Generic Tokens

Trying something new; structuring twitter dumps and rants into my phone into writing with Claude

I had to watch this Jack Conte talk after being baited by his entering into the AI data dispute. I find him a fascinating figure. He has that YouTuber preacher affectation where it sounds like he is crying and excited simultaneously.

In fairness, he attempts to qualify his interest in AI by pointing out that early rote usage of film cameras, recording studios and synths was initially protested before great artists found new ways to use the new medium. He then makes the argument that if music models were solely trained on music from before 1960, they would likely not be able to come up with the decade-defining records that followed.

That is inarguably true, but it discounts his prior point that it’s humans who use models — and of course that post-1960 new technology altered the sonic landscape. If musicians had all the tools of the 20th century prior to 1960, then humans using a model trained on that era could indeed come up with those sounds. Sloppy crowd-pleaser reasoning, not dissimilar to his claim about fair use claims for pre-training data being “bogus” due to AI companies striking partnerships with Disney.

I can’t escape how much a product of his time Conte is. Pomplamoose got big on YouTube making theatrical covers of popular songs from prior eras — themselves a kind of token of the ideology of YouTube, early adopters emblematic of the idea that anyone could make something popular from their bedroom, in part through clever SEO. Through no fault of his own, Patreon to my knowledge hasn’t facilitated any era-defining art in the idioms discussed. Absent early adoption from characters like Amanda Palmer — a token of the ideology of Patreon, emblematic of the idea that anyone could fundraise from their bedroom — it has been a peripheral force in music.

So in a way the excited/crying tone reads more accurately to me as the death of a particular story of the era that made Jack Conte. The era-defining records of the 20th century were well compensated, and likely still will be — through playing in coffee shops for eternity and their catalogue owners brokering equity deals and partnerships with big AI. It’s everyone else who was sold a shaky premise of sustainable noodling that is imperilled by this new tech, and that betrayal has as much to do with everything that made Jack Conte as it does with new AI models.

That is the more painful but more insightful story: so many people were let down by the promises of that era that never panned out, and AI is exposing their fragility. To his credit, I agree with him that the parasocial dynamics that lead people to pay to support friends in making whatever they are making will not disappear. But to invoke era-defining experimental art is to invite questions of how little of it was delivered as promised by the excited crying avatars of the attention economy.

In a way, the fawning enthusiastic theatre kid mashups — YouTubers beating the algorithm by putting “Daft Punk” and “Michael Jackson” in their ukulele cover video titles — was perhaps the era-defining art of the past two decades. It is damning that now the AI data question asks us to find a price for the cultural value of that work beyond just being generic tokens to train on. The unfashionable answer is close to zero.