The AI Unbundling

I'll believe it when I see it. How many self sufficient, independent digital artists are actually making a living doing something other than webcomics or furryporn? SD is amazing technology for sure, but this notion that it's going to incredibly disruptive is the latest in a long line of AI hype. It ignores the simple fact that there is already more art produced and uploaded onto the internet, every hour, than can meaningfully be consumed. Independent artists already had to compete with millions of art uploaded very minute, what is a few million more?

In the unbundling story, the premise is I gather that what is disruptive atm is that what's left as the god-in-the-gaps human role, is now democratized in the the technological acceleration in the the creation-substantiation phase at the head of the snake.

What we need to start preparing for is that the creation phase is next; it's a mere engineering problem.

Yes, that's somewhat sardonic; it's going to take some years. But only some.

What I am saying is, before we are ready, the aspects which remain human-domain in the cybernetic collaboration between humans and AI in what well call "content generation," namely, the agency and the intention—the "idea" which today is, the prompt; the editorial and curatorial functions; the insight and lightbulb and all...

...that is going to become the domain of the AI as well.

What inspires "art" or whatever else?

The filtering of media streams (aka cultural discourse, the zeitgeist) and the identification of symmetries, analogies, serendipity, humor... whatever that is, it like the "substantiation" phase is going to be ceded to the super-human in a few years.

I speculated recently to a friend that the way the fine-art market reacts to this is going to be interesting. We're going to have Wintermute on tap. I wondered if gallerists will represent AIs, or, spin them up long enough for them to emit some stream of work then be ceremoniously destroyed.

Per Ex Machina that may not go so well in various ways.

Anyway. The writing is on the wall.

And it's not in OUR handwriting.

The 20B images don’t look that much more impressive than what SD is already doing (aside from the ability to render text), and in some cases they look worse. It’s hard to tell because the resolution is so small, but even in the 20B “astronaut riding a horse through a pond” image, it looks like his hands are still nonsensical.

I really don't see it. What tools like SD, DALL-E etc do is essentially similar to what Google Image Search does, except that the results are presented differently - instead of showing you the results it has indexed in some order, it picks features from the results it has trained on with some priority, and presents a kind of amalgam view of the entire collection at once.

The only significant advantage compared to Google Image Search is that AI-generated art is getting around copyright, allowing you to publish it legally under your own name. This makes it very similar to Uber, which got around taxi regulations that were impeding market access.

Essentially, if we ignore copyright, there are relatively few illustration problem you could have where DALL-E/SD/... would help more than searching for an existing image. Branding is an obvious one, where the need for unique-ness is not only legal, but also practical. The other is that DALL-E/SD/etc can also sometimes produce combinations of unrelated images, thought that is offset by just how bad their results are in other places where there is plenty of existing art to choose from (especially anything involving images of humans).

So basically, you read over a book(with amazing art) over, know your terms(or maybe know how to use a neat visual dictionary), and now you can paint in many art styles.

Of course there will be some more learning while playing with stable diffusion and getting a feel how to talk to it.

But it's very far from the old skill of neing an artist, not much knowledge will transfer.