music is what we want for our children – the fourth best view on ai

I find it so amazing when people tell me that electronic music has no soul. If there’s no soul in the music, it’s because nobody put it there.
- Björk

A few years ago I was talking to a friend who was talking about his career options as a music graduate. The path that he felt was most likely to be lucrative, but the one that he couldn't bring himself to take, was education. People are willing to pay the price for a musical education, he reasoned: music is what we want for our children.

That phrase has been knocking around in my brain for years, resurfacing as I try to make sense of an interview with Mikey Schulman, the founder of AI music tech firm Suno. In it, he argues "It's not really enjoyable to make music now... it takes a lot of time, it takes a lot of practice, you have to get really good at an instrument or really good at a piece of production software. I think the majority of people don't enjoy the majority of time they spend making music."

That quote was brought into my field of view by music educator Max Alper, with the simple remark "These people are your enemies."

The idea that artists who have invested the time and effort to develop skill and mastery in their craft will instead turn their attention to Suno's woefully imprecise prompt-and-click mediocrity machine reads as an insult to musicians. Several years into the cycle of endlessly being told "but the AI will only get better" as generative AI systems plateau and we are left with slop machines, companies like Suno imagine themselves as providing us with something we want. When I started writing this piece, it went something like this: There is an audience for Suno, and maybe it's people making silly memes and those who want to make a quick buck from the dysfunctional music economy.

Schulman is hardly alone in taking on the ire of my favourite commentators: last year Daniel Ek stepped into the same shit when he declared "With the cost of creating content being close to zero, people can share an incredible amount of content". I feel like calling music "content" is its own sign of choosing the side of the enemy, but this text is probably accurate for the type of platform which Spotify sees itself as. One AI music tech company, Boomy, claims to have allowed users to generate an amount in the tens of millions of "songs", a figure which they absolutely insanely claim to represent 14% of all recorded music. Cast your eyes back to our notes on the impossibility of getting your music heard by an audience. It's getting worse! If you're happy to call this slop "content", then yes. The cost of content is in fact close to zero. The value of it is equal.

I am not saying every single "AI" tool ever built is purely useless. In my day job as a software engineer, I frequently use AI tools to automate the more tedious aspects of the job. They help, but I very, very rarely see AI get real-world tasks right first time, and I believe that they probably aren't worth the money a lot of them cost. Software isn't art, it's a lot easier to work with and modify, and even then, LLMs are a most frustrating programming partner and does its best work scribbling in the margins and working as a kind of predictive text for repetitive work. This is an observation largely stolen from Ryan Broderick of Garbage Day, but if AI cleans up a producer's library of drum samples like XO does or make some fucked up synth sounds like SynPlant2, good! If Schulman made SynPlant2 he wouldn't have earned the right to spout bullshit about musicians practicing their craft being a problem to solve, but at least he would have actually have met musicians where they were.

In fact I even like a song with generative AI a little deeper in its fabric, but it's one I think I can make a consistent argument for. Last year, JPEGMAFIA sampled one of those memes powered by a Suno-like tool. Someone on YouTube calling themselves "AI For The Culture" made a cover(?) of sorts where robotic artifacted vocals try to sing Future as a 1973 soul jam, and Peggy uses that cover as the basis for either on or off the drugs. What's funny is that Peggy's version adds a lot more detail - filling in the skeleton of the fake song with significantly more organic detail; the trembling robot voice begins sounding a million miles out of place. The effect is bizarre, and bears basically none of the DNA of the original Future song. Yet I don't believe there's any real point to synthetic crate digging when the crates we already have access to are already so deep ; this is maybe a point in favour of YouTube to MP3 tools more than it is in favour of generative AI. In Peggy's discography, it feels closer to his history of digital native sampling and getting into the weeds of YouTube than any win for Suno.

"I'm living inside your head. You've gone mad."
– "Effort" from O Emperor's Choice Music Prize winning album Jason

Every year, the team behind Ireland's highest honour in popular music, the Choice Music Prize, releases a particularly fascinating artefact - a master list of every single record they're aware of that is deemed eligible for the award. I believe they scour the Irish press for every last mention of any record. They make mistakes, and they welcome corrections from the public - at my glance right now I noticed that Princ€ss' self-titled didn't make it, but they're damn near close to perfect. If you want to know how many albums Laurie Shaw finished in a year (seven in 2022 btw), that's where to look.

The 2021 list, for reasons that escape me, even contains the year's culchie surrealist output of Dollar Pickle Records Kerry, the infamous North Kerry Noise label, who's fingerprints on the list need little examination.

