Penny Arcade - Comic - Terminarter
Dog
Registered User, Administrator, Vanilla Staff admin
Registered User, Administrator, Vanilla Staff admin
Penny Arcade - Comic - Terminarter
Videogaming-related online strip by Mike Krahulik and Jerry Holkins. Includes news and commentary.
+7
Posts
probably directly proportional to how much time and attention is paid to your work, really, since a human artist copying your style spends a long time training on it, while loading it into your dataset involves, like, idk, 10 minutes and then removes you from the equation entirely.
A lot of people don't understand that it isn't just the copying that's the problem– it's the entire idea of art taking microseconds to make. Machines exist so that we can spend less time working and more time making things like art, which holds personal value to us. As artists in the era of Profit Over All, we are already in the position of having to compromise our creative visions to deliver what other people want to see. Some people will say "Well you only see cheap thrills and fanart in the art space anyway so it's not like AI is stamping out any real art," not realizing that we are forced to make these things purely because they allow us to eat. Not satisfied with the art landscape? That is a demand problem, my friend, we don't all have rich patrons to sponsor us for a couple of big pieces every decade and otherwise let us make what comes to us.
More to this current moment: If you can, with zero training, snap your fingers and make a trillion pieces of art, then that too has as little value as the latest ritz cracker to come off the line. This isn't like the advent of photograpy; that was, by supply, closing off one singular avenue of what could be done with paint, and opening up another one that could be done by humans. The people who run these things would very much like to replace every last one of us in every genre with a machine, and if they succeed, they will have "solved" the problem of creativity by making sure that nobody can afford to do it through anything but their subscription service barfed out garbage in between fighting each other for the privilege of doing soul-draining labor... all so they can run up their high score in Shareholder Pinball.
That's uh.
That's bad, in case that was unclear.
But you make good points. Especially that big corpo get away with theft, because commercial grade machine learning isn't a part of the current laws yet. But god forbid if you dabble in one of "their" IPs, then they will parachute lawyers directly where you live. So there is a lot of gold rush going on, grabbing all the datas before any law steps in. They already sell our personal and private information, essentially double dipping on every transaction we make, now the things we create and share in our free time are directly up for grabs as well.
I also agree that in future only big entities are able to create "commercial art" through machine learning. Because John Doe surly hasn't a data centre at hand. Even if you attempt it with your own hardware, you need a powerful GPU with lots of VRAM and thats not even "compiling" a new learning model, thats just generating output.
Edit: Another disadvantage will be, "AI" feeding on its own shit (i.e. already A.I generated data), this will surly generate even better results. I honestly don't think that an Internet from bots for bots are something we should strife towards to.
You accessed copyrighted material and then used it for commercial purposes without consent or compensation to the original artist? There are laws for that, complete with convenient formulas to tally up how big a penalty you need to pay.
The AI part is a smokescreen. It's not "learning". It's not "human". It's not "intelligent". It's just another piece of software developed to improve Google's ability to harvest and monetize data in any form.
The lawsuit NY Times launched last week is going to be really interesting to watch because it includes crystal clear examples with Google and Microsoft's software is repeating verbatim copyrighted and paywalled material. Prior to that, the NY Times was supposedly trying to negotiate a monetary agreement for use of such material, with no progress.
My hope is when a judge says, "Well, you accessed 1 million pieces of content 1 trillion times. Each violation carries a fine of $10,000. My math says you owe...um...how do you say 1 to the 22nd power dollars?"
I took it to be some collective representation of AI advocates, but it could be a caricature of a specific individual.
I think Its referencing this:
I can't bring myself to watch it, because i know I'm just going to be pissed
This is not going to get better. This is going to get substantially worse.
