WotC WotC: 'Artists Must Refrain From Using AI Art Generation'

After it was revealed this week that one of the artists for Bigby Presents: Glory of the Giants used artificial intelligence as part of their process when creating some of the book's images, Wizards of the Coast has made a short statement via the D&D Beyond Twitter (X?) account.

The statement is in image format, so I've transcribed it below.

Today we became aware that an artist used AI to create artwork for the upcoming book, Bigby Presents: Glory of the Giants. We have worked with this artist since 2014 and he's put years of work into book we all love. While we weren't aware of the artist's choice to use AI in the creation process for these commissioned pieces, we have discussed with him, and he will not use AI for Wizards' work moving forward. We are revising our process and updating our artist guidelines to make clear that artists must refrain from using AI art generation as part of their art creation process for developing D&D art.

-Wizards of the Coast​

Ilya Shkipin, the artist in question, talked about AI's part in his process during the week, but has since deleted those posts.

There is recent controversy on whether these illustrations I made were ai generated. AI was used in the process to generate certain details or polish and editing. To shine some light on the process I'm attaching earlier versions of the illustrations before ai had been applied to enhance details. As you can see a lot of painted elements were enhanced with ai rather than generated from ground up.

-Ilya Shlipin​

​

 

[FrogReaver]

....as pithy as that saying is, I think that the fact that the AI found that these were the salient factors quite notable.

To give you an easier-to-grasp example, there is a reason that orchestras went to "blind auditions" (performance behind a curtain).

That the AI determined that this is what was most important ... should likely make us reconsider our own processes.

IMO it’s not what’s most important. It’s just correlated with what is.

 

[Umbran]

I think that’s just a case of correlation not equating to causation.

The first is a case of correlation not equating to causation. But, we know why the training set had rulers in lots/all of the pictures.

The second case should make us wonder why the training set had that correlation.

 

[FrogReaver]

The second case should make us wonder why the training set had that correlation.

Seems trivially explainable.

Jared at the time was a popular male name and especially so amongst relatively wealthier families. Lacrosse also is an indication of family wealth as it’s not a sport typically played by less wealthy class members.

Wealth in general correlates to better education, better health, better networking connections and better relatability to other wealthier people - like their would be coworkers and bosses.

We can talk about whether that is right/fair/just/etc - but it’s not a particularly surprising correlation.

 

[Jfdlsjfd]

Just because its not mandatory doesn't mean its legit to use for commercial purposes (thus why the law currently states that AI content cannot be copyrighted). Laws are in the wild west phase regarding datasets and training models so we don't actually know if this argument will hold up in court.

I don't think it's law, it's just a US-only court decision, as far as I know, lawmakers didn't explicitely looked into that (and with regard to US laws, I don't claim to have any more knowledge than what is indicated on these threads). However, this is a decision that affects another step, step four.

Irrespective of what happen in step one to three, the outcome of step four can't be copyrighted. It doesn't mean that it can't be used commercially. I can sell anything that isn't copyrighted, either because it was created before copyright laws even existed (I can sell you the text of many religious texts from the middle ages), because their copyright lapsed (I can sell you the text of plays from the early 19th century) or because their weren't copyrightable in the first place (I can sell you what a monkey types on a typewriter). You wouldn't have interest to buy it, of course, instead of acquiring this content from another, cheaper source (like project gutenberg for texts), but it doesn't make it illegal to sell things that aren't falling under the umbrella of copyright.

Their words, not mine.

Sounds like a company's standard disclaimer.

Why are datasets and models for music entirely copyright-free and voluntary, when ai-image generator datasets and models don't have this requirement?

Because it's not a requirement. It's just a happenstance that Stability had a copyright-free stock of music (old music records)? Also, generating randomly a few seconds of recognizable existing music is certainly more statistically possible than generating a existing image. If I sing a few seconds of a music of an easily recognized score (one can recognize Twinkle Twinkle Little Star with 14 notes easily, and the odd of generating it are higher than a graphic generator generating a recognizable part of the Mona Lisa, barring overtraining on Mona Lisas.

 

[Umbran]

Why are datasets and models for music entirely copyright-free and voluntary

Um... they aren't. You have one example in which the dataset is copyright-free and voluntary. One example does not support a generalization that for music datasets and models are so, or will be so going forward.

 

[Art Waring]

Um... they aren't. You have one example in which the dataset is copyright-free and voluntary. One example does not support a generalization that for music datasets and models are so, or will be so going forward.

Oops! Apologies, I meant in regards to stability ai models, I will edit this mistake.

 

[Snarf Zagyg]

Seems trivially explainable.

Jared at the time was a popular male name and especially so amongst relatively wealthier families. Lacrosse also is an indication of family wealth as it’s not a sport typically played by less wealthy class members.

Wealth in general correlates to better education, better health, better networking connections and better relatability to other wealthier people - like their would be coworkers and bosses.

We can talk about whether that is right/fair/just/etc - but it’s not a particularly surprising correlation.

Actually, that is an incredibly surprising result, despite your post hoc rationalization.

Look, if the AI had centered on, say, "Ivy league degrees," we might say that this is unfair, but understandable.

But if you went back to the people who actually made the decisions that the data set was based on, and told them- "Hey, I know you think you're being fair. But do you know what the two most important factors are? Whether the applicant is named Jared, and whether they were a Lax Brah ...." I think that they would probably be more than a little surprised. Again, this isn't even college lacrosse ... these are people that listed high school lacrosse on their resumes.

In other words, it did uncover something rather surprising. Which should lead to some self-reflection. Whether you think it should lead to self-reflection or not ... well, I can't control what you choose to do, right?

 

[Umbran]

We can talk about whether that is right/fair/just/etc - but it’s not a particularly surprising correlation.

I don't feel "surprising" is the real takeaway here. He and I have said that it is something that shows we should think about our own processes more. This holds whether or not the result is surprising.

 

[Jfdlsjfd]

Seems trivially explainable.

Jared at the time was a popular male name and especially so amongst relatively wealthier families. Lacrosse also is an indication of family wealth as it’s not a sport typically played by less wealthy class members.

Indeed. I didn't know the social status associated with the name Jared or playing lacrosse in the US, but it's unfortunately easy to guess why the training sample did inclure more of them, than say, people named from a disenfranchized social groups.

It's the same bia found in prompt-expressing: if you're asking the generator to draw a CEO, most of the images will show a white male, wearing a suit. It might not even be a bia, but a reflection of the current situation (on which the AI is trained).

On the other hand, it's also possible that's it just a cause of two things being correlated without any causation link. Tending to mention lacrosse on their resume is just a trait that wasn't selected by the human, but it tells us things about being in the mindset of mentionning high-school lacrosse on their resume.

For example, we as recruiters might exhibit a bia toward hiring younger persons, and those are more often (it's an anecdotal observation, not a fact) inclined to pad their resume with high school experience (because they don't want to give a half-empty page to the recruiter).

 

[FrogReaver]

I don't feel "surprising" is the real takeaway here. He and I have said that it is something that shows we should think about our own processes more. This holds whether or not the result is surprising.

I agree surprising isn’t the key takeaway here, nor was it the key take away from my post (it was one word near the very end of my post). Your challenge to explain the correlation and my explanation were the key takeaways. IMO.