A pragmatic approach to the current AI situation.
I've been torn about who to side with in the AI debate, since it's so emotionally charged. I have thus wanted to weigh my position properly before committing to a hill I would be comfortable dying on.
One of the first posts I made on Substack was about how I was cautiously optimistic about AI. In summary, it postulated that asking ChatGPT for answers was probably going to be an improvement on how we do things now, where an alarmingly high percentage of us basically just get all our information from social media.
And… Putting our trust in advertising networks masking as “sharing platforms” to provide us with news and information, was probably not the best idea in the first place.
The problem with AI
My optimism about AI has since been somewhat reduced, as I’ve been made aware of how the movers and shakers in the AI field do their actual moving and shaking. And by that, I am of course referring to the seemingly concerted effort by Silicon Valley’s tech CEOs to leverage generative AI as a mechanism to turn working artists into starving artist. Or by “democratizing art” as they like to call it.
By now it’s well established that none of the current generative AI companies - whether they generate images, music or text - are headed by particularly nice people.
It’s also pretty clear that the reason the current crop of generative AI models are so good at what they do - and some of them are quite impressive - is because they have essentially stolen their entire datasets by scraping everything they could find on the Internet, without seeking permission or providing any compensation to the original creators.
This is of course horrible, but probably also completely legal. Because copyright law has traditionally worked in favor of giant corporations, and rarely protected the artists themselves.
Taking a stand
So, as the art loving liberal that I am, it should be obvious that my standpoint on generative AI would be firmly planted in the “against” camp at this point.
Yet… I am still kind of optimistic about AI.
And unlike, say… Members of the gun-lobby, I’m not even going to justify my position with some “rationale” as to why my point of view is “righteous”…
I just think that the current crop of AI-tools are cool.
Immorally created? Absolutely! But still very cool.
Where it started for me
I first became excited about generative AI when some guy made the news for winning the Colorado State Fair's fine art competition on September 5, 2022 with a Midjourney-generated image he called “Théâtre D'opéra Spatial” (pictured up top).
Apparently the judges were unaware that using AI to generate images was now a thing, and even after finding out still stuck to their decision.
And I agreed with them, because it’s objectively a great image that looks like art. It’s specific enough that you can visually parse it, yet abstract enough to trigger the imagination to create it’s own version of what’s going on.
And this is of course where the pebble in the proverbial shoe of generative AI is currently located. Because the AI is not really coming up with its own “art” through creativity, or for that matter reasoning. It’s just really good at plagiarizing the work of the artists it has been trained on. Often down to their very signature.
As such, it has become very doable for anyone to use generative to AI cook up something that looks kind of cool, even if they have no artistic ability or talent whatsoever. This again, has led to the Internet being flooded with low quality AI images, from people that has been given a really powerful magic paintbrush that creates what - at least to them - looks like art. And also gives them some attention on social media.
Which understandably is infuriating to those who are trying to make it as artists using their actual art skills.
Imperfect creations
But although the majority of people somehow fail to notice it, these AI images are always flawed in some way.
Weird hands, dead eyes, too many limbs, lines that should be straight but aren’t… The list goes on for quite a bit…
There are of course those who claim that you can make generative AI produce perfect images if you can just formulate the perfect request. Or engineer a specific enough prompt, as the people who call themselves “Prompt Engineers” like to call it.
Personally, I think these people are delusional. Because no matter how specific you are when you prompt an AI model to generate something, the most determining part of what you get back is always going to be randomly generated.
Also: “Prompt Engineer” is not, and should never be, a real job.
But is it art?
This is something which is fairly easy to answer, and that answer is:
- No!
What generative AI makes is definitely not art. And that is also why established artists should not be afraid to loose their job to AI. It is however “art adjacent” enough, to be used as building blocks for something which may later become art.
And this is relevant in the case of “Théâtre D'opéra Spatial”. The creator claims to have used 624 text prompts and input revisions to get Midjourney to create the image, and then manipulated it with Adobe Photoshop to make it work.
And I see no reason to doubt this claim, because it has - in my estimation - clearly been worked on by hand.
And I strongly believe, that once you start working, deploying your “artistic eye” to separate what parts of an AI generated image has potential from those that do not, take these out, apply a healthy dose of art direction and the art skill necessary to manually fix all the weird errors and glitches that AI tools currently produce… You are in time going to end up with real art. Even if the stems of the final image may be rooted in plagiarism.
Besides, plagiarism is, and has always been a big part of an artists journey. Particularly for young artists with talent, taste and artistic ability, but who still have yet to find their own voice.
To illustrate this, allow me to paint the following picture:
What we can learn from Motown
It’s 1979 and the band “Chic” has just put out their disco soul single “Good Times”, which would eventually hit number one on the Billboard Hot 100 on August 18th the same year. It’s a massive success for songwriters Nile Rodgers and Bernard Edwards, but there is also some annoyance, as the bass-line is promptly stolen by brash hip hop upstarts “The Sugarhill Gang” for use in their debut single “Rapper’s Delight” (which would also chart the same year). Hostilities are however avoided due to the parties agreeing to share songwriting credits for “Rapper’s Delight”.
