Zompist Bboard Again

Posted: **Sun Aug 28, 2022 7:54 pm**

I seem to have forgotten how this forum thing is supposed to work.

zompist wrote: ↑Sun Aug 28, 2022 6:44 pm Like most every technology, it could be used for good instead, as a tool for artists rather than against them. Just like modern illustration and 3-D modeling tools allow more people and not just AAA studios to create great video games, AI imagery could allow artists to do even more amazing things.

This is a valid point, and one worth considering (alongside punishing the "Techbros" for stealing in the first place). What jumps immediately to mind (something I had considered before the revelation of how the AI is trained) is using AI to produce something approaching a reference for characters in a work of fiction, which one could then provide as a reference while paying an artist to produce illustrations for a finished product. I do understand references to be enormously useful things.

Posted: **Sun Aug 28, 2022 8:15 pm**

Again, there is no simple answer to "how an AI is trained". For example, there is one technique where the AI asks for input from human experts for a relatively small fraction of the data to keep it on track, while doing its own thing elsewhere.

You could (pre-)train a model on a corpus of classics to let it recognize the basics of the medium, and pay a number of human critics to evaluate some of what's learned over the internet. You could produce an understudy AI that way. I don't know if there's any money in that market.

Similarly, the data could come from anywhere as long as you have a lot of it. Eg. AlphaZero trained by playing millions of games against itself. It didn't look at any human games.

PS. Also, who stole what, exactly?

Posted: **Sun Aug 28, 2022 8:51 pm**

Every so often someone claims that AI will take jobs away from programmers such as myself. However, the thing to remember, is that someone needs to program and train said AI's, and in the end "AI" will end up just becoming a more high-level form of programming. Also, programs have to be exactly correct and not merely approximate to function, and this is not something traditional machine learning is good for (and machine learning is much of what we have seen flourish in today's "AI spring"). Rather, for AI to succeed at programming, it needs to be rooted in symbolic logic and processing, and this is an area where one inevitably runs into "AI" simply becoming higher-level programming as someone will need to give said symbolic logic system input.

Posted: **Sun Aug 28, 2022 8:59 pm**

Computers can already write code (optimizing compilers). What they can't do is understand the client. Although it's debatable whether humans can understand programming clients either, if NLP improves in the future, then maybe a lot of programming will come to look more like Inform 7.

Posted: **Sun Aug 28, 2022 9:16 pm**

Also on the art side, the current AIs might need to be retrained without using copyrighted art or art under some other specific licenses AT THE MINIMUM, depending if some of the bigger art uploading sites decide to take legal action on behalf of their users, and how that ruling turns out. But the basics of that situation is just that neither the artists nor the sites the art was uploaded to were asked permission or compensated for the use as AI fodder, so a lot of visual artists and companies catering to them are pretty angry rn with art AI programmers.

Posted: **Sun Aug 28, 2022 9:32 pm**

There is no respect for copyright laws in the Third World. If that's already happening, then all you can do is police big corporations and make sure they don't use any copyrighted material. I don't know if you will ever get all copyrighted material off the internet.

Posted: **Sun Aug 28, 2022 9:51 pm**

There's far more public interest in policing big corporations, whose violations are likely to have the most painful impact (not an apocalyptic one, but in a way that makes the world for creators just a little bit "shittier"). If I were a well-known author, and some person wrote a piece of fanfiction, I could issue a DMCA takedown. It's easy enough to punch down already. The problem is that it's far too difficult to punch up, whether it be pursuing a tech company, or some big studio whose scriptwriters engage in brazen plagiarism.

Of course, I really don't think corporations should be able to own these sorts of creative properties (I find the notion that they can be legally divorced from the people who create them at least somewhat unnatural) at all (I don't feel I know enough about patent law to comment on that side of intellectual property, however). Logically, however, that which does not possess an intellect would not have any intellectual properties (if the pun may be pardoned).

Posted: **Mon Aug 29, 2022 2:32 am**

rotting bones wrote: ↑Sun Aug 28, 2022 9:32 pm I don't know if you will ever get all copyrighted material off the internet.