The North Kerry Noise movement's manifesto is worth a read in its own right, but here's what's relevant to the article:

1. You Must Have Your Song Titles Written Before Any Music Gets Written
2. You Can’t Spend More Than Half An Hour On Making A Track
3. All Tracks Must Be Created In One Go, You Can Not Save And Come Back To It At A Later Time

Spending anything longer than 30 minutes on a North Kerry Noise track is ludicrous. The whole point is to put the least amount of effort possible into creating music, thus contributing to how terrible the entire thing is. A lot of proper, credible musicians might not be comfortable with this aspect, but it is how it is

It shouldn't come as a surprise then that this fluke wasn't repeated, and these albums, valuable as they are to my enjoyment, didn't make the list again. After all, everything else on the list probably was the result of a sum total of endless hours of work. Regardless of what myself, or the judges think of them, their genre, their approach, music making is a process of immense and extreme effort, and I imagine someone looked at the deliberately lazy joke music and drew the line some day. Fair enough.

I'm not going to get through the 211 records that the Choice has identified in 2024, although I like checking in with the list anyway, to see if there's anything I meant to check out but slipped through the cracks in the year. I somehow missed the release of Dan Walsh's Fixity 8, and somehow missed all news about Súil Amháin & Bantum's athPhORT.

210 records I can reliably be sure were brought to this world with some serious amount of time and effort. They might not all be for me, but they're all for someone. And there's one I'm less sure of.

The Reality, by KAWAII HOE

10 track album

KAWAII HOEBandcamp New & Notable Feb 5, 2018

The main character of today's story is the Wexford-based Kawaii Hoe. Their album The Reality came out on their 27th birthday last year, according to press releases. I'll be abbreviating the project name a lot. It seems to be a part of a wider foray of theirs, which they've named "AI Music Label", and which has its own limited company called Epistle Ltd. I had offered them anonymity before setting out on the piece, but they has expressed a preference to be named. Ultimately, while I don't think their music is for me, there isn't the case to be made in the same way as against someone like Daniel Ek or Mikey Schulman.

I reached out to KH for this piece, and excerpts of our conversation will be included below. (They will appear out of order from the actual conversation and my own questions may be edited for clarity and length.) They opened by asking me what the Fourth Best angle on AI even is, and if I was coming into this with fair intentions for them and their work, or if I was approaching this without even giving the work the time of the day. I kind of stumbled into an answer for myself:

Fourth Best: I guess the angle I'm approaching this article from is that I kind of see it as something that's maybe unfortunate for the music industry at this current moment, with how competitive and difficult the streaming environment is for artists at all levels. I do think there are harms with AI music, but I think those harms come from the tech and music industries rather than from people like you who are interested in the technology, experimenting with it, maybe discovering something about yourself and your tastes through the process.

I think I'm still somewhere close to that, coming out of the experience. I think that I respect that KH can have fun with a weird piece of tech but I'm worried about AI in the long term and what happens when music made with it is brought to the marketplace.

I don't know who put their album on the Choice's master list, if it was them, a PR firm, or if it showed up in the research. See, KH has actually received a handful of spot radio plays in Ireland, and the kind of positive early coverage that's given in good will to artists who show decent technical promise but maybe haven't fully found their feet yet. RTÉ gave them a by-the-numbers interview for their website. They use a PR firm that is shared by some pretty good artists. Nialler9 gave them a feature - talking to me for this piece, Niall admitted feeling "duped" by the whole thing, but we'll get to that later.

exploring the world of Kawaii Hoe

I had to reach out to Kawaii Hoe to write this piece. There's a lot of text online about them; on the website for AI Music Label, on Bandcamp, in press releases, even in RTÉ, but I couldn't really put together a clear picture of what makes someone want to take on this task.

FB: I've helped artists put music out there, do the process of PR, trying to get media attention, playlisting etc. and it's been stressful, something that I think a lot of artists get burned out on! So I guess I'm curious what inspired you to take on that task.

Kawaii Hoe: I got inspired by Charli xcx to do this since she authorised YouTube Music AI to use her voice. I love her and how she says everyone could be an artist now. I’m glad that AI music exists because I can venture into a business I’ve really wanted to do, starting from scratch with my own finances. It’d be great if I could make money but after all, I’m satisfied, having fun, happy that there are people in the world especially in Ireland who would listen to me.

I’m glad that AI music exists because I can venture into a business I’ve really wanted to do, starting from scratch with my own finances. It’d be great if I could make money but after all, I’m satisfied, having fun, happy that there are people in the world especially in Ireland who would listen to me.

