Earlier I was told that there was no difference between how a human learns and how a large language model learns, and when I pointed out differences was mocked for being pedantic and pointing out irrelevant differences. In that spirit, I have an answer:
“There’s no difference between a goldfish and a laser beam, because they both exist in this universe and sometimes show up in offices.”
I thought it was because neither one can whistle.
Because there is a B in both.
You’ve said this many times. Is there actually something in US copyright law that explicitly calls out the labor required to make something as a reason for its protection?
If I am walking down the street and see something interesting and take a picture of it, there was really minimal labor involved. But if someone directly copies the image (which may take as much or more labor than originally taking it), they are still guilty of plagiarism.
On the other hand, if I create a 3D model of an object that is very similar to something in the real world, I can make it for a fraction of the labor required for the original, but it still isn’t plagairism because it’s different enough to be a real work.
Also, some artists can create terrible crap that nonetheless took them years to finish, and it still has no value. On the other hand, a prolific writer might produce a great book or a script in a month.
Labor isn’t a value. Labor is a cost. The value of any good or product has nothing to do with how much labor went into it, other than that if the labor costs too much for the market to bear, the product won’t be made.
Paintings and books and scripts are never evaluated by critics based on how much labor it took to make them. The work stands alone. So where in copyright law is the labor required to build something enshrined as a requirement?
Training on data is not plagairism. Nor is learning from the people who went before you. For example, lots of people have criticized AI for allowing works to be drawn ‘in the style of Greg Rutkowski’. But where did Rutkowski get his style from? Perhaps Kelly Freas, or some other fantasy artist. And where did they get their style from? Probably the artists that they liked and studied.
Cavemen drew stick figures not because they were dumber than us, but because they didn’t have thousands of years of human artistic development to lean on.
“If I can see further than others, it’s because I stand on the shoulders of giants.”
That’s a pernicious request. It implies that if can’t find it in the legal code and it’s not explicitly called out, what I’m saying isn’t true. Which is, of course, bullshit. What I’m saying is based on reading about theories of copyright, and the rationale behind it.
That said, it’s implicit in the US Constitution:
That clause doesn’t make sense if there’s no labor involved.
The rest of your post is weak sauce that betrays a fundamental misunderstanding of terms like “plagiarism” and of the labor it takes to build a skillset like photography, so I won’t bother with it.
Okay, so the idea that labor matters isn’t in the law. Can you point to a court case that decided a plagiarism complaint in whole or even in part based on the labor that went into the original work?
Your quote of the constitution says nothing about effort. But it DOES say that the purpose is to promote the progress of science and the arts. Restricting AI that can do that would then seem to be against at least the spirit of the constitution.
What? Now we’re talking about the effort needed to build a skillset, not the effort required to actually create the art? Where did that come from? Where it comes from is that photography breaks your argument, because good photography requires lots of knowledge, but the act of taking a photograph can have a labor component that ranges from weeks of setup to the effort it takes to push a little button, depending on the photograph.
I like Yousef Karsh’s portraits. So I studied them, read a bit about the way he used lighting, then duplicated his setup and took some portraits of people. Did I plagiarize him? Obviously not. If a friend uses my setup and takes his own picture, is he plagiarising me? Of course not, even though he had no knowledge of how to do it himself.
Plagiarism involves taking other people’s work and calling it your own. But it isn’t plagiarism to make derivative works in the same style. Not if people do it, and not if the AI does it.
Go look and see how many variations of Edward Hopper’s ‘Nighthawks’ are out there. Both in parody, and serious derivations of it. No one plagiarized him, even though some of the paintings are very close.
None of those derivative paintings could be made if the authors didn’t study and learn the original.
As for the success of the writer’s strike, the cat is already out of the bag. If AI is truly superior at writing scripts, and WGA affiliated productions can’t use it, independent films and programs will.
AI is coming for a lot of studios, not because of the writing but because very soon an AI will be able to ‘shoot’ a movie that looks like a huge blockbuster. So if you don’t need a studio to make a movie, competition will be fierce. If the studios and streaming services refuse a superior product, they will start losing market share to those who don’t. There’s no putting this genie back in the bottle. And if AI can’t produce a superior product, the writers have nothing to worry about. The good ones, anyway.