No, but one might be able to put it all behind a paywall or mandatory advertising.

Posted: **Mon Aug 29, 2022 3:05 am**

Richard W wrote: ↑Mon Aug 29, 2022 2:32 am
rotting bones wrote: ↑Sun Aug 28, 2022 9:32 pm I don't know if you will ever get all copyrighted material off the internet.
No, but one might be able to put it all behind a paywall or mandatory advertising.

Um, y'all know how copyright works? Everything on the web is copyright. Everything on a writer's manuscript sitting in a drawer is copyright. Doesn't have to be owned by a megacorporation or published in a book or for sale, doesn't have to be protected by letters from lawyers. It's copyright the moment you make it, and protected by copyright law. This is true both in the US and the EU. All this stuff really is there, no fooling, to protect creators.

As a corollary, if you are a techbro, grabbing art off the web is almost always stealing. There is plenty of public domain art, but things are not public domain because "it's on the web!".

Posted: **Mon Aug 29, 2022 3:46 am**

zompist wrote: ↑Mon Aug 29, 2022 3:05 am
Richard W wrote: ↑Mon Aug 29, 2022 2:32 am
rotting bones wrote: ↑Sun Aug 28, 2022 9:32 pm I don't know if you will ever get all copyrighted material off the internet.
No, but one might be able to put it all behind a paywall or mandatory advertising.
Um, y'all know how copyright works? Everything on the web is copyright.

Not the shape of glyphs in a font, at least in the US. And the US notoriously fails to recognise the perpetual crown copyright in the Authorised Version of the Bible. Texts in WikiSource are usually not subject to copyright - whereas Wikipedia articles are. (However, I think quite a few Wikipedia articles have been pirated off Wikipedia.)[/quote]

Posted: **Mon Aug 29, 2022 7:10 am**

zompist wrote: ↑Mon Aug 29, 2022 3:05 am
Um, y'all know how copyright works? Everything on the web is copyright. Everything on a writer's manuscript sitting in a drawer is copyright. Doesn't have to be owned by a megacorporation or published in a book or for sale, doesn't have to be protected by letters from lawyers. It's copyright the moment you make it, and protected by copyright law. This is true both in the US and the EU. All this stuff really is there, no fooling, to protect creators.

As a corollary, if you are a techbro, grabbing art off the web is almost always stealing. There is plenty of public domain art, but things are not public domain because "it's on the web!".

I think what the Weird Boys Arguing About Trains mean when they say copyright is "unlicensed use of copyrighted material."

Posted: **Mon Aug 29, 2022 8:33 am**

zompist wrote: ↑Sun Aug 28, 2022 6:44 pm FWIW I think the original worry is correct: under our capitalist system, the likelihood is that less money will go to artists who need and deserve it, and who as a class produced the training data, and more money to techbros who don't need or deserve it. It's not the end of the world or the outmoding of humans, it's just a way to make the world a little shittier.

Like most every technology, it could be used for good instead, as a tool for artists rather than against them. Just like modern illustration and 3-D modeling tools allow more people and not just AAA studios to create great video games, AI imagery could allow artists to do even more amazing things.

I agree with both points. We're getting into philosophy, but I don't see DALL-E and other similar AIs as creating art. They're tools you can potentially use to create art.
In itself they're pretty boring. You can get something quite similar to a Rembrandt painting -- which isn't terribly interesting -- though always a little off. What's interesting is the neat ways people have found to interact with it. I've tried to use Craiyon for conworlding art and it's a pretty interesting tool -- but I need to come up with an interesting prompt and practically reword the image from scratch.

As an aside... a colleague of mine, who's that kind of techbro who won't shut up about AI -- did some market research. There are truly staggering sums going into AI... for very little practical returns. How long have the Tesla self-driving been in development hell now?
I suspect AI right now essentially belongs in the same category as blockchain and the internet of things -- useless tech that's mostly there to separate investors from their money as fast as possible.

