I don’t think it would be challenging to make an image generator that “respects copyright” (you could train it on public domain art and photos but you could also license massive libraries of stock photos and TV shows and book/album covers etc. from the media companies that hold the rights to them) and I think the existence of such a generator would not lead people currently mad about AI to suddenly be cool with it because it’s really not about copyright.
machine-saint said: adobe and gettyimages in fact both have generators trained on their stock photo catalog!
machine-saint said: there’s also mitsua diffusion, which is trained on public domain/permissively-licensed art
cadmusfly said: @machine-saint there’s also CommonCanvas, an open source model that is apparently comparable in output quality to Stable Diffusion 2.0 and is Coming Out Soon ™
#huh#I do estimate some people are actually worried about job displacement#and the concern of intellectual property is more having to do with their assumption that otherwise#these models could not be created in the first place… but as far as I can see it would only delay it by max 5 years or so#the part where I pick some pics by a specific author to specifically replicate their style feels like a dick move#for the same reason people tracing art and passing it off as theirs is a dick move#but also this seems to be a separate thing from crawling the internet and picking everything that goes (via @mlembug)
I do think it’s worth noting that there are definitely some people explicitly taking the stance of “AI art through public domain and licensed datasets is fine but other kinds aren’t”, but I imagine it will be hard for that to hold up as a meaningful distinction in the long run. AI art will continue showing up in more and more places, with the more limited datasets still causing all the economic concerns people fear from the tool in general, and over time people trying to draw a line between “ethical AI art” and “unethical AI art” will have to figure out if their default assumption upon determining an image is AI art is one, the other, or neither.
It ends up not really being a stable position - as we saw with the whole antis situation, a reference point I keep finding pertinent. People would go “you can’t do these things in fiction unless you have an excuse” but they end up having to react to things without knowing an “excuse” is present, so they end up either tending to react with disgust or wariness and ending up spreading that uncharitableness to the whole category regardless, or relaxing and inevitably reaching the conclusion that it’s fine either way.
ultimately I think economic considerations will determine the outcome more than attitudes: if AI makes things cheaper and faster then damn right it’s going to get used everywhere.
I think this is kind of ignoring that art is always following taste. and taste has shifted to have anything that’s AI registered as being cheap or low quality. there is no assessment of quality without assessment of method. we have seen that artists whose style looks “kind of like ai” have seen a sudden drop in success because of the shared negative association. there’s a good likelihood that artistic direction in a lot of cases, even corporate one, will shift to a visual style that intentionally would be harder to AI generate to prove effort and therefore legitimacy
we always exist in a cultural context and there is not actually a neutral “just liking how something looks”. art doesn’t serve the purpose to look good at least not on a grand scale. it’s for self expression of course but arguably more important than that it’s point is to legitimize whatever it is attached to. status through skill or monetary means or connections. doing a fanart of something for example isn’t just producing another image of a character somehow. it’s expressing a thought and it’s expressing a connection of the artist to the work and giving the work more legitimacy by showing dedication to it and it’s giving more legitimacy to the artist by them displaying their skill. if there’s no skill, effort and means displayed that makes art in practice pointless. we see this to some degree with art we have no personal connection to and can only abstractly guess the context and effort. expressionist paintings and traditional church chants will do nothing for you if you don’t have a cultural connection to them. AI art being faster and cheaper makes it worth less. the value isn’t derived from isolated visual appeal. there’s an infinite amount of things that are nice to look at. with oil painting, simply paintings of landscapes are very cheap because they’re easy to do, even if they look very nice. with digital art, a detailed full render is held with higher regard than a simple sketch. not because of how much nicer it looks, but because of what it communicates
when we don’t like a show we don’t say “i don’t like how it looks”. we complain about cost cutting measures.
one way or another, art is a way to connect with people
mhm. luckily because this is the case i don’t think artists have to worry much in the long term. ai can be used in whatever ways and culture will dynamically respond and adjust the value of how it is used depending on association. i personally think it will largely disappear from foreground prominence in the next few years and probably be used as a cost-cutting measure in places where it’s not as obvious. we are currently in a time where taste is rapidly adjusting and after a while it’ll settle
much like the transition from hand drawn advertising to photograph based advertising, or the transition from hand drawn cel animation to computer animation, etc. etc.
Not to be That Bitch, but:
Me.
I’m the one who would still object.
You can train your AI entirely on images from 1743 and I will still be screaming about the environmental impact and the fact that you were too fucking lazy to think so you outsourced the thinking to a machine and want to pretend the result is on par with real artists who actually do the work.
the environmental impact is lower than getting humans to do the work and getting cheaper all the time, and the last Disney movie animated with old school cels was The Little Mermaid in 1989, technology moves on.
Every image generated by AI wastes 16 ounces of water.
And I frankly don’t give a shit what Disney does. In case you haven’t noticed, they’ve been horrible to their cast, crew, local environment, and customers forever and it’s only getting worse.
If “cheaper” is your goal, fine. Drench the world in shit. It’s not like we’re already oversaturated with it. This argument to me sounds no different than the people who ask me to make them a blanket for $30 because that’s what it would cost them to buy one at Walmart and then get mad when I tell them $30 wouldn’t even cover the yarn but unlike a Walmart blanket the ones I make will last for decades.
That stat doesn’t sound real.
Specifically, how does AI waste water? The cooling? That’s a closed loop, the 16 ounces get recirculated until they evaporate and return to the water cycle. It’s not wasted, you cannot waste it, thermodynamics make it so it always returns to nature. Lots of sensationalist media are trying to make it sound like that water is disintegrated and leaves the planet immediately.
Beyond that, it’s getting more efficient over time. Complaining that it’s inefficient and hating it forever is like never wanted to get in a car because the Ford T puts out too much exhaust. They’re not always going to be that bad, obviously earlier models are inefficient, that’s just how technological development goes. First you make a big, bulky, wasteful thing that gets the job done and then you refine it over time until achieving maximum efficiency. We’ve been doing that since the wheel.
Also, the development of energy consuming technology is in itself a driving factor for its efficiency improving. Right now Microsoft is in talks to fund a nuclear power plant to feed its AI, meaning eventually all their output can be pulled out of the grid. Google just developed a more efficient multiplication method that will over time make computer hardware as a whole less wasteful. Solar panels keep getting cheaper and developing new tech dependent on more abundant elements to counteract rare metal mining, because rare metals are expensive and companies would prefer not having to pay for them. Everyone is working towards cheaper, more efficient AI because they benefit from cutting energy costs as much as possible.
As for copyright, yes, it will eventually be indistinguishable and arguably already is. That’s not an argument against AI; it’s an argument against copyright. We found an unpatchable leak in the system, and the only way around it is to forgo copyright altogether. That is a good thing. It will have to happen within our lifetime.
We are being asked to reexamine our definition of art. The conversation cannot begin and end with “latex paint is bad for the environment”.
that statistic is absolutely not real because you can generate dozens of AI images on your laptop without filling it with water.
and you can do so while using less energy than someone would need to spend an hour drawing in photoshop
I think it was very much a strategic mistake to base opposition to AI imagery on its energy consumption!
The best and only real criticism is the fact that it further alienates the worker from their labor and its products. All those companies investing in AI are just attempting to get in on a corporate power grab to decrease worker power, they do not care about anything else.
We should be focusing on that when it comes to the “why is generative AI bad?” conversations.