It’s possible that people won’t need any outside content at all, because an AI might be able to make bespoke movies just for them. That’s more than a few years down the road, but not that far.
Now is not the time for anyone to cut themselves off from the potential benefits of AI. Especially since we don’t know what they are yet. We can guess, and there are some obvious ones, but the true impact of AI is yet to be felt, and probably wildly unpredictable.
Okay, so that’s not what I said. You’re repeatedly misrepresenting me in pretty obnoxious ways, and I don’t think that there’s a productive conversation with you, absent a retraction of this and a promise to stop the behavior.
Yeah, I don’t think I am. You said the Constitution implies that the labor content of the art matters. I disputed that, and asked for a court case in which that argument was made. That’s not disingenuous, it’s directly on point. It can easily be the case that a copy of a work takes far more labor than the original. No one cares.
Is the law completely up to date on AI? No, because it is about the relationship between people, and AI didn’t exist when it was writtern. So there will have to be updates to the law at some point, or some Supreme Court cases to establish precedent. Maybe the law will be changed to define AI training as some form of ‘use’ that requires permission of the authors. Or maybe it won’t, and the origin of AI derived works won’t matter at all unless the art or text is a verbatim copy.
Until then, I think you are making a spurious argument to justify supporting the WGA. The labor content of a work is irrelevant to its protected status. All that matters is whether or not the current work was directly copied in whole or in part from another one without attribution and outside of fair use.
We are at an impasse. I don’t care to talk with folks who are more interested in winning than in paying close attention to what the other person is saying, and I’m unconvinced that you’re approaching the discussion that way, and nothing in this post convinces me otherwise. I think we’re done.
It’s worth noting that when I cite things, I read my cites to make sure they actually say what I think they say. It is, ironically, a fair amount of labor to cite things in this way, as opposed to just linking to something and claiming it says something without regard to whether it actually says that.
I’m loath to do this labor for someone who is not, I believe, making the request in good faith. I’m especially loath to do it when the demand is ill-formed and only tangentially related to my claims, as in the request for “plagiarism” cases–not what I’m talking about.
That said, my failure to comply with an inappropriate and ill-formed cite request may be construed by good-faith readers as an admission that what I’m saying is wrong. So despite myself, I’ve spent some time digging through court cases.
Here’s the Supreme Court decision in Twentieth Century Music Corp. v. Aiken:
Here’s Fox Film Corp. v. Doyal:
Here’s Kendall v. Winsor (about patents, but the reasoning is relevant):
…one of the reasons why I’m so adamant that AI Scriptwriters will never be able to replace the human writer is because of this. There was a great quote that I read yesterday that summed it up:
I’ve posted the story about how Vince Gilligan came up with the idea for Breaking Bad. It basically started with a joke. And that joke put an image in his head, and that image turned into the idea of an "everyman character who decides to ‘break bad’ and become a criminal.”
So how does a language learning model go from that, basic idea to the opening scene of a show called “Breaking Bad” of Walter White driving an RV wearing nothing but a gas mask and white undies with two dead bodies rolling around in the back?
It might randomly come up with that idea. You could have the AI prompter asking for a list of potential openings scenes for the show, and " Walter White driving an RV wearing nothing but a gas mask and white undies with two dead bodies rolling around in the back" might just be one of the many options the AI Writer will generate.
But then somebody needs to figure out that this is the best out of all the other potential ideas. They need to recognize it for the gold that it is. Because the language learning model doesn’t really know, nor does it care. Perhaps there will be an algorithm that will filter the list down to the relevant “taste clusters.”
So instead of this iconic scene, we get my version where Walter White goes to school, collapses in the opening 30 seconds of the show, is diagnosed by the doctor and literally a minute later is in bed with his wife, telling her his plans to become a drug dealer.
What is more likely to happen is that you would need AI prompter would need to have at least be better than an intern. It wouldn’t be the likes of Vince Gilligan. Because they wouldn’t cross the picket line to do something like this. But it wouldn’t be someone that would be unskilled. Perhaps, like many of us, they are an “ideas person.”