This is already pretty interesting - I hadn't heard of that collaboration before. I looked at what Charli had to say about it; to tell you the truth, I'm not sure if she wants to be there.

"When I was first approached by YouTube I was cautious and still am, AI is going to transform the world and the music industry in ways we do not yet fully understand. This experiment will offer a small insight into the creative opportunities that could be possible and I’m interested to see what comes out of it."

Perspectives from the artists collaborating on our early music experiments

Hear from the artists who have joined hands with YouTube and Google DeepMind on our newest experiment: Dream Track and Music AI.

YouTube Official BlogThe YouTube Music Team

I'm also a Charli fan, but I've been searching after this interview, and I can't for the life of me figure out what they were referring to when he said Charli said that "everyone could be an artist now". In the wake of BRAT, Charli xcx's worldview seems to be something of the opposite. On an episode of Subway Takes, she noted "music is not important" and opined on musicians who don't take time to craft a greater world - an artistry - around their work. In other words, artists are a rarity, not the everyman. Maybe this worldview could be read in reverse, however - if music is separate from artistry in Charli's eyes, does the source of the music even matter?

An off-the-shelf AI music detector identifies Kawaii Hoe's work as AI with "100% confidence". The way in which some AI music detectors work is maybe ever-so-slightly revealing in telling the story of how AI music is usually generated. When a piece of music is compressed to MP3, is picks up usually-imperceptible artifacts. AI systems trained on libraries of probably-stolen MP3s include mathematically impossible imitations of these artifacts in their outputs. A similar story repeats itself in image generators mimicking JPEG compression. There's a good YouTube video about it by Benn Jordan.

FB: What sort of tools you use to make music - do you use fully generative tools like Suno or Udio, do you use a mix of those tools with more traditional music production tools like DAWs?

KH: I use Suno and you have options to download more than the direct output if you subscribe. What I do is that I edit it to the way I want it to sounds like.

For the music style, I make sure each song is what I wanted and cohesive. Sometimes I have to start over again and again because AI makes mistakes too. For the lyrics, it is written based on my own true life experiences and edited by myself.

To me, the difference is that you use traditional or digital instruments and I use AI, and we are both making music - does it matter how music makers create the music if the music is good? For me, AI is a tool like traditional instruments (no disrespect). AI could be nice and helpful for some creatives who don’t know how to compose a song but only know how to write lyrics.

All the website, artworks, videos, distribution, metadata, are created, designed, handled by myself without AI.

FB: Do you feel like the music you've made so far is a meaningful expression of your own style, taste and ideas?

KH: Absolutely yes. The way I see AI is a tool kind of like an instrument. I use it to bring my ideas to life. For example, all my lyrics are based on personal experiences and feelings, and I make sure the final output matches the vision I have in mind. If the AI generates an idea that doesn’t feel right, I’ll tweak it or scrap it and try again. So whilst AI plays a big role, the creative direction is always mine. I think it’s important to clear up the misconception that AI means unoriginal or fake, a lot of thought and effort still goes into every song I make.

Kawaii Hoe's music is rather... uncanny. I don't really know how far we can take this conversation without acknowledging the fact that I don't particularly care for it; there's some strong pop bonafides in there, but at times it gets very weird and off-kilter. It uses a breathy female vocal that sounds quite consistent across their discography, and the backing tracks feel like Deadmau5-style soft progressive house tracks, the sound of the internet in the early days of always-on broadband. If that's what they're going for – it's pretty spot on.

Colours everywhere, by KAWAII HOE

track by KAWAII HOE

KAWAII HOEBandcamp New & Notable Feb 5, 2018

I've tried messing around with Suno mostly for the purposes of making joke songs and to get a feel for if we're all fucking cooked. Everything I touch coming out of it sounds a little bit too much like Imagine Dragons, no matter how hard I prompt it to go a certain direction. Ultimately, I didn't want to pay for it to stick around at it long enough to get good results. Mainly, I felt frustrated about how little control I had over the outputs.

To borrow yet again from Ryan Broderick's Garbage Day newsletter, he described the experience as yet another slot machine for our broken internet:

Last month, I started poking around AI music generators like Suno and Udio. Both had more in common with Spotify or Apple Music than they did ChatGPT or Midjourney. There are subreddits for these apps where users share what they’re making. They treat AI music like a gacha game, paying for credits, hoping to generate the song they hear in their head, and then agonize on Reddit about it when it falls short. [...]

