consequently, it turns out that AI is actually very good at, for example, playing dirty electric blues. (If you don’t believe me, spend a few minutes with AI music channels on YouTube.)
It’s a truism that AI is limited by its training material, and thus can’t be original.
At the same time, it also feels as though AI currently isn’t doing anything more than rehashing existing musical styles. It’s a truism that AI is limited by its training material, and thus can’t be original because it can only draw on what it already ‘knows’. However, when sequencing became mainstream, human musicians turned its limitations into virtues. Precision and repetition became the hallmarks of new genres. Does AI music not have similar attributes that humans can abuse in imaginative ways? Are there no glitches in the matrix that can be chiselled open?
If it can’t do more than pastiche rock & roll, maybe AI is not such a threat to creativity after all.
Sam Inglis Editor In Chief