Krita Commitment to Empowering Human Artists (CEHA)

[Tymon Dąbrowski]

> Based on the conversation that we had yesterday, my impression is that
Krita might end up using Stable Diffusion's model (or something like that,
it's not very clear to me).

With all due respect, it was *you* starting the conversation about
generative AI (by alluding to it in CEHA). None of the currently suggested
projects include Stable Diffusion. I think we should cross that bridge when
we get there, and we haven't yet. Even if our users want Stable Diffusion,
there is that Python plugin someone made, we don't need to do it just now.

I suggested a project that I believe is ethical and doesn't have the
harmful traits you talk about. I think (though I still need to check it)
that it would be useful as well. I believe we should do it if someone funds
development for it (if not, we'd have some more pressing matters, but we're
broke, and someone paying us to do useful things means that the revenue we
get during that time can go to different things... like an emergency fund
or travel fund or for more developers or raises. If it wasn't useful, or
wasn't ethical, I'd have a different opinion).
To be extra clear, again: the lineart project would be based fully on the
pictures that come from artists who agree for their pictures to be used
that way, and won't do anything creative, it's supposed to merge several
sketch lines together into a nice simplified sketch/lineart. I don't think
it will harm millions of artists. I can argue with you on that particular
topic. There is an article that I linked above, so the issue is very well
defined.

I think that I realized that I *don't have* a fully developed, final
opinion on the topic of generative AI (like Stable Diffusion) in general
and I wouldn't want Krita to do anything about it just yet. Therefore right
now I'm both against implementing generative AI in Krita in the near
future, and against posting CEHA as our mission statement (I mean I guess I
might agree with the sentiment behind it, but still need to figure too much
stuff out, and I just don't want to support it yet, I'm sorry, I know you
have good intentions). For now, I'd prefer to state opinions on a
case-by-case basis.

I know, "let's not do anything", "I'd prefer us to wait" might sound a bit
cowardly, but frankly, I just want to get more data on what its effects
will be and how the world will look like in a few years. The generative AI
stuff is still very new. I feel like making a decision right now would make
it be based on my own fears or desires or hypotheses, instead of what it
actually is and how harmful it actually is.


wt., 20 lut 2024 o 22:57 Emmet O'Neill <emmetoneill.pdx at gmail.com>
napisał(a):

