Part 1: Yes, You Should Probably Hate This
AI in art is a loaded topic. As it should be – plenty of the complaints about it are merited. Aside from the issues that apply to AI more broadly (environmental impact, questionable value vs. the insane levels of investment being funneled into it), AI in art raises a few specific concerns.
For one, AI is entirely dependent on past, human-created works. The technology isn’t capable of creativity – anything it produces is the result of an algorithm assembling bits and pieces of actual artists’ actual art in response to a query. Of course, human artists borrow from other artists all the time, but with AI the human may be responsible for as little as a single prompt – a few words that kick off a mechanical process of cut-and-paste that’s complex enough to fool us into seeing the end product as something more original than it can possibly be. Without a human artist, this is just fancy theft.
The second big problem is AI being used by non-creative groups and individuals who are looking to acquire an “art-like” product without paying human beings who possess creative skills. If the artist is no longer a necessary part of art creation process, then they are just an inconvenient middleman. Meanwhile, the technology taking over the process is, again, simply mechanically mining past human creative works for free (fancy theft).
There are other problems with AI in art that are more subjective but no less legitimate. Most of the art being pulled out of these tools is absolute slop. It can be churned out easily and endlessly, and it’s rapidly occupying more and more space in our lives both on- and offline. It’s hollow and soulless by its nature. God knows what overexposure to it is going to do to us long-term. The idea of children being raised on AI slop on YouTube, having infinite hours of machine-made content injected into their developing brains, is utterly dystopian, and it’s happening right now.
This is ugly, nightmarish stuff. Artists, probably more than anyone, have every reason to be anti-AI.
However – if we set these ethical nightmares aside (which we should not! But stay with me here) – I don’t think it’s completely true that art can’t be produced with AI. Most of what’s being made, what I’ve described above, is not art. But I also believe artists can make interesting use of AI as something like a chaos sampling tool. AI can be used to experiment, to build and tear down, to inject seemingly random echoes of past art into the present in unpredictable ways. As long as a human creative mind is pulling the levers and actively curating the output with artistic intent, the end product can be art.
Part 2: OK, So I Did Actually Write Some Songs Though
I made a lot of music in my teens and early 20s. Some of these songs were meant to be funny, and were recorded under the appropriately dumb name Fidel and the Castronauts with my pal John. The number of people we successfully entertained was probably in the single digits, but we had fun.
One Castronauts song I’m sure no one enjoyed but me was called “Computer Vomit.” This track was a product of the audio freeware I’d been using on my family computer crashing; inexplicably, it then output a 12-minute audio file comprised of mangled fragments of recent recordings and harsh digital noise. I doubt anyone else has ever listened to it from beginning to end, but I still find it fascinating.
Other songs at that time I wrote more sincerely, in the late ‘90s/early aughts indie pop vein I was then obsessed with. I never really found a vehicle for these, partly because I didn’t have the confidence to share them too broadly. I’d perform them solo for friends on occasion, though I didn’t think I was a good enough guitar player to form or join a real band. I recorded basic demos, a handful of which I’ve managed to hang onto as low-quality MP3s over the years. Some are more embarrassing than others, but I’m glad I still have them.
Part 3: Where Suno Comes In
Since its launch in late 2023, Suno AI has mainly been used to generate songs based on simple user prompts. You can feed it original lyrics (or just have ChatGPT do it), suggest the style/genre you want, and then with one click you get a song.
It’s pretty entertaining! This past spring when Kyla was on the other side of the country managing a serious family matter, I was able to entertain her long-distance with silly Suno-generated songs. Is it art? Not really, but it served its purpose.
Last month I discovered Suno had added a new “cover” feature in beta. Now you could upload your own original recordings (limited to two minutes long) and prompt the AI to perform a cover version. I uploaded some of my old demos, and the output was pretty jarring. Unlike those earlier, wholly AI-generated novelties, these still sound like my songs – mechanically interpreted, produced algorithmically, and yes, products of fancy theft – but still, my words and my melodies.
Playing with this feature became my obsession for a few days. I tested the limits of what Suno could do with my songs, generating hundreds of iterations, and I came away from it pretty impressed. Most of the output was garbage, of course, but here and there it was not.
