I've been waiting a
while to get my hands on this topic...
So there was a recent issue of
Admin Magazine in which the letter from the editor,
Artificial Intelligence in Review, summed up my thoughts much more eloquently than I'd be able to write myself--so I'm going to use a few quotes from that as my jumping-off point.
"...[M]any see AI applications such as ChatGPT as enemies and threats to our existence. The problem with that thinking is that these applications only know and can use what we give them... A person might build a program that can cross-reference, compare and contrast, and even draw some primitive conclusions, but it has to be fed the data from which it draws those conclusions. The inner sight that only the human mind can experience is that spark of genius that sees relativity, bent space, and the possibility of subatomic particles... Only the act of falling in love or losing a loved one can inspire the human heart to write great song lyrics... Remember that artificial intelligence is artificial. If it had existed 1,000 years ago, the printing press would not have been invented any earlier than it was, the Americas wouldn’t have been “discovered” any sooner than they were, and powered flight wouldn’t have happened any earlier, either... " (ADMIN Senior Editor Ken Hess for Admin Magazine Issue 77: Secure CI/CD Pipelines)
Now, I'm no expert and I'll probably accidentally misuse some tech vocab words here, but hopefully my point gets across!
I first want to focus on that part about AI only knowing what we give them. There's been many reports of AI image-generation recreating either offensive imagery, or even
being inable to depict non-white people. These kinds of issues have been around for a
while, which you'll know if you remember certain facial-recognition software having issues identifying non-white faces, or the Twitter AI bots that started spewing hate speech as soon as they launched. In the NY Times article I linked, the Senegalese artist Linda Dounia Rebeiz says about DALL-E 2, "It defaults to the worst stereotypes that already exist on the internet." It seems to me that AI is forever limited by its input data. I'm not alone in noticing an uptick in articles that are completely devoid of any original content, and you can usually pick them out as being written by AI immediately, like in this
example. When using Google Images to find photo reference recently, I came across some pretty horrifying AI-generated results. Not only are the resulting images unsettling, they're not even useful!
That being said, a lot of people are fearmongering about something they pretty much should be familiar with already. Machine learning has been around for decades, we've heard complaints about the Facebook and YouTube algorithms near constantly, and predictive text on phones and email services continues to be passable at best. Labeling a lot of stuff we've seen before as "AI powered" definitely feels like slapping buzzwords on stuff for viral marketing. AI
art in particular feels like a very similar grift to crypto and NFTs. Sure, it may get a bit bigger and it will probably get better, but it's not going to replace human-made art in the same way Bitcoin has not replaced the US dollar. It will continue to "flourish" mainly in grifter communities as people continue to search for greater fools. At least that's all I see happening...
Anyway, I'll get to the positives now. A few folks have already mentioned using AI as a tool, but I've been a bit stumped trying to find examples of it being used well (read: in a way that doesn't feel like a shallow cash grab). My partner told me about a good one though! Oneohtrix Point Never's newest album, Again, makes use of
OpenAI Jukebox, Adobe Enhanced Speech, and the Riffusion neural network to create these incredibly otherworldly, dreamlike tracks. The songs I listened to (which may or may not be the only ones that made use of AI, I can't really tell from
what I've read) are Krumville, The Body Trail, and On an Axis. Of course, you probably don't
need AI to create these effects (the cut-up speech, for example, could be from recordings of real people talking and I wouldn't know the difference), but it seems like a neat thing to have in your toolbelt as a musician.
Ending with Daniel Lopatin's own words on this seems fitting, so here's his take on his use of AI on his new album:
"It’s so banal. To me the thing that’s exciting about AI is that they seem to behave a lot like we do, in that they confidently misrepresent ideas, they misconstrue history. Those two things are so on point to me that I was like, ‘Yes, this absolutely belongs in the OPN toolbox."
"This is like the mediaeval period. It’s really funny to me. It can be quite haunting too. I don’t even know where the eeriness ends and the humour begins. I’m not so impressed with it sonically, but using this stuff influenced the arrangements, particularly Jukebox. That’s the one that seems to do the weirdest stuff with rhythm and time. It fails so majestically that we have to take note of it." (
Selim Bulut interviewing Daniel Lopatin for AnOther Magazine "How Young Adulthood and AI Shaped Oneohtrix Point Never’s New Record")