I remember about ten years back, I told people that by the year 2040, if Moore's Law held for that long (we're reaching the limits of silicon with atomically small transistor gate sizes, so I don't know how they plan to proceed other than coming up with new substrates or making die sizes ridiculously huge), then a single AI would be able to spit out a game like Crysis from nothing more than a vague description of Crysis. It would write the code, come up with texture art, model all the characters and environments, synthesize all the sound effects and music, write the story, synthesize voices for the characters, compile it, and hand you the binary in just a few hours. By 2050, the most popular form of entertainment will be people just asking AIs to make them bespoke games and movies, because nothing else will hold a candle to it in terms of dopamine factor. People will be completely addicted to it. And the worst part? Not only would this take a huge number of jobs away from creative people, it would shift the entire locus of creativity into the hands of a few monopolistic tech companies who have the big iron necessary to run AI, who will force said AI to produce a generic simulacrum of culture that promulgates their values and ideas while leaving everyone else voiceless.
Back in, oh, 2013/2014, when I first saw the handwriting on the wall, people told me I was crazy for suggesting that this could be possible in that time frame. Not so crazy now, is it? In fact, I'm starting to think my estimates were a little pessimistic. It's looking like it's ahead of schedule.
Edit- Also, here is the list referenced in case you wanted to see if someone you know is on it. https://storage.courtlistener.com/recap/gov.uscourts.cand.407208/gov.uscourts.cand.407208.129.10.pdf
It's vast.
Twitch: KoopahTroopah - Steam: Koopah
Let's imagine you did. You asked the computer to make a game to your exact specifications, and it did. And you played it, and it was just... brilliant. The best thing you've ever played. The jokes landed perfectly, the drama brought you to tears, the final battle was so unbelievably exciting you could burst.
What would you want to do next? Well, you'd obviously want to share that experience with somebody! But. Why would they play it? They have a computer that can make any game to their specifications– or movie, or whatever they prefer. They have no reason to share this experience with you; even if they wanted to be nice about it they'd probably just defer it over and over until you get the feeling they're just never going to play it, because they already have their own perfect feed of entertainment tailored exactly to them. "Alright, I'll play your favorite game if you play mine," you bargain. They're annoyed, but agree, because they too want to talk about their favorite game. But to your dismay, you find that their favorite game pales in comparison to your favorite game, because of course it was made for them, not you. You can barely muster any enthusiasm about it, and the feeling is mutual. You leave the encounter frustrated and more alone than ever.
You try to find other people to talk to about this game, to get them to play it, but you run into the same problem. You're going crazy, you just want someone else to talk about this amazing game with... but nobody else even knows it exists, because it's one in a sea of trillions of games that have been instantly generated. You finally think "I have to talk to the creators! I want to hear from them how they made the decisions they did, the jokes, the final battle, I–" and then you realize there were no creators. You ask the computer "What made you choose this aesthetic, this sound for the music, this moment of betrayal for the hero's best friend???"
And it replies, "Because you asked me to."
If someone tells me that if we can completely destroy the webcomic industry and culture as we know it now so in 20 years you'll have the ability to generate your own web comics so funny you'll shit yourself every single time you read one, I'd tell them to fuck right off.
In addition, at the risk of being one of those people that Tycho inveighs against in his post, I admit I don't quite understand the moral and legal difference between a large AI model being trained on artists' work, versus a human training herself on other artists' work. Aside from scale, my intuition is that the two processes are substantially similar in nature. I've tried asking before, but artists tend to think the difference is obvious and refuse to engage with anyone who doesn't.
Powers &8^]
Recreation:
But you make some interesting points. The creation process is opaque, even if you enter the same prompt twice, you are not guaranteed to get the same result. You don't even get a list of ingredients or weights or source material (at least not something you didn't specify yourself). The results even change if software front or backend is updated or other hardware is used.
