D&D General DMs Guild and DriveThruRPG ban AI written works, requires labels for AI generated art

[Maxperson]

If I understood @Art Waring post above, the issue is that the imaged were "licensed" for non-commercial purposes and then used for commercial purposes anyway, without an opportunity for copyright holders to opt out. Given the methodology of training midjourney (it is the only one I am familiar with, and I am certainly no expert) this constitutes a use of copyrighted material without compensation. I'm not sure whether that is a crime or civil violation or what, but it certainly isn't ethical.

I'm not sure learning from looking at an image and learning from what you see is really "use" of that image. That goes way beyond my paygrade.

 

[Vaalingrade]

And while one half-baked tech is driven into the ground by tech bros blinded by greed and zero patience to wait for its maturity before it's actually viable, real science is quietly making room temperature super conductors a reality.

I can't wait until they buzzword that into uselessness.

 

[Charlaquin]

I keep seeing this claimed and I am trying to figure out how this is theft and not say copyright infringement?

Legally speaking, copyright infringement may be a more accurate term. Colloquialy, the word theft is used to describe all kinds of unrightful use of others’ property (whether actual or intellectual), including copyright infringement.

 

[Reynard]

I'm not sure learning from looking at an image and learning from what you see is really "use" of that image. That goes way beyond my paygrade.

But the people who created midjourney weren't looking at something and learning, and neither was their algorithm -- because their algorithm isn't a mind or entity. They programmed a tool for which they were charging a fee using references that were not theirs to use. We can't get around that by pretending midjourney is a person, even if the way it "learns" is (very) superficially similar to the way a person might.

 

[Art Waring]

What you describe here doesn't meet the definition of theft.

Well, I guess this is an agree to disagree, as we both have our own perspectives on the matter.

For me the primary question I pose is what should be done with ai-tools that seek to use non-commercial datasets like LAION to create content for commercial purposes.

I think that ethically, giving credit & some form of compensation for use of an artists work should be addressed.

2. There's alot of 'free' art out there. Categorizing which is which might be an immediate problem, but it won't be for long. As long as there is sufficient 'free' art then you don't have to worry about the degradation those articles mention.

And that's totally cool by me. Anything in the public domain should be available for anyone to use. Training AI on data solely in the public domain however, has yet to be a precedent. This argument grows particularly weak when you see that the highest amount of prompts use a living artists name (like Gred Rutkowsky), rather than using the name of an artist who is no longer alive and their work is in the public domain.

(I addressed point 1 in my prev post).

Yea, it's quite a conundrum. How do you get people to hire you without showing your work and how do you show your work without having it potentially copied.

IMO. In the past the risk of having your works copied was relatively low. Even if someone did so it likely didn't impact any potential jobs you might get. But AI is different and even if it doesn't copy your art, it can still disrupt your means of making a living by potentially being vastly cheaper, supposing it got enough free or cheap art elsewhere.

At the end of the day though - the real issue with ai isn't it copying your art - it's ai putting you out of business - and it can do that without ever needing your art.

I am trying to deal with this challenge ATM, so I really don't have any solutions. Indeed its not just ai putting artists out of work, but the industry itself. That's part of the reason why I am steering away from freelance work and using my skills in the future as an independent publisher (you won't find my portfolio on Artstation).

 

[Maxperson]

But the people who created midjourney weren't looking at something and learning, and neither was their algorithm -- because their algorithm isn't a mind or entity. They programmed a tool for which they were charging a fee using references that were not theirs to use. We can't get around that by pretending midjourney is a person, even if the way it "learns" is (very) superficially similar to the way a person might.

What is learning? I'd argue that when I learn something I'm just using references that aren't mine to inform me on how to do something. Painters learn by looking at what other artists do(references) and then use those to inform them on how to do something, say copy a style of another artist. Just because the tool isn't organic or alive doesn't necessarily mean that it isn't learning.

As I said, this is all way above my paygrade. It will be interesting to see how it plays out in the courts and legislatures.

 

[Thomas Shey]

No. I can create brand new things. Let the ink flow. Not based on anything anyone else has ever done. That’s literally what creativity means. And AI can’t do it.

When it comes to visual art, demonstrate that the original work most people will do is distinguishable from the recombinant work an AI can do. All evidence I have to date is you can't.

I really do get the fact that people would seriously like not to have their creative work displaced by automation. But magical thinking, and I see a lot of what seems to be that in these discussions, is not going to do it. That's as true about distinguishing "true" creativity from recombinant work as it is about there being some distinction between what an artist does when he learns to draw by looking at a lot of other artists and what a evolutionary program does when its feeding in from other artists and outputting things that are new is.

Its a set of distinctions I can understand why creators want to be significant, but I'm not sold that makes them significant.

 

[Thomas Shey]

I mean, if you're just after a generic medieval village scene, odds are good AI can (now or in the near future) bang something out for you that'll do well enough. But if you're looking for a detailed image that incorporates specific elements in exact locations to match with the adventure module you're writing, then AI will let you down big time; and paint on canvas is a whole lot faster of a manual image-creation process than is pixel-by-pixel on a computer.

Though I'll note you have to qualify that is it will let you down if working by itself. There's a lot of capability for a user to set up where they want specific things in locations (or even just tell the program to put them there) and let the software fill in the rest. For some uses that's probably inadequate, but for others, its just the program doing things that otherwise would be busywork for the user anyway (in a lot of cases I may know roughly where the mounts should be, but I don't feel a need to know where every range and peak is).

 

[Reynard]

I don't think it is useful to argue about what is "real" creativity and what is just regurgitation, because we don't make those distinctions in the commercial arts beyond a) is it any good, and b) does it violate copyright. There is a huge amount of derivative work in the arts -- I would argue that MOST of the material put out by people is blatantly derivative. So it doesn't actually matter.

All that matters here is are we going to require corporations intending to use this technology for their own enrichment compensate the artists on whose works they train their algorithms? I think we must. Let artists opt in or out on their own terms, and let companies utilize existing public domain works within the scope of the law.

Whether we want an endless stream of AI generated and minimally edited material to flood every available market is a different question.

 

[Reynard]

Though I'll note you have to qualify that is it will let you down if working by itself. There's a lot of capability for a user to set up where they want specific things in locations (or even just tell the program to put them there) and let the software fill in the rest. For some uses that's probably inadequate, but for others, its just the program doing things that otherwise would be busywork for the user anyway (in a lot of cases I may know roughly where the mounts should be, but I don't feel a need to know where every range and peak is).

I don't know if I dreamed it but I thought I saw an ad for a background generator. Like, you put a person or whatever in the foreground and then tell the algorithm to create a "city nightlife" backdrop or whatever.