On generative ai

I was planning to write this post a bit later, once both law and my own views had settled more, but a self-appointed art policeperson forced my hand. I don’t believe in answering strangers’ probing questions online, but this is an important topic. So here you have them, my half-baked thoughts on the topic of generative ai.

Many folks have strong opinions about LLM generators. You know, the thing popularly called “ai art” – though I pedantically quibble that it’s neither intelligent nor art per se (it’s a mindless tool).

I believe the luddite approach—to proudly deny one could or would ever use these technologies in any way is doomed to failure. Whether or not such an ethical absolutist stance is correct, the economics of this technology will have its way.

And let’s be clear, the economics are astonishing. Textual ai tools can serve as powerful editors, reorder lists alphabetically, suggest processes, provide a sounding board for ideas, and more. Visual ai tools allow quick prototyping, testing compositions, generating character concepts, backgrounds, mood boards, expanding backgrounds, removing artifacts, and more. At the very least, they’re the equivalent of a free assistant.

Personally, I also think the new possibilities for generating surprising, unexpected, and never before seen assets, are quite astounding. The technology opens up the prospect of millions of people without the skill or ability or time to wield a brush to see their ideas visualized. It gives a person with a rudimentary grasp of a language the chance to turn their bullet points into poetry. I think this is wonderful.

However, I don’t think the assets generated by these programs are, in and of themselves, art. They may be images or texts or what have you. But, they are not art. To me, art requires sentient aesthetic creative action (I say sentient rather than human, because I do consider some examples of works by other animals and hominids to also be art and keep open the possibility of artificial sentience in the future being able to create art).

As such, I don’t think these assets can reasonably be either protected by copyright nor have much values as works on their own.

In all this, my position broadly accords with that of the United States Copyright Office* and the recent precedent-setting federal case (Thaler v. Perlmutter) that AI cannot hold copyright.

That said, the assets produced by these programs can be used to create art. From the very crude (a midjourney picture of a religious leader in a balenciaga jacket + a meme font) to the more sophisticated (composition tests, variations on an artists’ own existing works, collages materials, etc.). After all, du Champ could turn a urinal into a fountain. However, du Champ is also instructive: though one fountain became a urinal, the myriad other millions of urinals have remained urinals.

So, for example, one could assemble visuals generated by AI plus texts generated by AI into a cohesive book and own the rights to that book as an assembly of art and texts that tell a story as a collection in the same way they could own the rights to leaves picked up off the ground and clippings from a 200-year-old newspaper glued into a scrapbook to tell a story. However, neither the images nor the texts would be covered by copyright – that is, someone could copy such content verbatim so long as they did not assemble it in a largely identical manner.

Fountain by Marcel Duchamp. Source: Wikipedia.
Marcel_Duchamp’s Fountain photographed by Alfred Stieglitz. Photo from Wikipedia.

Basically, the more human effort and creativity one applied to rework and refashion assets created by AI into something different and more, the stronger their claim to have made art would be.

In this, again, I am following pretty basic copyright law on compilations and minimum requirements. From the Berne convention (esp. art. 2.7.) onwards (and yes, also the US copyright office compendium on practices, esp. section 309).

On a personal note, I think the ethical thing to do is for people who use ai generated visuals as their finished product, to say so. Further, if they’re using ai to specifically copy another artist? That’s downright scammy if they’re doing it without crediting the original.

On the other hand, I don’t think ai generated visuals used as part of story boarding, generating ideas and concepts, trying out lighting and compositions, are very different from an artist leafing through a magazine or trawling pinterest, and hardly worth mentioning.

Come on Luka, easy words, would you buy “art” made by a human artist using AI generated assets?

If I thought it was good and well-made and surprising and I enjoyed what represented? Yeah. I would.

But, that’s honestly a high bar.

So, would you buy a poster of a picture generated by AI? No text, nothing else. Just a prompted piece?

If it really spoke to me and seeing it was one of the most visceral experiences I’ve ever had and it was cheaper than just taking the file printing it myself? Yeah, I suppose.

Gotta keep an open mind, you know. Might get surprised. Who knows – the future is an alien country.

Ok, Luka, would you use AI art in the works you sell?

