The invention of ML tools is as significant as the printing press. These technologies will underpin the creative practice & economy of the 21st Century
Permanently moved is a personal podcast 301 seconds in length, written and recorded by @thejaymo
New Cultural Technologies
A year ago (Episode 21-29) I talked about my first tentative steps with AI models and image generators. I’ve had thoughts over the last 12 months. Recently I’ve come to the conclusion that the short term impact of these technologies are under estimated and under discussed
In the last year alone things have gone from CLIP and VQ GAN running on Google’s Colab cloud compute platform. To DALL·E 2, Midjourney, Dall-E mini etc. ML models are now household names.
I’ve been playing with Nvidia’s NeRF software recently. You take a bunch of photos of a location from different angles and it figures out a 3D model based on the set data. It’s wild. There’s also some code where you plug in a photo of a dog, and it spits back a 3D model. It’s not great, but it will be. 12 months.
The speed and rate of development is staggering. Google’s Magic Eraser or Photoshop’s Content Aware are pedestrian.
Articles at the moment are about the aesthetics of DALL·E 2. Or discussing the copy and moral rights of the artist a model has been trained on. The open source community up in arms about Github’s copilot.
For me conversations about the training content of AI models are the same as the ones about our collective cultural imagination. Both ML models and the MCU are shaped by the same capitalist logics.
AI Models are complete worlds. Snapshots of The Culture.
Each one is different. Shaped by the preferences and bias’ of the neural network used and the content it looked at. DALL·E 2 is a complete world the same way Wikipedia is.
These models are a new substrate. Cultural technologies.
Different models produce different images. Each has its own vibe, pallet, shape, quality. Regardless of the prompt. That’s CLIP, that’s Guided Diffusion, that’s Midjourny etc.
AI models are like film stocks. Kodachrome vs Fujicolor Pro.
A photographer’s style is as intangible as the filmstock they use. But until recently we wouldn’t have dreamt of saying any of that should be under copyright.
Dan Hon, talking about style transfer in his newsletter recently said:
You pay to have it [the model] create images in a certain photographer’s style and, well, nobody’s stopping you from then using those images in whichever way you wish.
Replace photographer with artist, musician, fashion designer, architect whatever. The sentence still holds.
We are in a really weird place in culture. In 2018 everything changed. The Blurred Lines lawsuit.
The worst copyright decision ever made in history.
Until 2018 the law was clear. A song is the: Chords, Melody, and Lyrics. Blurred Lines sounds nothing like Marvin Gaye’s Gotta Give It Up. But they feel the same.
A jury of Pharell’s peers thought this is something that can be owned. They judged that he’d infringed on Gaye’s vibe.
So now, apparently, you can own a groove. Or rather, own the copyright over how a song makes people feel.
Pharell, producing Blurred Lines, did style transfer.
Of course I have sympathy for photographers who find that the addition of their name in a prompt causes an image to look a certain way. But do we really want to live in a world where the style and vibe of a creative work is copyrightable?
Spawning from AI models is a totally different thing to sampling.
Gregory Coleman is the most sampled musician in history. The drummer who played the Amen Break. Yet he died homeless in 2006 at the age of 62. This is a huge failure of copyright. The system is completely broken.
Let’s say that style is copyrightable. Let’s think it through.
It’s 1969. Coleman having just recorded with the Winstons picks up his sticks
Tist Tist ta tss, tist ta tss. (Makes spang-a-lang noises)
Ah ah ahh. That’s swinging buddy.
You can’t do that.
That’s been Kenny Clarke’s since 1939.
What does this mean for Reggae, trap, blues, folk? Whatever?
Post Blurred Lines, this is the world we are in already. Calls to remunerated artists whose names are style modifiers in prompts under the current system – is straight up right accelerationist.
The development of ML tools is as big a deal as the printing press. The technology that begat copyright in the first place. But models are a medium. They aren’t copied with paper and ink. New things emerge (are spawned) out of them.
It’s clear too, that these new technologies will underpin the creative practice and economy of the 21st Century.
Over the next 20 years the creation of, and ownership of these new cultural technologies has to be redrawn. The entire framework. Culture, copyright, creative remuneration .
But this future speculative paradigm will emerge out of the conversions and positions we take today.
The current media environment encourages us to take sides. To develop coarse arguments based on reactionary headlines. The biggest cultural contribution anyone can make to benefit the artists of future, today. Is to develop a nuanced position.
Think about the coarse arguments about sampling back in the 90’s. Then think about Coleman dying homeless and alone.
The script above is the original script written for the episode. It may differ from what ended up in the edit.
If you’d like to support this blog and its various projects, please consider making a regular contribution here. It genuinely helps me keep things up and running, so thank you!