Posted: **Mon Aug 29, 2022 10:25 am**

linguistcat wrote: ↑Sun Aug 28, 2022 9:16 pm Also on the art side, the current AIs might need to be retrained without using copyrighted art or art under some other specific licenses AT THE MINIMUM, depending if some of the bigger art uploading sites decide to take legal action on behalf of their users, and how that ruling turns out. But the basics of that situation is just that neither the artists nor the sites the art was uploaded to were asked permission or compensated for the use as AI fodder, so a lot of visual artists and companies catering to them are pretty angry rn with art AI programmers.

I presume you believe that artists should not look at other art that is in copyright for inspiration for their own art, because this is essentially the same thing, aside from that you are dealing with real neurons rather than artificial neural nets.

Posted: **Mon Aug 29, 2022 11:18 am**

zompist wrote: ↑Mon Aug 29, 2022 3:05 am
Richard W wrote: ↑Mon Aug 29, 2022 2:32 am
rotting bones wrote: ↑Sun Aug 28, 2022 9:32 pm I don't know if you will ever get all copyrighted material off the internet.
No, but one might be able to put it all behind a paywall or mandatory advertising.
Um, y'all know how copyright works? Everything on the web is copyright. Everything on a writer's manuscript sitting in a drawer is copyright. Doesn't have to be owned by a megacorporation or published in a book or for sale, doesn't have to be protected by letters from lawyers. It's copyright the moment you make it, and protected by copyright law. This is true both in the US and the EU. All this stuff really is there, no fooling, to protect creators.

As a corollary, if you are a techbro, grabbing art off the web is almost always stealing. There is plenty of public domain art, but things are not public domain because "it's on the web!".

It should also be remembered that being free-as-in-freedom does not preclude being copyrighted by any means. For instance, all the software I create on my own is FLOSS, but it is just as much copyrighted as Microsoft Windows is. It just has licenses with quite lenient terms associated with it that allow one to do effectively whatever one wishes with it provided one preserves the copyright and license notices and agrees to not claim a warranty.

Posted: **Mon Aug 29, 2022 12:48 pm**

Travis B. wrote: ↑Mon Aug 29, 2022 10:25 am
linguistcat wrote: ↑Sun Aug 28, 2022 9:16 pm Also on the art side, the current AIs might need to be retrained without using copyrighted art or art under some other specific licenses AT THE MINIMUM, depending if some of the bigger art uploading sites decide to take legal action on behalf of their users, and how that ruling turns out. But the basics of that situation is just that neither the artists nor the sites the art was uploaded to were asked permission or compensated for the use as AI fodder, so a lot of visual artists and companies catering to them are pretty angry rn with art AI programmers.
I presume you believe that artists should not look at other art that is in copyright for inspiration for their own art, because this is essentially the same thing, aside from that you are dealing with real neurons rather than artificial neural nets.

This isn't about how the AI works, it's about the decisions the programmers made to get there. The AIs are not conscious, the people making them are. Whatever happens in AIs making art versus how a fully conscious brain makes art is not the matter at hand, the conscious creators of the AIs using questionable or immoral practices to obtain training data IS.

I think you're overlooking that the human programmers were the ones deciding what art to include in the training data, not the AIs themselves, and they did so without permission (unless it works differently than I've been told, in which case things are probably different than what I've read). I can't unsee the Mona Lisa, but assuming that I had the art skill to do so (I don't), I could reference the Mona Lisa in an art piece or do a completely different subject in the same style. But I would not trace the Mona Lisa exactly (or a digital image) and say that was my original piece "inspired" by the Mona Lisa. However the people making AIs can choose to include training data or not. They can choose to include art from living people who never gave them permission to do so (which I would say is immoral), or they can collect art that is under licenses that allow such use, or get permission from artists to use their art in the training data. In either of the latter cases, they should give credit to the people or at least places that the training data was obtained from.