They want us to to pay to subscribe to their background noise app and they want us to populate it for them, now that they’ve made, as Spotify’s CEO put it, “the cost of creating content close to zero.” And then they’ll want us to pay more to ask the AI slot machine to generate something we actually want.

[source]

I find it interesting that this familiar repetition, the lack of control and making progress by trial and error is there in KH's explanation of their practice, how it feels like it's them getting better at expressing their ideas. I don't know for sure if I personally believe that. Here we are, stochastic outputs drawing from weighted virtual neurons. Put your signature on it. It's yours.

FB: I guess I'd like to hone in on what you said about using AI to bridge a skill gap; have you tried making music through more traditional means in the past, given that you said you'd always wanted to try it out?

KH: I’ve always wanted to try my hand at music but traditional methods felt a bit out of reach for me. I grew up singing in a choir so music has always been a part of me but learning instruments and mastering production software seemed overwhelming. When I discovered AI could help, it opened up a door for me to finally explore this passion without needing to go through years of technical training. It doesn’t mean I don’t respect traditional music making. I actually admire it a lot. But for me, AI became a very helpful tool to make that dream of creating music a reality.

There’s definitely been a learning curve. It wasn’t always perfect. It took me a while to figure out how to get it to align with what I wanted. I see it as part of the process just like any other creative work. I’ve gotten better at shaping the AI results and making sure everything feels cohesive and true to me. At the end of the day, I’m just having fun creating music, designing artwork, promoting it and seeing where it takes me. 😊

If I got lucky on the Suno slot machine and I found songs that I actually really liked coming out of the randomness, would I feel the same?

Cards on the table: I'm not a musician, and I have felt a similar frustration myself. I feel like I don't have the rhythm for playing instruments, and I've found myself overwhelmed by production software. I feel like with other digital skills I've found myself trying to acquire, I often get to the point where what I can make gets myself over the peak of acceptability fairly quickly. I wouldn't call myself a professional at Adobe Premiere, but I can knock together an edit I'm proud of for my own work. I don't consider myself a graphic designer, but I think the Fourth Best Instagram page doesn't look half bad. Yet I feel like whenever I try to learn how to make music, I spend too much time cringing at my own ineptitude and I fail to mount the hurdle into the practice to get past that point.

I feel like if I approached a tool with Suno with fewer moral hangups and maybe some lower standards, it would be a pretty powerful exploit for that part of my brain. If I got lucky and made something I thought sounded nice, all of a sudden I made this. If the AI put out a bunch of nonsense, well that's the tool's fault. The actual solution to this is drilled into children: you practice, you study, you get over it. When we say music is what we want for our children, we understand that it is not a natural process. It is an investment to give them a fluency. Suno is a tool that gives that illusion of fluency.

Of course, I found my own illusion and started DJing.

So yes, I have a bit of sympathy for Kawaii Hoe. I mean their music isn't my cup of tea and I feel like we might be at odds when it comes to their entire practice. It's not that an instrument must be difficult, but an instrument must give control to its user. I use the word "tool" here a lot, and I don't think Suno even is that. Suno is a black box. We know the tools that created it, but it by itself is just kind of an object. Sometimes it works. A lot of the time it doesn't.

a tangent: the beginner's guide

We'll return to Kawaii Hoe soon, but I want to walk the line around my own past with AI tools.

I was pretty close with the technology society in UCC for a few years after I graduated giving a couple of industry talks over the years. In April 2021, long before all of this went to shit, I didn't have anything interesting to say about my day job, so I gave a talk on art and AI. It was pessimistic.

I was enjoying exploring taking my coding skills and doing something artistic with it - I'd already learned how to live-code visuals for gigs. Ahead of this talk, I learned how to train models based on datasets, built a dataset out of my partner's artwork and watched a tool interpolate through an infinite space of variations on that work. I thought it was cool as hell, and I was seeing a lot of possibilities for it. I saw a lot of artists building their own tools, training models, exploring the opportunities.

Of course, I saw the grifters. People just stealing others' models wholesale, generating an output, selling zero effort works at auction. I saw people getting very weird about NFTs, got disgusted by that scene and in my talk I joked "The Revolution Will Not Be Tokenized". I just hoped that people would stay cool, stay curious, keep making newer ideas, and steer the progress away from the grifters.

My talk was largely just a plea for people to do some cool shit.