> Hi Thorsten, sorry for the late reply.
>
> I avoid looking at Krita stuff during the weekends if possible and since
> we had a long discussion about this stuff yesterday morning I was a bit
> spent.
>
> Drawing used to be my hobby (still is to a lesser extent). Now it’s
>> part of the toolbox that I earn money with. Because of this, I am
>> somewhat torn, but can’t afford ideology over pragmatism.
>>
>
>  Understandable.
>
> That said, ignoring what is ethical, sustainable and healthy for society
> is a dark path that we also can't afford to go down, in my opinion.
> There are a lot of bad ways (via exploitation of technology and/or people)
> that society could *optimize for money* *in the short term* by choosing
> not to care about anything else.
> In other words, *the path of least resistance to the highest financial
> return isn't always the best path*.
>
> Without getting into anything too heavy, we know that from history, so we
> might as well at least try to learn something from it.
>
> But I do see a parallel between a human learning from the work of others
>> and ML training.
>
>
> The reason you see a parallel between *human learning* and *machine
> learning *is because of marketing.
>
> Human beings clearly don't function like an "AI" does;
> we have agency to seek out our own desires,
> every waking moment our senses are bombarded with unique experiences that
> shape us,
> we have a limited capacity to produce,
> we have our own creative x-factor,
> and maybe most importantly of all, *we aren't property*.
>
> I'd argue that the companies that currently dominate the "AI" business
> want to conflate these things, specifically because it evokes science
> fiction, drives hype, and muddies the water around copyright.
> A camera or microphone could be marketed as technology which sees and
> hears just like a human does (and there would be some truth to that) but
> that doesn't dismiss the ethical questions surrounding how those tools
> might be used.
>
> Anyway, If we *are* going to start personifying "AI", then doesn't logic
> suggest that the output should be the intellectual property of the AI
> itself, and not the person prompting it or the company that "owns" it.
> This all seems like an example of "wanting to have your cake and eat it
> too" around "AI".  Either the "AI" is the artist or it isn't.
>
> So where do we come in? If you don't own the meat, and you don't own the
> meat grinder, then how can you claim to own the sausage?
>
> It is not the same as
>> flat out copyright infringement (unless a work is actually reproduced),
>> though it could be judged a different kind of infringement.
>
>
> I think that remains to be seen, and I'm not really in a position to argue
> it anyway.
> I personally do not see AI training as "fair use", but I'll let the highly
> paid lawyers and judges figure that one out.
>
> Now if the forces in favor of banning this kind of use succeed, the
>> likely outcome is that there will be nothing like Stable Diffusion any
>> more, free, available to everyone. There will still be commercial
>> generators from the few big corporations that can afford large content
>> libraries. It could well make it impossible to work on a professional
>> level with Libre Software tools (for all but a selected few with the
>> most discerning clients).
>
>
> We aren't in a position to ban anything, only to moderate our own actions
> by what we and the community think is right.
>
> I simply don't want Krita to become yet another tool that maybe helps one
> artist by hurting thousands (or even millions, or really some unknownable
> number).
>
> Based on the conversation that we had yesterday, my impression is that
> Krita might end up using Stable Diffusion's model (or something like that,
> it's not very clear to me).
> I have a couple of open questions about that:
>
>    1.  If Krita was to use another company's pre-trained (I assume as an
>    opaque, black box trained on data that is totally unknown to us) model to
>    implement various features, what would happen if that binary blob model was
>    eventually monetized in some way?
>    2. If we wouldn't be comfortable scraping millions of copyrighted
>    images from all over the internet ourselves, then why should we be
>    comfortable using someone else's model that has trained that way?
>
> Finally, the implication here that it is only possible to "work on a
> professional level" by using exploitative "AI" tools, hints at the
> magnitude of the ethical problems around this technology: It exists purely
> as a shortcut to money--not for you but for the keepers of the dataset.
> The idea seems to be to create more content with less money. Paying less
> money to human artists in the future by technologically exploiting the
> artwork of the past.
>
> Nevermind the fundamental idea of supply and demand, which says that "In
> any market transaction between a seller and a buyer, the price of the good
> or service is determined by supply and demand in a market
> <https://www.imf.org/en/Publications/fandd/issues/Series/Back-to-Basics/Supply-and-Demand>".
>
> Let's imagine that generative "AI" can produce an *infinite supply* of
> high-quality artwork and "content", do you believe that *demand can scale
> infinitely* too?
>
> Even putting the *artistic merits* of "AI" art aside, I don't see any way
> in which this path leads to anything resembling economic sustainability for
> artists--even though it's their own damn artwork that has made any of this
> technology possible.
>
> Regarding the CEHA, on the side of to be expected actual outcomes, I
>> don’t see anything empowering here. Only that Krita will be a less
>> capable tool.
>
>
> If the goal is to merely produce an image fitting a description as quickly
> as possible, consequences be damned, then Krita is already "less capable"
> than generative AI.
>
> If the goal is to help humans develop skills and create unique and
> expressive works of art, then I think we should continue to find every
> ethical and sustainable way of doing that.
>
> I feel as if there is *a lot of hand-wringing from proponents of "AI"
> around the simple question of "ok, cool, but can it be done ethically with
> consent or permissive licensing?", which maybe implies that it cannot*...
>
> Anyway, sorry if I seem a bit jaded, but I've had a lot of permutations of
> this discussion over the last year.
>
> On Fri, Feb 16, 2024 at 1:07 AM Thorsten Wilms <t_w_ at freenet.de> wrote:
>
>> On Thu, 15 Feb 2024 20:36:45 -0800
>>
>> > https://invent.kde.org/graphics/krita/-/merge_requests/2071
>>
>> Hi!
>>
>> To add a different perspective, if I may:
>>
>> Drawing used to be my hobby (still is to a lesser extent). Now it’s
>> part of the toolbox that I earn money with. Because of this, I am
>> somewhat torn, but can’t afford ideology over pragmatism.
>>
>> I am not happy about the devaluation (both senses) of art due to the
>> arrival of ML image generators. But I do see a parallel between a human
>> learning from the work of others and ML training. It is not the same as
>> flat out copyright infringement (unless a work is actually reproduced),
>> though it could be judged a different kind of infringement.
>>
>> Now if the forces in favor of banning this kind of use succeed, the
>> likely outcome is that there will be nothing like Stable Diffusion any
>> more, free, available to everyone. There will still be commercial
>> generators from the few big corporations that can afford large content
>> libraries. It could well make it impossible to work on a professional
>> level with Libre Software tools (for all but a selected few with the
>> most discerning clients).
>>
>> Regarding the CEHA, on the side of to be expected actual outcomes, I
>> don’t see anything empowering here. Only that Krita will be a less
>> capable tool.
>>
>>
>> --
>> Thorsten Wilms <t_w_ at freenet.de>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/kimageshop/attachments/20240221/91e64b05/attachment-0001.htm>