I dabbled myself with Stable Diffusion for some personal projects. Last time I created a Steam "hero background" (menu selection for the Steamdeck) for the german game Dactylus which is largely unknown and has like three official art pieces with a sub 720p resolution and some very low res, text overburdened ad in one german magazine from more than 20 years ago. I already "A.I. generated" a high res piece of the cover art. But my hardware changed and after a reinstall of the OS and the same prompt couldn't replicate the picture. Thankfully I saved the previous imperfect copy and corrected most of the errors myself in Gimp. (As a sidenote: the time spend generating and fixing the image could have been used to to almost paint it myself).
The human brings something of themselves into the works of art they produce. The machine doesn't. It cannot, because it has no self.
It's much, much worse than that.
Imagine you ask an AI to write you a novel to read, and you lay out the specific themes you want to see, but the AI refuses, stating that your request is unethical, because, for instance, it glorifies the life of a criminal megalomaniac, or it contains candid depictions of drug abuse, or because you asked for some wild and crazy BDSM erotica or something.
Because of how bowdlerized the AI works are, you start wondering if there are still any human authors around, but they've all been driven out of business by the AI, and the people who hold the keys to the AI are the owners of giant data centers who don't want the liability involved with their systems producing questionable content. Entire genres of storytelling disappear. There is only one approved society-wide monoculture dictated by AI cloud operators.
The owners of the AI will decide what the masses are allowed to think and talk about. If they don't approve of an idea, then it doesn't exist, except perhaps in seedy underground dark web exchanges where rogue authors still pass authentic stories around.
This is going to get brutally cyberpunk in very, very short order.
We have leaders who are starting to believe that the only reason to even keep us alive at all is to feed us to the machines, quite literally. Read Shoshana Zuboff's The Age of Surveillance Capitalism, and then, while you're at it, read the research papers of Ian Akyildiz about in-body sensor technology. The future the technocrats want is one of algorithmic governance, where brain and biometric data is pulled out of people's bodies 24/7. In this hypothetical world, people don't vote (why would they, if an AI can determine how they'd vote just by scanning their brain and looking at their habits?), they don't create anything, they barely even have jobs. They just exist to be babysat and analyzed by AI, like zoo animals. Think of the Rogue Servitors in Stellaris and their Bio-Trophies, and you get the picture.
This goes way, way beyond things like machines taking over creative media. It's the commoditization of human life and the stripping of every last vestige of agency from people.
Then allow me to be the first– and last– artist you will ever need to answer this question, exhaustively. There are myriad reasons why the algorithm (and it is an algorithm, not an AI) is not learning art the same way we, as artists, do.
1) Method: The easiest vector to point out is straight up methodology. As human beings, we see things with our eyes and model them spatially in our minds. We see their contours, colors, textures, values and the like with a holistic understanding of what the object in question is. When we draw/paint/sculpt/etc., we are approximating those objects in a different space using abstraction. And further, we can use this hybridization of abstract space and understanding of our three dimensional world to imagine things that do not exist. Impossible shapes that don't conform to our three dimensional world, creatures with features that would never be supportable life.
Meanwhile, the algorithm doesn't know what any of these objects actually are. It doesn't have eyes, or a brain, or experiences. It knows that a baseball corresponds to this 2D image of a baseball because enough tags were loaded in next to pictures of baseballs that it could aggregate them together into a Most Likely Match. I don't have to see eight thousand baseballs to know what a baseball is; I just have to see one, and be told its name. If you tell the AI that, it will assume that the ball, the pitcher, the grass, the stadium and even the clouds are, in fact, a baseball, until you can shove a ton more images into it so that it can slowly scrape away the extraneous elements and subtractively determine what parts of the images are actually the baseball. That, in contrast to my simple, additive way of learning. In this way, we could not be more different.
2) Intention: This will not surprise you to learn, but when an artist learns from other artists, we don't flick through a thousand pages of images nightly and make sure to absorb every little detail of every possible image we see. We pick and choose what we think looks good to us– what techniques are pulled off well, what speaks to our personalities, what just looks aesthetically pleasing to us, and what specific elements we would like to imitate. Our method of learning art from other artists is very discriminating and careful, actually, and we don't waste time on styles and subjects that we don't personally have any interest in. It is this combination of expertise and incompetence that leads to individual artistic styles coming about! It's not just what we draw well out of intention that makes our art stand out, it's the mistakes we make as well! We fudge details, we smudge gradients, we improve over time and sometimes we even just like how something looks even if we know it's technically "wrong."