Instead, the models we got were mass-plagiarism machines at scale. About a year following my talk, we started seeing a bunch of projects like Midjourney, Dall-E 2 and Stable Diffusion. Once we got those, pretty much everyone stopped caring about tweaking complicated code, training their own models, all of that. All the prompt-based stuff was everywhere, it started getting stale and samey. It was good for making a meme, I guess, but the public at large got bored, then angry, and before long, seeing visual AI in a project was a mark of poor quality almost by default. This is now the mainstream opinion: When Nicki Minaj started tweeting out some of the worst promotional art for music ever, it was a sign she was washed. When Kanye started using AI in all of his music videos, it was a sign that he was washed, etc. Even Kawaii Hoe will take care to remind you that they don't use AI for their artwork, videos, website etc. Funny to automate just the making music part of making music.

I still think there's a place for AI in visual art, when it's largely artists working for themselves and shaping the technology themselves. I do not feel the same about the planet-scale tech companies. To me, it's somewhere on the gradient between musicians stringing together tightly controlled generative music with code or complex modular synthesis systems and Suno inhaling the entire history of recorded music and spitting something out stochastically. I think the same can be true of visual art rather than music. But also, I think the models were just far more interesting, far more fun when they were in their infancy. When results took effort, and effort wasn't copying lists of other artists names and buzzy keywords.

A thought that hadn't crossed my mind at all when I wrote the talk was the environmental cost of training AI models, in part because back then, the models we were discussing were tiny, often trained within hours on a single GPU or in a free Google notebook. That isn't the case anymore. I went to a talk at SRECon simply titled "Energy Consumption of Datacentres". The slides in that one look, uh... Scary.

But I do enjoy some art that uses AI for visuals. It isn't Kawaii Hoe's music that inspired me to have to come to terms with a Fourth Best view on AI. It's not even really music at all.

tangent tangent: a trading card game made me write this?

CloudCore, a fascinating UK based label with releases by a couple of Fourth Best favourites from Ireland like Sloucho, Rory Sweeney and Brawni, has a trading card game. Cards and their effects are inspired by their roster, or are homages to the community around the label. I've been to two of the meetups around the game; I attended a CloudCore TCG tournament where the prize was getting a card about you in the expansion pack. I didn't win but I have thought about what what a Fourth Best rift would do. Yes, I really do think the game is class. It's very mechanically satisfying - some cool strategic ideas in there.

CloudCore Cards Starter Kit from CloudCore

CloudCore Cards Starter Kit from CloudCore

CloudCore

There's a lot of cards in this thing and a lot more to come, and they all have pretty detailed artwork. There's at least four/five pieces of Sloucho art in there; you can buy a shiny Sloucho by entering a password found in a Boiler Room set. I saw some cards passed out at a DJ set at Primavera Sound.

CloudCore's Discord server is an insanely complex thing to think about, consisting of a bunch of mechanics I've never seen anywhere else, including "transits" between different zones for events, and they encourage users to create their own custom "transit bots", providing Midjourney prompts for how to create something that fits in with their art style but also gives users room for customisation. This feels like a use case for AI art models that sits fine with me, actually; reduced barriers to entry for things that are mostly for fun and for free and aren't products to be sold or anything. I'd rather the models weren't built in the first place but if we have to suffer them, at least that's a way to use it that's not grifty and exploitative.

But the card art had similar vibes, so I asked around at the TCG event what the story was with it. The answer, unsurprisingly, was yes. AI was used to get over the enormous task of generating unique art for the game's significant number of cards and to allow a team of 3 to put together the game, and for cards made for special events to be a possibility.

It kind of sits in this awkward space for me. I really enjoy the game, the label, the artists around it. I love the fact that events can get their own cards and still fit into that whole world. When I think of the AI aspect of it, it does make me feel like it's a shame that the card art wasn't an opportunity to collaborate further, within the community or around it, to make something that even if a bit more chaotic, felt like a celebration of the wider scene around it, or as an opportunity to bring new contributors in. Yet I also see the argument, pretty clearly, that this game might not exist at all, or be as appealing, without that card art. I want CloudCore to win in spite of it, and hopefully in the future without it.

The question I actually want to figure out for myself, the unfinished aspect of the Fourth Best view on AI, is this: What do we do with meaningful art that for lack of time or resources or people ends up using AI to cut the corners, to make the thing itself possible? I don't mean giving a pass to artists who clearly have the budget to not use AI. I don't mean giving a pass to artists who phone in their album artwork on Midjourney either.

