AI Art is the Insincerest form of Flattery?

What feats would humanity achieve if it networked together the largest amount of computing power in history? 

Found on Twitter

A day or two ago I suddenly noticed a ton of memes re-done in a Studio Ghibli style, and I figured some new AI tool was to blame.  Sure enough, this turns out to be the result of a new feature put out by OpenAI's image generation tool.

It apparently hit the right spot of images people will enjoy enough for half a second to hit share or like, and so they exploded all over social media.  Someone even put what must have been many hours and some actual money into redoing the Fellowship of the Ring movie trailer, shot for shot, in this AI generated Ghibli-style.

A number of people were also pretty angry about it, pointing out that Ghibli's hand drawn animation style is one of the few artistic bright spots in a movie landscape over the last 20 years increasingly dominated by formulaic and CGI fare.

It seems to me this actually underscores the sense in which "AI art" is different from what humans can do.  After all, the way that this type of AI works is it ingests a huge number of examples of something, whether that is writing on various topics or images, and then on command it puts together elements which it associates with the request.

So in respect to images, what AI can do pretty easily is put together elements that are commonly found and easily identified.  I pretty commonly use AI generated images to illustrate my pricing substack posts.  If I didn't have that option, I'd be using free stock photos, and the only reason I illustrate the posts at all is because LinkedIn and Facebook are show a post a lot less if it doesn't have an image.  But one of the things I've found in generating those images is that AI really is only good at combining very simple existing concepts.  So, if you ask it to show a Starbucks cup of coffee sitting on a counter, it can do pretty well.  If you ask it to show a traffic light with a dollar sign imposed on each colored lens of the light, it's a little shaky. Ask for an airliner leaving a jet trail made of dollar signs and it completely fall apart.

This is why trends like these Ghibli style images underline now the replicability or artists but how much they are needed.  AI can rip off Ghibli's style because it already exists.  But AI is not able to generate a new style.

And indeed, something like the AI Ghibli Fellowship of the Rings trailer attempt just ends up underscoring how little this is like anything Ghibli would do. After all, Peter Jackson films have their own visual palette, whether done moderately well as in Fellowship or with utter silliness like in the Hobbit movies.  Ghibli has a whole different design ethic. The Ghibli movie of Wizard of Earthsea (not one of their best efforts, honestly) underlines that Ghibli is not just a drawing style, it's an entire approach to visual imagination. That AI Fellowship trailer is not how a Ghibli LotR movie would look, because Ghibli does not imagine the way that Peter Jackson does.

To be honest, I think this is something that most people who consume images (whether on their phones or in the movie theater) don't consciously think about. For a lot of people, seeing a Ghibli style image results in a quick "Oh Cool!" reaction, and they move on.

That's a bit unfortunate, but probably pretty universally how things have been. Most people do not think deeply about how the things they enjoy seeing come into being.

But underneath, the thing that draws people into a piece of art for more than a half second is that it has some originality. Indeed, real originality of vision is special enough that even copies of it can fascinate at least a little bit.

You can't have derivatives without originals.

And similarly, if you don't have original writing and original images (both art and photos) you can't generate these probability-based AI blends which put together several concepts and produce some new thing which draws on various sources but doesn't truly create.

This means that AI image generation is necessarily pretty ephemeral, and if you want anything with real creativity or quality of vision, you need to have a human create it for you. So, for instance, while I use throw-away AI images to illustrate posts so they'll show up on social media, when I wanted a logo for Pricing Evolution I went to two different artists and commissioned original drawings. Both, I think, ended up having a personality which you would not get at all from AI, because they're originally created by humans.

In the realm of writing, I've heard several writers joke that they're now "writing for AI". The sense in which they mean this is that while AI is very useful at searching the wide world of writings available on some particular topic and putting all of that information together, what AI is not able to do is provide original insights and research. So they mean that they are focusing on writing original work, the sort of thing which will be synthesized and summarized and linked to even as the swarm of bots continues to serve up information to people.

And I think similarly, when it comes to any kind of art, original styles and creations will continue to come from humans, even as AI allows people to throw up quick imitations or fusions which hold the eye for a moment. 

Without humans, there would be nothing to imitate.