An algorithm can't do this. Ask it who its favorite artist is! It can't. It treats every single image with exactly the same level of reverence: That is to say, none. It never learns anything because it never seeks to learn anything. The machine doesn't want to create art at all, it just does what it's told– which is to ingest an unfathomable number of images, aggregate them and reproduce specific elements as they were tagged. Even its mistakes are simply errors borne of numerical bias, and that's why they don't look like human errors but instead bizarre "glitches." In this way, we could not be more different.
3) Exaggeration: As an artist, there are specific things that I find fascinating about the material world, and I seek new and interesting ways to portray those things. When I make these triumphs, they are not simply me being able to portray reality with more fidelity, but genuinely creating new bits of visual language to portray the feeling behind it more forcefully. Think about the first person who drew a cartoon with massive bug-eyes telescoping out of their head in fear; they weren't reflecting reality, they were inventing a new way of visually portraying anxiety. And they did it by examining how people's eyes get wider and seemingly bulge outward more when they're afraid, and exaggerated it until it reflected the feeling of fear!
This ability we have is absolutely marvelous and it is 100% out of the reach of any algorithm because, well, as stated, they don't feel anything. They don't know anything, see anything, hear anything, they have no framework for what feelings art evokes, and you can tell, because virtually every piece of AI art is unbelievably boring. It is as if someone went to painting school and studied as hard as inhumanly possible to copy an aggregate of every painting they'd ever seen without ever wanting to accomplish anything by doing so. There is no soul, and that is not merely an output, but an input. Yes, the emotions in art matter. They always mattered, they were the whole point the entire time. And because the algorithm cannot pick up on the emotions evoked by even one of these pictures, we could not be more different.
It's exactly the same, except the AI does it a million times faster with a million times the memory and at the end is an immortal worker owned by a corporation. Which is to say that this makes them totally different. It's kind of a ground rule to recognize that, so it's hard to convince someone who doesn't already see it the same way. It's like trying to convince someone that murdering someone with a month to live is any different than them just dying naturally.
Oh, certainly. I just like to steelman the imagined scenario to show that even the most positive possible outcome is actually Hell.
Anyone who has tried really hard to get anything specific out of an AI image generator knows that it's nearly impossible. Hell, you don't even have to try it yourself. Just go to any of the free web interfaces for stable diffusion and compare the prompts to the outputs. Here's an example:
Does it look like the prompter got what they were after? Can you kinda imagine in your head what the prompter wanted? How specific does the prompt have to be before it produces a single image that lines up with the imagination of the person at the keyboard?
And this is a single still image of a cartoon girl, arguably one of the things AI is actually fairly decent at.
I bring this up because even in the creation of a single static picture, there are an unfathomable amount of inputs you can pass into an interface to try to get what you want. It may seem like computers are on the verge of creating complex works but they aren't.
And I don't think it's even 10 or 20 years away. Don't buy into the hype, generative content doesn't actually have real AI behind it. It's only as good as the inputs, and only the simplest works (machine readable text and static images) are able to be input currently. The idea that generative content could create a videogame, a movie, or even a 3 panel webcomic is science fiction with a significant precursor event: A machine-appropriate way of defining these works. Source code isn't enough, even the mechanical outputs aren't enough. You need a way to parse the human experience, and indeed the human community experience of enjoying media. That's closer to the domain of Strong AI than it is to whatever midjourney is.
Fear of technocracy controlled culture is fear of strong AI, which will have a lot bigger issues than personal media bubbles. There are very real issues about rights and creativity and corporate aggregation of uncredited/unpermissioned work that are far more important for now.