But that brings me back to KH.

duped - chatting to a journalist who covered Kawaii Hoe's music

I started on this piece well over a month ago, trying to make sense of stuff like the CloudCore TCG and after being reminded of The Reality by the Choice master list. The first time I heard about KH's use of AI was last summer, on the Discord server for Irish music news site Nialler9, one of the site's Patreon perks. I had seen Nialler9's coverage of the work from time to time and actually listened to a few of the tracks before the curtain was pulled.

Niall Byrne was able to answer some of my questions on his coverage for this article and provide some further thoughts on the AI music story as a whole. I'm extremely grateful for his contributions.

FB: What first convinced you to write about Kawaii Hoe's music? How did it fit in with the rest of your coverage?

Niall Byrne: Much of my work is spent trawling through various platforms and email folders listening to new music. I heard [KH's] 'badbad goodgood' when I was doing this and from an Irish perspective, an ultra-bright hyperpop song with a super strong melody is rare in the Irish music inbox. I felt it deserved its own little shoutout, rather than the weekly Irish songs series which is usually the first port of call for new Irish artists. The mystery of who the artist was and the effective branding of it suggested someone who had their shit together so I always want to encourage that by featuring and offering advice if I think someone can benefit.

Irish hyperpop artist KAWAII HOE is lusting on ‘badbad goodgood’ | Nialler9

The Irish electronic artist impresses with single #2. Kawaii Hoe is a new artist debuting music in 2024 with the first single ‘Kaleidoscope Love’ and

Nialler9Niall Byrne

FB: How did you feel when you learned the music was AI?

NB: It was the first time I had been duped by AI-generated music to my knowledge, so it was a new feeling so I had to sit with it and think about it a bit. My gut feeling though was that it wasn't something I wanted to be associated with. It just didn't feel right to me. Presenting with female vocals also, when it was not revealed that it was a dude producing also is a part of feeling duped, like being catfished by a new artist in a way.

FB: Did you ever suspect something was off about the work?

NB: Part of the Nialler9 work is to encourage new artists through features, and after about three months of some track features I got a bit more curious about how the music was made. Firstly, it was being released at quite a prolific rate and then the song subjects appeared to be reacting to zeitgeisty things like Charli XCX. I think finally there was a mention of AI Music Label somewhere and that's when I copped it, and the artist told me when I asked. 

It took me a bit longer than it should to realise the songs were fully AI but I hadn't heard anything that was convincing before. Upon realising the truth,  I think you can hear artefacts or amalgamations in the vocal, a slight shiny metallic hissing sound, that I took for vocal processing but it's actually the composite AI vocal. 

I get the sense Niall's inbox is an absolutely terrifying thing to see. He aims to get 100% coverage of the Irish music he receives and does weekly roundups of the best of the best. Fourth Best gets far, far less and I struggle to get to half of what I receive. He is forever an OG and someone I look up to when it comes to this site and its mission. So of course, I figured he'd be the one to know a bit more about if Kawaii Hoe's case is a wholly unique one.

FB: Are you getting more AI pitches? Are they being honest about being AI generated, or can you tell something's off?

NB: It's hard to tell if I'm getting more AI pitches as if there is AI used it hasn't been mentioned directly. Any of the other AI stuff tagged as such has been uninteresting to me or just bad or bland. KH at least, had created their own specific sound world here but is that art or a kind of prompt programming? I think it's the latter currently.

What's to stop someone from replicating even Kawaii Hoe's tracks with AI ultimately? Pop will eat itself is already a thing.  There are of course, the moral issues of training a model to sound like another artist, or the whole history of recorded music. Musicians are inspired by others but those influences and inspirations usually come out in a different nebulous way once it travels through the human condition, it's not a simple output, and that's what makes music so fascinating isn't it ?

Niall also offered some thoughts on AI generated music that to me harkens back to the earlier points that I made - we had an option to create a world of useful tools for artists, and there are tons of examples of those out there in the world, but the decision to centralise on The Prompt as a basis for all interaction basically just means we're stuck with slop machines.

I think it's chalk and cheese to compare AI as a vocal effect to creating entire songs with AI. They are not the same. We all have to figure out how we feel about listening to music with the only human input being some smart prompts.

it all comes back to charli xcx somehow

I noticed a few times in our conversation, KH mentioned part of their motivation is to transform their writing into songs. They did hint that they started out writing with AI's help, but has been writing their own lyrics on later material.

FB: How do you approach writing the lyrics for your work? Is AI also a part of that process?

