The Ethical Generation Of AI Art

There's been a lot of discourse in the TTRPG space as regards AI art, and I've spent a lot of time thrashing out arguments about it's presence and use in that space.  Eventually I decided to compile my bits and blog it out so each thought can find it's proper length.  So here's that.

(This piece has been updated.  Updates are in dark red.)

...

First thing, I want to settle my basic ground on some stuff:

1. Nobody is obligated to hire or not hire artists.  Single instances of the "stole a job" argument are nonsense in both directions; it's not a strong point to say that it did or a defense to say that it didn't.  If you're arguing on the basis of Bob Hiring An Artist Or Not, you're not on useful ground; Bob gets to hire who he likes or not.  However....

2. Everyone is to some extent responsible for the systems they patronize, fund, and enable, and their wider impact on how things work.  If AI art pushes an local/sub/whatever economy in the aggregate that centralizes money in the hands of tech company owners who do not pay living, working artists, while drawing on the labor of those same living, working artists, that IS a problem - and Bob, if publishing, is a contributor to normalizing that problem.

3. Referring to a shift to AI art with metaphors regarding industrialization may very well be apt, but doesn't actually tell us if it's good or not.  Kicking your clogs into the machine can legitimately be the ethical decision, and "someone's going to be the villain, may as well be me" is a negative ethical stance, not an acceptance of a new neutral.

4. Having an AI sample from artwork appears to be legal, so long as it doesn't breach terms of service for hosting sites or the like (which...   scrapers have done, so put a pin in that), but if it reproduces something especially close to a trademark or copyrighted piece, that could easily be infringement.  While we should properly be concerned firstly with whether it's ethical, so far as legality goes, the answer is that it's not fully settled, at least not so far as stock art sites are concerned:



5. Following up from the infinite digital arguments of the past on Theft vs. Piracy, it's probably worth getting the term of accusation correct - what's the specific bad thing AI art generation is generally being accused of?  In most cases, the terms are almost certainly remix and plagiarism, so that's what I'll be using later on.

This leaves us with two big questions:

Will AI art worsen our local/etc economy as in (2), and should it be avoided on that basis?

Are AI art programs basically plagiarism or unethical remix engines, and should they be avoided on that basis?

So let me bash on those separately:


Our Local Economy

We don't know with any certainty what the scale of the impact is here because the reactions we've seen are people enthusiastically jumping on it and sharing it around, and people reacting to that, and a lot of shouting.  We don't know if buyers of games will buy things made with this art as a novelty, as a regularity, if they'll get sick of it quickly.  We don't know if it will improve and do better on things like eyes and hands and cross the uncanny valley forever, become indistinguishable from other art, all of that.  There are a lot of sellers of AI stock art suddenly, and will likely be a rush of things with AI art in them, yes, but scale also requires a lot of buyers, or it's just a failed fad people wasted time on.

I'm also unsure of the tone of the impact economically.  If the people using it are, in the aggregate, using it to replace free or public domain art, it's not doing a damn thing to the local artists.  If they are, in the aggregate, using it to replace stock art, then it's a loss to to stock artists. If they're most often replacing their own efforts, then economically it's just a labor-saver (this is me, by the way; the "artist I would be replacing" is also me).  If it's taking the place of commissioned art, again, it's a loss to commissioned artists.

I suspect that it's going to be a loss to stock artists, after a lot of shaking and tinkering and etc etc etc, but not nearly so much so to commissioned artists.  And even this is somewhat in question, given the very recent (as of this update) rejection of AI art by a number of Stock Art sales sites.  Still, though, for many designers, generating their own good-enough AI art is often cheaper and faster than finding a good-enough piece on stock sites.

If the art improves technically and is normalized to an extent that it creeps up the ladder of where it's accepted, it may bit by bit edge out commissioned art.  Of course, many artists work both sides of that fence, even using stock art to attract commissioned clients, so to them it doesn't matter if it's a load of bricks or a load of rocks that looks like it might be hanging above them.

On the other side, the writers are also often broke, and AI may prove to be an economic gain to them; getting to do their thing at the level and in styles many consumers demand even though they're not walking in with significant capital.  

It may be a win-lose deal, such that eventually what we get is that the writers are less often broke but there aren't half as many artists working in TTRPGs...   And I would say that as an end state that's ultimately neutral or even maybe mildly positive (everyone still here is making, on average, more money) economically, but the path from here to there is littered with vanished opportunities for artists or (if the scale is large enough) even dead careers - a potentially grim road to a very mid-range end.  