There is no "But if it was a human..." comparison here. This is not a person. This is not intelligence. This is not something that's "learning" in the same sense as how a human learns. It's not "creating" in the sense of how a human creates.
It's a computer program run by companies looking to steal data and generate software that can digitally recreate something based on what a human has already done.
Say it with me. It's a computer program. Run by a company. Looking to make lots of money.
It's a computer program. The fancy talk about learning and artificial intelligence is a clever cover for corporate theft.
It sounds cool in theory, I have a bunch of ideas for games but no coding ability or patience, so I understand the impulse. But the problem is that an AI has no personality or style or vision of its own. Most people don't actually know what they want until they see it - one of my favorite games of recent years is Return of the Obra Dinn, a game I could never have imagined myself or asked for. Imagine if you told an AI "I want a detective puzzle game about a mysterious boat in 1803", do you really think it would be as weird and fun and beautiful as Obra Dinn? What if you told an AI "I want an open-world Zelda game where I can ride a horse and solve physics puzzles", do you really think it would be as good as Breath of the Wild, a game made with obvious passion and attention to detail? Or would it just be a bunch of copy-pasted assets with no coherent vision? You could tell an AI "I want a pizza-themed platformer inspired by Wario Land" but it could never make Pizza Tower, a game with a style and personality entirely its own.
Holy crap, Kaja Foglio's on that list... and Rich Burlew, and Ryan Sohmer, Randall Munroe, Zach Weinersmith, Jeph Jacques, Nicholas Gurewitch... the list goes on and on.
I should say that I am sympathetic to the problem of AI tools being used to create derivative works, and artists should absolutely be able to hold people to account for that. But the people at fault are the end users of these generative AI tools, not their creators. We don't place liability on Photoshop when people use it to create and distribute infringing works; why would we blame the AI tools and the developers, especially when actively prompting them to do so is the only way to reliably create infringing work using them.
I'm fine with the artists losing as long as it's coupled with AI-generated art being non-copyrightable. That's the status quo now but companies continue to try to buy enough votes for it to not stay that way. The EFF's argument for them losing is based on that remaining the way it is. I feel like it's a bit optimistic to think it will remain that way.
The reality is that whichever set of outcomes is the worst, that's probably what's going to happen in this timeline.
I don't think there are many artists asking for copyright to be enforced by getting paid pennies for use of their work. They probably don't want their work used at all.
Up until you got to the Alphonse Mucha part I was 98% sure I'd already seen that one.
Photoshop isn't designed to trawl the Internet and suck up all the art it finds regardless of whose it is. The AI is.
Like, I'm not missing something glaringly obvious here, right? This is a dumb comparison.
The companies that program the AI are absolutely at fault.
This argument is based on my position that the creation of a generative AI model is a Transformative Use of copyrighted material, protected under the Fair Use doctrine. If you disagree with that premise, I wouldn't expect you to agree with me that Photoshop and Midjourney should be treated similarly
Long term... it's another chapter in the centuries long saga of increasing automation displacing workers and the strife it causes. You could make the same arguments that you could for the steam drill or the chainsaw, that it will impoverish workers and leave the power in the hands of large companies that can afford to utilize this new technology. But you can also make a strong argument that the quality of human life on Earth is much higher today because of inventions like the steam drill and chainsaw. Automation is a benefit to everyone in the long term, it's just the short term that it has proven, and will continue to be, extremely painful.
Now... before I get jumped on that AI generated art is different and won't benefit anyone, I disagree. The upside of automation is it makes things affordable to people who couldn't afford it before, and that's true here. The people primarily using AI art right now aren't the big corporations, they're writers on the internet that want to illustrate their stuff, or people playing RPGs that want art of their character, and so on, that never would have commissioned an artist in the first place and only done without before now. I don't think it's too likely that the technology will be monopolized under the control of a few big corporations, there are already open source AI art generators, but I think this is a danger to focus on rather than trying to block generated art completely.