I think ai generative tools can be excellent tools when used correctly. They can boost productivity and creativity both, from help with editing to use a sounding board for testing ideas. I also think ai generators can be great fun as simple play, with no end goal in mind. They’re quite excellent on the fly when you’re running a roleplaying game on discord and need a quick face to pin to a character.

However, I take immense pride in the craft and personal creativity I put into my published work and I wouldn’t put my name to words that I had not personally typed nor to art that I had not personally drawn. The layout I might outsource to a better graphic designer at some point.

I’ve never, nor do I intend to, sell works created by other entities – whether human or machine – under my own name.

Luka, why wouldn’t you just say this when I demanded you say so on [social medium]?

It’s tempting in the age of social media and parasocial relationships to simply demand answers from strangers. This is a temptation that we would all do well to resist.

*Let’s leave aside my willingness to grant copyright to any sentient organism. Likewise, my pedantic quibbles on definitions of art and intelligence. All those are currently and practically irrelevant. Pace.

Usually I’d have a small paragraph here about joining my patreon, but this time I’m just going to invite you to make art with any and every tool you can find. Go on. Give your sentience free creative reign.

8 replies on “On generative ai”

FORBES: Did you seek consent from living artists or work still under copyright?
David Holz: No. There isn’t really a way to get a hundred million images and know where they’re coming from. It would be cool if images had metadata embedded in them about the copyright owner or something. But that’s not a thing; there’s not a registry. There’s no way to find a picture on the Internet, and then automatically trace it to an owner and then have any way of doing anything to authenticate it.

Yeah, I’ll just leave this here. No one cares Luka. They just don’t. Discussing “ethics” is a waste of time.

I politely disagree. I think a lot of people care very much about ethics, but it’s a complicated issue with little precedent, where values, interests, and points of view collide. As a result, people end up with very different stances.

I have experience with trying to set up contracts with other writers and artists, that would automatically share royalties ensuring everyone has ongoing revenues from their work. DTRPG has a surprisingly good, automatic system, where you can set a royalty split. Itch.io does not. Let’s not even start a discussion about setting it up with other platforms or physical retailers. The end result is that even if I ethically want to set up a system that shares royalties to everyone, doing so is beyond my (and most other publishers’!) abilities: if I set things up the way I wanted to, I would soon (after about 5 published multi-author titles), end up spending half of my time just doing accounting (I exaggerate, but since I’m not an accountant it would take a lot of time and energy). In the end, that was part of the reason I chose to stick to sole authorship and came to grudgingly accept the practical reason for the work-for-hire or the one-and-done global licensing fee model in publishing.

David Holz, in your excerpt, says two things: 1) the practical difficulty (effectively, impossibility) of seeking consent from artists for pictures available for free on the internet and 2) that they did not seek [individual] consent from living artists. He does not actually talk about about ethics or not in this excerpt, so we are only guessing at the ethical background or ellipsis that connects these two statements. He may:

1) think it is impossible, therefore unnecessary.
2) think it is regrettably impossible.
3) thinks it is unnecessary, and the difficulty is irrelevant.
4) thinks it would be good, but the balance of practical impossibility and the legally weak position of artists, makes ignoring this ethical precept an acceptable risk.
5) thinks it is ethically unclear if it is necessary or not in the absence of regulatory precedent and therefore permissible until explicitly prevented by law.

But, overall, Holz’s position is not very important right now. There are multiple organizations and individuals working on visual generative ai, and they may all have different ethical and pragmatic positions. People in favor and against these tools may also all have different positions. For example, even an artist arguing against these tools is not doing so solely from a disinterested ethical position, after all – these tools may well be a threat to their livelihood (they have already eliminated much of the $5-a-piece art market, that’s for sure).

In the end, this is a new technology, and its use will end up regulated by courts, law, and custom. Ethics will certainly impact how it is used, therefore, for people with a strong interest in specific outcomes, discussing ethics is a very important task.

Thanks for this piece – I wouldn’t call it half-baked by any means. Having just exited many similar conversations this week that touched on similar concepts and then finding this directly afterwards was interesting – I thoroughly enjoyed reading it as a nuanced and balanced take that I happen to find myself largely aligned with.

Thank you for the comment 🙂 I’m glad you enjoyed my piece!