KH: For [post-album single] ‘charli xcx’, I wrote the lyrics 100% by myself since no AI can reproduce my feelings. For The Reality, I just want to be happy and money can definitely help which is why I daydream about being rich and famous; but it goes back to reality eventually. All the songs are based on my real life experiences. You have to have your original idea for AI to output something meaningful that belongs to you so I’m sorry to disagree with the misconception that AI means fake and unoriginal.

I didn't really know how to respond to "no AI can reproduce my feelings" in a conversation about their AI generated art, and I kind of ended up just letting that one slide.

I've kind of noted above that I find KH's music difficult to listen to. 'charli xcx' is the weirdest thing they've ever done, in part because they're using a machine made to make formulaic pop music and giving it lyrical content that feels nothing like that. It sounds like it's completely falling apart. It's one of the things that raised suspicion as to what they were up to. Maybe that's what appeals to them about the work, although I'm not sure.

charli xcx is written as basically prose, with long run-on sentences. It paints a story of isolation, fear and pain, trying to use music as a means of connecting with community. One of the lines reads:

There's something autobiographical in Kawaii Hoe's work, peering out from the layers of algorithmic separation, its twisting of gender and technology. It would break your heart a little bit. It's there too in that album, with its more escapist and fantastical imagery, daydreaming about being rich and famous, although there they just seem like pop songs. charli xcx has this bizarre closing line that the AI voice almost appears to struggle to read, the word adulthood coming out all wrong:

I don't really know where to begin with this song. Before speaking with KH, I couldn't even begin to picture what it is or what it's trying to do. Now that I can see this a little closer; a moment from the artist's own life, trying to make sense of all of this: making silly little pop songs with a very weird piece of technology and trying to participate in an alien world. The song itself is an abstraction of... I don't want to say cry for help exactly. A longing that does not know how to describe itself.

I think I come out of the other side of this thinking... the tools Kawaii Hoe has probably shouldn't ever have been built and the sort of music they're putting out is making streaming platforms worse and all of that. Yet I can't blame them for trying to figure something out about themself through this strange new medium. I just don't know how to feel about the means of getting their voice out there.

Kawaii Hoe's music shows up in some weird places. They have 30,000 monthly listeners on Spotify. Their music shows up in playlists which pretend to be other playlists, with titles like "DJ Mag Top 100" and "RA Top 100" or a blatant rip-off of Spotify's Fresh Finds by probable scam merchants TPA (The Playlist Agency). It's hard to know what's going on there. It's an entirely different article I'll do some day, but artists: never pay any organisation that offers playlist placements for money. This is a scam and it puts your music's position on streaming platforms at risk.

I would guess, based on some data I have access to, that TPA often surfaces music and spams it with bot traffic without the consent or foreknowledge of the artists, in an attempt to pick up new customers for their dubious playlist services. As such, I am not accusing KH or their PR firm of buying playlist services or bot traffic. In fact, the more I looked into why they pull somewhat respectable streaming numbers brought me to a weirder fact: they might genuinely be big in Indonesia.

I caught this note in the press release for their Christmas song.

Kawaii Hoe's music has sparked massive engagement on TikTok, especially in Indonesia, with user-generated videos (UGC) featuring their songs accumulating over 19 million views and 715,000 likes.

I'd actually also noticed some of this during my research. When I googled lyrics from their work, I genuinely kept finding Indonesian TikToks. Their lyrics have been translated into Indonesian on MusixMatch. There's a real audience there.

And I don't know what to do with that either.

time to get brutalised

This article needs a conclusion. We've been circling the idea that AI is cheapening music, duping listeners and the media, and while I don't want to point fingers at any given user of these tools, I agree with Alper's "these people are your enemies" take aimed at the men in the ivory towers. Then again, I think about this dilemma when I see news stories about people throwing eggs at Tesla Cybertrucks in the States. Maybe the most efficient approach is to remind people how fundamentally weak all of this is, and maybe a bit of shame for your participation is a good thing. If you use AI visuals, people are very very open about hating that. The same will be the case for AI music.

One of my favourite films of the year, The Brutalist, received a surprising amount of backlash over two alleged uses of AI. One count was from an interview where it was mentioned that Midjourney was used in some of the development phases of the film (the director denies any of this material was used in the final product), and one from the use of a tool called Respeecher that was used to correct Adrien Brody's performances in Hungarian at points in the film. I feel like railing against big, ambitious, difficult works of art because machine learning was involved somewhere down the chain is tilting at windmills. It feels like we're not going to win the war on prompt-based slop if we're going after weird niche stuff that is improving the art. When I told my girlfriend about this, her response was something like "nah, I think the actor should just do their job". A friend of mine argued that Brody should be ineligible for an Oscar because of the use of Respeecher. I sort of just don't care.