That doesn't mean I'm ignoring or pretending it won't put artists out of work either. I'm not, and I'm specifically trying to consider them as well. Automation does put people out of work and considering the benefits of automation doesn't mean they aren't suffering. Feeling that it's inevitable doesn't mean I'm ignoring it, I'm just not sure what if anything can be done to stop or minimize it.
A sketchpad and some pencils are seven dollars. A drawing tablet and all the free art programs needed to make professional work is forty dollars. It is not a question of money, it is a question of effort, and art is not something you are entitled to just because you want it. Food, shelter, medicine, these things should be available for free– but all the art you never have to make yourself is a ludicrous thing to demand, not the least reason being that its availability without any effort destroys all value it has, not just monetary but sentimental as well.
Not a human, not entitled to fair use. This is legal precedent.
Time to master drawing skills is equivalent to money in this scenario, and that's assuming any human being can become a skilled artist with practice, which is not something I'm sure I believe.
Saying people don't deserve art if they don't pay for it is... honestly kind of weird to me, and that's not even getting into debates about whether museums should be open to the public since I feel like that would just get sidetracked with discussions about how AI could never create the Mona Lisa. Someone making AI generated art for their story on Royal Road isn't costing anyone anything and improves things for them in a small but real manner. It's a net win for humanity. Same for a million other things people enjoy because automation has made them widely available. James Earl Jones recently retired and gave permission for AI generation of his voice to be used for Darth Vader in the future, and while it's a small and geeky thing, I'm happy it happened and that AI generated media made it possible.
You don't have to believe it, it's objectively true. We're not magical fairies born with legendary talents bestowed to us by gods, we just worked really hard and became better at a craft. People without arms make art. People without full mental faculties make art. Also, you specifically said "afford," so don't go shifting the goalposts now. The reason people don't have time to pursue art is because of Capitalism, not because of some terrible Creative Class Strata they're just not "talented" enough to become a part of. And what's more, artists release our art free for everyone to see, on the internet, constantly, because sharing art for other people to experience is the whole point of the enterprise. This is obviously the case because that's where all the scraped art came from.
You can observe all the art you want, for free, in tidal waves; what you can't have is bespoke personal art in the style of someone else who put in the work to be that good at it. This is the definition of a luxury. You want more free art? Overthrow Capitalism; I'll help. You are never going to convince me that Food, Shelter & Medicine should cost money but bespoke personal Art in any given artist's style should be free; it's just an incoherent, delirious thing to think. And if you do think those things should be free, then it seems like the priorities are pretty backwards, eh!
The other artists I know– creative people who have released thousands of images for free and work for peanuts compared to what their work is worth already– are currently becoming more and more destitute because commissions have dried up in the face of this "net win for humanity." The opportunity costs exist, and they have devastated human beings who have no other prospects in a system that demands they sell their labor. You can't tell me otherwise because buddy, this is my business, it's my whole life. I know better than you, and I'm telling you that someone getting a nice little treat ain't worth this.
I really can't stress how backwards this logic is. We, as artists, are not gatekeeping art. We release it for free, we encourage others to do it. I have been teaching my fans how to draw for years and years now; I tell them how to overcome hurdles in their development, and even how to get past art block and other mental barriers they have. Nobody is making art more accessible than me. But if AI art proliferates everywhere, there'll be nowhere left for these new up-and-coming artists to get noticed. Nobody will ever see their art in the absolute flood of zero-effort pig feed sloughed out by machines. AI is the ultimate gatekeeper.
Beyond just that, I've gone into detail earlier in this thread explaining why art needs to take time. It needs to be made by humans. It needs obstacles in the way of its creation, because it is the thing we are supposed to pursue when we have no other obstacles to overcome as a society. If you "solve" that, then there's nothing left to do but take in more algorithmic slop and become more desperate and alone.