Posted: **Mon Aug 29, 2022 1:15 pm**

linguistcat wrote: ↑Mon Aug 29, 2022 12:48 pm
Travis B. wrote: ↑Mon Aug 29, 2022 10:25 am
linguistcat wrote: ↑Sun Aug 28, 2022 9:16 pm Also on the art side, the current AIs might need to be retrained without using copyrighted art or art under some other specific licenses AT THE MINIMUM, depending if some of the bigger art uploading sites decide to take legal action on behalf of their users, and how that ruling turns out. But the basics of that situation is just that neither the artists nor the sites the art was uploaded to were asked permission or compensated for the use as AI fodder, so a lot of visual artists and companies catering to them are pretty angry rn with art AI programmers.
I presume you believe that artists should not look at other art that is in copyright for inspiration for their own art, because this is essentially the same thing, aside from that you are dealing with real neurons rather than artificial neural nets.
This isn't about how the AI works, it's about the decisions the programmers made to get there. The AIs are not conscious, the people making them are. Whatever happens in AIs making art versus how a fully conscious brain makes art is not the matter at hand, the conscious creators of the AIs using questionable or immoral practices to obtain training data IS.

I think you're overlooking that the human programmers were the ones deciding what art to include in the training data, not the AIs themselves, and they did so without permission (unless it works differently than I've been told, in which case things are probably different than what I've read). I can't unsee the Mona Lisa, but assuming that I had the art skill to do so (I don't), I could reference the Mona Lisa in an art piece or do a completely different subject in the same style. But I would not trace the Mona Lisa exactly (or a digital image) and say that was my original piece "inspired" by the Mona Lisa. However the people making AIs can choose to include training data or not. They can choose to include art from living people who never gave them permission to do so (which I would say is immoral), or they can collect art that is under licenses that allow such use, or get permission from artists to use their art in the training data. In either of the latter cases, they should give credit to the people or at least places that the training data was obtained from.

But this implies that it is immoral for sentient humans who make art to look at other artists' copyrighted works as inspiration, which is no different except that they themselves are making art rather than training an AI to make art. Of course, one could argue that what is actually immoral is plagiarism and not simply being inspired by others art, and that the real immoral action on the part of the AI's designers was to not program it to specifically not plagiarize others' works. (Of course, such is easier said than done in reality.)

Posted: **Mon Aug 29, 2022 1:30 pm**

It seems to be that a number of us (myself included) consider the act of training an AI on copyrighted work, without proper licensing, the preparation of a derivative work (the AI's ability to make images seems to be directly derived from the images on which it has been trained), and consequently an act of plagiarism in itself.

Posted: **Mon Aug 29, 2022 1:33 pm**

Rounin Ryuuji wrote: ↑Mon Aug 29, 2022 1:30 pm It seems to be that a number of us (myself included) consider the act of training an AI on copyrighted work, without proper licensing, the preparation of a derivative work (the AI's ability to make images seems to be directly derived from the images on which it has been trained), and consequently an act of plagiarism in itself.

The problem with that argument is then works created by an artist who has viewed others' copyrighted works is automatically making derivative works from them simply by having seen them, if one wants to be consistent.

Posted: **Mon Aug 29, 2022 1:59 pm**

I think the matter here is that some people view the human brain as being special, whereas to me the human brain is simply a machine made out of meat, whereas a computer running an AI is simply a machine made out of silicon and interconnects. Sure, an AI running on a computer is designed by humans whereas the human brain is not, but that should not exempt the human brain in any fashion. Why is an AI being trained on copyrighted works automatically seen as creating derivative works (even when it does not specifically plagiarize others' works) whereas a human brain trained on copyrighted works is not (unless it actually commits plagiarism)?

Posted: **Mon Aug 29, 2022 2:02 pm**

Travis B. wrote: ↑Mon Aug 29, 2022 1:33 pm The problem with that argument is then works created by an artist who has viewed others' copyrighted works is automatically making derivative works from them simply by having seen them, if one wants to be consistent.

Isn't that already a problem with music copyright?

Zompist Bboard Again

AIs gunning for our precious freelancers

Re: Venting thread

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: Venting thread

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers

Re: AIs gunning for our precious freelancers