Now, if it enables independent designers to compete in such a way that it lowers barriers to entry and make the kind of money that lets them hire artists, and they do so, then it's a gain for both.  But here's where that creep up the ladder might fit, driven by "Well, I could hire an artist now, or run ads?  I'll run the ads." and similar line-item decisions where the projects get better but the artists stay out in the cold. 

Ultimately, to me, this all reads as a reason to be wary overall, but it's all shaky, filled with maybe-if and maybe-not, and as a result it doesn't set a lot of bright, rigid lines about acceptability or action.  It does draw one, as an absolute, and it goes like this: 

Even on purely local-economic grounds, it's definitely good to frame AI Art as being, at best, for experimental, starting-out, and no-budget work, and it's definitely good to sneer loudly at projects that can afford commissioned art that are using AI Art instead.  No, they don't have an obligation to hire any set way, but we probably should treat them as if they did, so that usage "up the ladder" won't be normalized.  


Remix Engines

At this point, it's common to introduce some analogy to photobashing or collage or DJs, and I've done that before, but I want to bite that particular hand for a second.

If Jimmy walks into a room with a copy of Shakespeare's Romeo and Juliet, and walks back out later on with "Jimmy's Great Romance", which is plainly a R&J with minor changes, we don't need to give a shit what happened in the room.  If Jimmy has a thousand monkeys at typewriters in there, and he merely selected the closest matches to his book and then sent those around for monkey-editing a hundred times, he has invented a really astonishing way to plagiarize Romeo and Juliet, but he's still doing that.

It doesn't matter if the AI cuts them into confetti or vectors or alien algorithmic patterns Which No Man Can Know.  If it later reproduces a work it disassembled to a close enough extent to be a minor reskin, we can reasonably call it plagiarism.  If it often reproduces works somewhat, enough that it's notable but clearly not the same work, we can reasonably call the product a remix, and judge it by standards common to other sorts of remix art.  If the influence exists only really at the level of influence, we can call it "inspired by", and not worry about it much.  And yes, these are subjective levels, and always will be.  So it goes.

As an aside, at this point, there's often someone in arguments that says "AI can't possibly reproduce a piece closely enough that it'd be plagiaristic!  That's not a thing, because [discussion of the monkeys in the room]".  To which I can only say, here's what happens when I type "Van Gogh Starry Night" into Midjourney:


Everything I have seen to date leads me to believe that AI Art programs fundamentally act as remix engines, which on (at least seemingly) rare occasions pull too much from a given source and push over into the realm of plagiarism, but are likely sometimes are original enough by virtue of randomness to claim that no influence rises above being an inspiration.  The designers would like them to reach the level of pure inspiration, and their most ardent fans claim we've already arrived, but honestly it doesn't seem like we're there yet.  Moreover, when such a program happens over into plagiaristic territory, it doesn't know that, and doesn't pass on enough information to the user for them to know it.

So, this means that most AI Art has three major ethical issues in general use, which when they're taken together make it at least "kinda shady" and at worst "straight bad".

  1. Users may arrive at plagiarism incidentally.
  2. Only willing or public domain sources should be used at the level of remix (being inspired by others is inevitable in art, and AI can be allowed that standard, I'd think). 
  3. When remix is done, at least the primary sources of the mix should typically be credited, and this isn't built-in.

AI art programmers can work on these (and bloody well should, and in some cases definitely are or aren't) by limiting their datasets to willing and public domain artists, working to get their thousand-monkey room to "grind finer" toward inspiration, and ideally to toss up a credit when a major source exists.

But users can also make significant strides towards or against these things.  Take a moment to read about Greg Rutkowski, then come back.

...Okay, so, that's not cool, obviously.   Right?  That's users actively calling an artist in as a major source on the remix - an artist who feels threatened by what's being done with that remixing, and then drowning that artists search results with bad crediting.

But then, on the flip side, consider the following piece, which I quite like (you might hate it, but hey, taste):


That was generated in Midjourney by feeding it the address of a photo, a subject, and an artist for style, specifically:

/imagine prompt: https://www.publicdomainpictures.net/pictures/450000/velka/george-town-clock-tower.jpg clocktower in stained glass by Edmund Dulac --iw 1 --ar 11:17

Which means that it's a remix whose primary sources are Alix Lee (the photographer who took this picture and put it in the public domain) and Edmund Dulac (a classical illustrator in the public domain).  I know who they are; I know I'm not in plagiarism land, my main sources are public domain, and I can credit them.

All of which goes into my view, which is that on a personal level, AI Art is ethically much like a power tool with deeply insufficient safety guards.  It can be used ethically, but it requires an active effort to do so, and we need people to actually be putting forth some effort towards doing so and improving it along those lines, users and programmers both.