My biggest issue with how this, ah, generation of AI tools is the training data acquisition. Generative AI tools are amazing, there’s much more legitimate value from them than any of the crypto (and related) nonsense of the last several years, anyway.

I know it was mentioned in a previous comment, but just because it is hard (very hard!) does not make an action allowable, in most societies. And so just because the internet makes it hard to determine the origin of art, does not make it allowable to use it unfairly, in my opinion (and this is a point many people on both sides gloss over, as “Fair Use” is a common intellectual property concept in most countries). I just don’t think mass processing of art to update algorithms is any more fair use than taking the sales contact list or demographic data of a company to use as your own (which is essentially how most social media companies actually make money).

Every gold rush is unfortunately bloody, as you seem to be implying, and I agree. That said, every gold rush’s damage is felt long after the mines are closed, and the towns turn to haunting grounds. In the USA, we are still dealing with the ramifications of settling and developing the nation; people are still caused undue suffering (in small ways, and also sometimes big). The opportunity here is that we are able to minimize or mitigate the damage here, which is quite likely a serious collapse in the low and middle end of the art economy, and the animosity that that will engender in those artists (and rightfully so!). Which I am sure will have knock on effects for a long time to come. We should make the companies do the right thing on training data, or force them to take down their models.

I may seem harsh, but these AI and AI-leveraging companies are already using secretive and underhanded ways to gain training data. For instance, the MEE6 discord bot (so not even discord, but a bot!) is almost certainly harvesting discord conversations; turns out they added a GPT-powered chat or to the bot. If you @ it talks in clear AI prompts. Now since this bot is used for leaderboards based on user activity, it is highly likely they scan messages to train their model at the same time. And this feature was only noticed when interacting with the bot, and then going to their site. Who knows when it was enabled? I quite that server, because even if my content is valued at small fractions of pennies (I have no illusions about my prose), I was not directly asked, and I was being taken advantage of.

So I say, go forth! Train your models! I am one of those artistically hopeless (in the visual arts, at least) types who could get a lot of benefit. But also as someone who admires people skilled in the arts, make damn sure their permission is gained. If it is hard, make it easy for customers by developing a tool artists can submit art that they are okay having you train with. Like the AI guy quoted in the other comment, he already knows of a solution, and it can’t be harder to develop an “Ethically Sourced Art” registry than, you know, developing a working LLM. Especially with all the investment money going into these AI companies right now.

Agreed – it’s a jungle. That said, it’s all rolling more and more into the mainstream. Generative AI is now bundled with Photoshop. Honestly, they moved faster than I expected.

Finally a reasonable take on AI. It is so difficult on the internet to impart the idea that AI only puts out what you put in, that ultimately its creativity and quality of output are dependent on the user. Just look at Corridor Digitals AI animation! They had to put hours upon hours of effort just to get it looking half decent to achieve their desired effect, and that is just the technical parts of that project.

I’ve respected and enjoyed the art & worlds you’ve created long before any of these nascent technologies were an accessible tool for creators, or even a popular talking point. So I respect that you recognize the spectrum of use cases — and importantly, the open-ended potential to expand expression for people who might not be gifted with the talent or privilege to readily realize their imagination into visual media.
I think you also hit on a major point that is often missing from the discourse as we try to make sense of this rapidly developing digital culture, which is that all these platforms we already publish our works on have a wide array of intentions, functionality, and compensation models. They range from benevolent & convenient for collaboration, to benign but unwieldy, or outright predatory. Or some weird permutation of any of those qualities.
People shout beautiful and/or terrible things out into the void of the internet constantly. Sometimes it’s a public work free for anyone to consume, and sometimes you should take care to seek compensation for all the energy you poured into your thing.
This has always been the case though, long before digital connectivity. We’ve always continually expanded on the collective consciousness to make weird new art, and heap new ideas onto the pile that someone else started. If anything, we are better equipped now to share, debate, and support each other than ever before.
LLMs and new ways to create won’t upend the human qualities of art and culture overnight, but the time we are afforded to adapt comfortably and make sense of the loop is exponentially tightening.
Nerds, scientists, artists, venture capitalists and governments aren’t going to stop developing the technology, it’s already happening.
We get to use it however we choose though, just like we always have with the little worlds we build and inhabit with our art.
All I know for certain though, is that shared fiction is cool.

Comments are closed.