Now “The Sugerhill Gang” predated what would later become the practice of sampling - which is something of a “best practice” in the world of rap and hip-hop - so they actually had to play the base-line themselves on real instruments, instead of using computers and samplers to do it for them. But “Good Times” would eventually become one of the most sampled songs in the world as well.
And whilst claims certainly could be made - back in the day - that sampling wasn’t real musicianship, and that therefore rap and hip hop wasn’t art… It would be very difficult to convincingly make that argument today.
Old world solutions that could still work
The music industry eventually worked out a procedure called “clearance”, which is the process of acquiring permission to use a sample from the original artist in exchange for compensation. And this helped smooth things along immensely.
And I would argue that generative AI is in a similar situation now, as sampling was before it became a regulated practice.
The problem with generative AI, is that the tech-bros in Silicon Valley are nowhere near as willing to collaborate on solutions as the pioneers of disco soul and hip hop were back in 1979. Primarily because they are not so much in the business of making money, as they are in the business of making all the money.
And this is where things get tricky for me, because whilst I absolutely agree that artists are getting utterly shafted as the situation is now, I can also see how much potential AI has as a tool to facilitate the creation of art, and I don’t want to lose that.
So I would really like for the fronts in this conflict to agree on some common ground where we can have both AI tools and artist who are paid fairly for their work.
I would love - if any of my generative AI projects ever go commercial - to be legally required to compensate the artists whose work has invariably been ripped off by the AI tools I’m using. I would love it even more if the companies making the tools were required to pay artists dividends for using their art to create the tools in the first place.
New rules
So as a person who actually likes living in a society based on the concept of a social contract, I believe its high time for some appropriate legislation on the matter.
Now, I am no law maker, but if I was to advise anyone on how to solve this (at least for cases of visual art) through legislation, I think the following set of rules would get us a fair bit of the way there:
1. Commercial AI models must maintain a public database of every individual image it has scraped, and who created them.
This might sound like hard work for the companies producing the AI models, but I’m pretty sure they have this data. How else would they be able to mimic a particular artist’s style so well when prompted for it?
There is also the fact that we know Midjourney probably has a database like this already, due to the list of scraped artists that was leaked last year.
2. Every AI generated image should keep a record in its metadata, of which artists works and how big of a percentage, were used as references for the generated image.
Any commercial usage of the image must also require this metadata to stay intact. This will of course require some cooperation from actors like Corel and Adobe, to make sure that saving images after working on them with art-tools retains this meta-data, but this is still a viable endeavor to pursue.
3. Artists whose work and style is being used in a commercial image should receive a viable royalty payment.
Organizations similar to the ones we already have in the recording industry, should be responsible for collecting royalties from use of these images based on their metadata, and distributing these royalties to the artists.
This will of course also require cooperation from a bunch of commercial actors, and require artists to unionize. In addition, a working framework for the size of royalty payments must be put in place. But this will create new revenue streams for artists, and also jobs for people in the collection and distribution companies.
How practical is this?
Is this going to happen? In an ideal world this would be a no-brainer.
Unfortunately, in the real world there is a general reluctance by lawmakers to regulate the tech sector in any way. So I don’t have high hopes for any good laws passing anytime soon. We are for instance way overdue on regulating social media.
So while we wait for appropriate legislation, I will continue to actively voice my disdain of the current practices of the generative AI industry.
But since I am being pragmatic about it, I am also going to continue experimenting with generative AI, learning from it and incorporating it into my workflow. Because to not do so, is just plain silly at this point. The genie is already completely out of the bottle, and it’s unlikely to go back in again.
To ignore the potential of generative AI, would be like textile workers during the industrial revolution refusing to use sewing machines. It would just put you at too great of a disadvantage.
Generative AI is an immensely powerful tool that is already here and available for everyone, and the race is currently on to see who can harness it. And the people who have the best potential to harness it, are the ones most adept at making art: Artists.
It is however, important to be upfront about it when using generative AI, and not pretend that you are in fact not. As such, I have and will continue to mark all my own images that incorporates generative AI with the following signature:
As for my AI experiments, I will sometimes post about them here if I think they are worth sharing.
At the same time, I will of course continue building my actual art skills in the old fashioned way. Unfortunately, most art-portfolio sites and social media has already been saturated with works of generative AI, so it’s unlikely to be noticed.
I have therefore started putting my “organic” art up on Cara. Which is an art-sharing site where there are strict rules against posting AI generated images, and where the goal is to create a community of real artists and art.
And like Substack, it’s still relatively new and therefore still fun and exciting. So let’s all hope that it manages to retain that “new website smell” for a while, even after the finance people eventually come sniffing around looking for investment opportunities that they can invariably “monetize” and then run into the ground.