Part 4: ARCHIE WHAT HAPPENED
Here’s the end result: Archie What Happened. A collection of songs written at various times over the past ~25 years, rapidly remade by Suno AI in different styles over the course of a few days.
A track-by-track breakdown:
- “Intro (Archie’s Theme)” – The earliest version of this one was a loop of the chorus melody I made when I was 17 (it appeared on a Fidel and the Castronauts album as “Flava Loop”). I wrote the verse chord progression a year or two later, but never came up with a vocal part I liked. This version hews closely to a MIDI arrangement I created sometime in college under the very stupid working title “Enemy of the Thermostate.”
- “Helpless” – In my teens I always imagined this would be the first song on my someday-album. This is also the first one I tested with Suno, mostly because it’s short and simple but still requires the AI to handle some unexpected chord changes. Pretty obvious this was written by a dreary Radiohead kid, but I stand by it.
- “Paragüero” – A light, bouncy Beck-inspired track. The title and melody are super old – like, Beck didn’t have an album called Guero yet, that old. Twenty-plus years later I got an idea for how this could work as a complete song.
- “Better” – This is a newer one. I’ve had the chorus bouncing around in my head for a couple years, and it’s stuck with me mostly because I think the line “I can’t believe it’s not better” makes for a funny hook. I came up with verses on the spot, intended in part as a test to see how Suno would handle crazy key changes (pretty well, it turns out). I always figured this would be more of an uptempo, New Pornographers power-pop kind of song, but I liked this Sarah McLachlan-y take the best.
- “To Die Well” – This is a much more recent song than most here, entirely written in 2022. While a lot of these naturally represent an angsty teenage perspective, this song looks back on that mindset from an adult point of view. It’s also written as a dialogue with a friend who committed suicide in his early 20s. Heavy, but one of the few lyrics I’m really proud of. This track had to be a little longer than the others, so this is the one instance where I had to play around with Suno’s “extend” feature and do some stitching together in Audacity.
- “Vim” – A fun little quarter-life crisis song. This might be the one I’ve played the most for people over the years. I always heard it as this New Wave kind of thing in my head, and this is the first time I’ve been able to actualize that.
- “Metronome” – I believe this is the oldest song here, written and recorded when I was 15 or so. The lyrics were very cringe and teenagery and didn’t mean much, but I still like the melody. I cleaned up the lyrics and stumbled onto this ethereal sort of Enya version, which I like a lot.
- “Friend” – I wrote this one in 2010. This is the only track where I feel like Suno’s robo-vocals get a little distracting, but it totally nailed the jangly Peter Buck guitar thing I was going for.
- “To the Nines” – I recorded a version sometime around 2008 that serves as the basis for this. The vocal part has always been there, but it never had real lyrics, so this was a test in Suno’s ability to process my gibberish. I really like how the AI repurposed the single-note guitar riff that runs through the verses, and there’s a wild psychedelic crescendo section that it came up with on its own, as far as I can tell.
- “No Big Surprise” – Like “Archie’s Theme,” this one is based on an instrumental arrangement I created in Reason sometime in my late teens, so Suno didn’t have to interpret too much. I believe the song had lyrics at one point but those are completely lost to time, so I came up with new ones that adhere to the original melody.
- “Sunlight” – I wrote this one at 23 or 24, shortly before Kyla and I moved to NYC. I didn’t want to stray far from the original demo’s sound on this one, which can actually be more of a challenge with Suno vs. doing something radically different.
- “Outro (What Happened)” – Going through this cover process with Suno leads to all kinds of unexpected output. This, for example, was intended to be a Mulatu Astatke-style take on “Archie’s Theme,” but the AI went off and did its own thing. Fascinating and unknowable, a new (but much shorter) “Computer Vomit” for our times.
On no level do I think of this as a real album. There’s an excitement, though, in hearing these songs fleshed out to a degree they’ve never been outside my head. There’s still a hollowness and a soullessness to it. It’s best to consider these still demos, just leveled-up via new tech; I’d much rather hear them performed by real people.
Anyway… is this art?