I can't look at The Brutalist with its 215 minute runtime, shot on a previously-dead film format and its immense scale and think "yeah. this one's the enemy".

Spotify's an enemy and I finally cut ties last month. I had been talking about this for about a year and investigating other options. Eventually I just jumped ship when they threw that party for Ben Shapiro after Trump's inauguration. I use Qobuz now. I like Qobuz. They actually write about music! They seem to pay a less-terrible royalty rate (the Benn Jordan video I posted above seems to imply they pay more than a fraction of a penny)! I saw a Tony Bontana album in their monthly roundup and I wondered if I'd ever see the likes of that from Spotify. Not after they laid off hundreds and committed to being a slop factory anyway, that's for sure.

Suno and Udio and the rest are enemies too. Be it the theft at scale or the fundamental hate for their art form.

Niall Byrne has his own, more familiar version of "music is what we want for our children". It's Mrs. Doyle's refrain.

NB: It's part of the end result of the tech bros taking over music, trying a cheat code or to game the system or win at art by creating content to share and make money. It loses the human toil and work that is rewarding in the creation of art. Something which the tech and AI evangelists and the likes of Daniel Ek have never had an inkling of in the first place.

Like that Suno guy's logic, it's alien to him that someone might enjoy the process of making music - MAYBE I LIKE THE MISERY!

The Teamaster comparison is apt. A lot of people want to take the misery out of making music. A couple months back, I saw Max Alper post on BlueSky about a growing trend of of DAW plugins with names like "SynthGPT", allowing for prompt-based direction inside a traditional music production environment.

The flattening of music culture isn't happening just at its distribution via streaming, but also at its root of production. Tools that sell themselves as a one click easy solution to instantly create xyz style or genre for the sake of creating quick turnaround audio content that is completely removed from the culture in which that style originated from are so fucking cynical.

These music tech companies participate in the death of their own culture by presenting their products as sonic wallpaper generators, effectively teaching aspiring artists that "this is how you should approach the process of creation: quick and easy and from a template we design for you." open the MUSIC SCHOOLS.

Artists: love of process must outweigh love of final product. Why else are we doing this if we don’t love every fucking second of creating it? If you view creative process as a grind and not the closest you’ll get to divinity then idk what to tell you, there are plenty of more profitable grinds out there besides art.

On that list of 211 Irish albums of last year, I can imagine all the others among them took effort and maybe just a little bit of the "misery" that tech firms think need to be replaced. And that's to be celebrated.

In my conversation with Kawaii Hoe, I met someone who I think genuinely believes that what they're doing is a meaningful expression for them. And I don't think mild buzz on Indonesian TikTok videos is delivering them much income, so fair enough. I'll let them have it. It's why I wanted to avoid naming them at first.

The problem with these tools is that if making slop is mildly profitable, there'll be people linking up the work like sweatshops. Doesn't matter if it's lofi study beats or Shrimp Jesus. It's the new Axie Infinity. Scrapping for crumbs at the money machine. These are their peers, even if they're convinced they're in it for more than a get-rich-quick scheme. So I'm worried.

KH pointed me to a Reddit post from a couple months ago of someone trying to start up an AI music label. The post reads "We have an AI record label called Lost Echo Records. The artist photos, bios, backgrounds are all AI generated. We use Suno to create all of the full songs from beginnings to end. [...] We have everything from country to metalcore." and it goes off on a rant about how they've been getting zero followers anywhere. KH said "You can tell the difference and my dedication comparing to this".

It's true, I think. I reckon KH might have a future with this. Not with the AI songs themselves but if they're going to speak the languages of distribution, PR, design and marketing, maybe there are ways that they could participate in music and thrive. I can't say the same for people who just pump out content and complain about the lack of traction.

I don't think there's an existential threat to music from AI. I think it's going to harm people's livelihoods, but music will continue. I don't think we'll ever reach the point that we'll hit the "death of music culture" at the hands of the snake oil salesmen. This tradition will survive even if all profit is stripped from it. In spite of the misery, it is what we want, and what we aim to protect. If you've made it to the bottom of this screed, you're in the fight as well.

What do we do when we find art that we see value in that uses wholesale generative AI models? I think we call it a shame. I think we call it a missed opportunity. Any time you use a prompt-based engine for images, video or music, you're missing a chance to pair up.

Music is what we want for our children, and music has the right.

Kawaii Hoe's new song is out this Friday, but I think I'll be scouring the list.