So the AI prompter has this idea of “Walter White driving an RV wearing nothing but a gas mask and white undies with two dead bodies rolling around in the back” and he enters that prompt, and gets a half-way decent scene. The AI prompter tidies it up, and it looks pretty good. The questions remain:
The easiest thing to do here is to ask the AI writer to come up with answers to all of these questions. Which it will. Those answers might pose problems though, when it comes to continuity. And from the experiments that writers have been doing in the real world, AI writers are really bad at this. It’s why they struggle to get beyond a few pages of script.
In the real world finding answers to these questions is called “breaking the story.” If you don’t know what breaking the story is, here’s the man himself, Vince Giligan, to talk you through the process. I think that its crucial to watch this video because it really sums up my position on things.
The thing that most people in this thread (and apparently the executives at the studios) seem to think what writers do is “just write the script.”
But the hardest thing to do, and the most crucial thing to do, is to “break the story.” And that is something that a language learning model simply isn’t designed to do.
It’s like trying to construct a building with no blueprint, and no engineering sign-off. You might be able to build a simple lean-to. But anything more complex and you risk having the whole thing fall apart.
And it isn’t just “prestige drama” like Breaking Bad that needs to go through this process. Every scripted show on television breaks the story. Its part of the process. Vince talks about how the breaking the script can take anywhere between 1-2 weeks per episode. A show like The Rookie with on average 8 writers in the writers room might take 24 weeks to break 24 episodes of television.
How would a language learning model replace that process?
I would argue that it can’t. I think it would require an entirely different model that doesn’t yet exist.
What a language learning model might be able to do is “write the script at the end.” All of the hard work has been done: the characters have been established, the plot beats all structured, the continuity set. Feed all of that into a machine and it spits out a script at the other end.
But here’s the thing: that AI generated script will still need to have a human rewriter at the end of the process. That’s something that everyone here in this thread have already conceded. So how much time and money are we actually saving here? What value does the AI writer bring to the table? Why not just get the writer who has been part of the process from the beginning, who knows these characters back-to-front, who has the entire back-story firmly implanted in their brain, and knows how it is all going to end, write the script?
…AI still hasn’t even been able to write a single filmable script yet. It won’t be “shooting a movie that looks like a huge blockbuster” any time soon.
I note that you got in a shot at me outside of the pit without naming me. Nicely done.
It appears that what you did was just find some commentary that uses the word ‘labor’. Nothing in any of that says that the labor that went into something determines its value. I think labor here is being used to mean ‘creator’. The one whose labor created the thing in question.
But in the spirit of fairness, maybe you meant it that way, that the law establishes ownership by who did the original labor to create a work. This might be to distinguish it from ‘found’ art, or ‘art’ made by monkeys with paintbrushes or something. It doesn’t mean that the art has any value derived from the labor, but that without someone’s labor the art wouldn’t exist at all.
But AI art might apply here, if it can be established that a human provided no help whatsoever. The thing is, it gets fuzzy because the AI won’t do anything without prompting, and a prompt that might make a good story could be hundreds or thousands of words long, and ‘prompt engineering’ is becoming a field in its own right. So someone who is generating AI stuff with prompts applied their own labor to the project. This would be opposed to an AI randomly drawing something woth no input, which wouldn’t be protected because no human can claim any part of the authorship process.
I will admit that this goes a bit into the weeds for me, as I am not a lawyer. Can I claim a copyright on a programmatically generated work? Say I write an algorithm to generate an image made by various random functions. I wrote the algorithm, and created a unique piece of art. But I never created an actual pixel of the final product. Does that matter? Or if a dog runs across my carpet with muddy paws, can I copyright the result as art? Even though I didn’t plan it, and didn’t have any part in making it? I’m honestly not sure.
So if you were talking abiut merely the act of creation as ‘labor’, then yes, we were talking past each other. I was not attempting to deceive or gaslight you. And you have a point, at least for some AI generated stuff.
…in the US:
Thanks for that!
I predict there will be court cases over AI generated content created by prompts. A prompt can vary from “Write me a neat story”, to 8,000 tokens worth of plot, characters, descriptions, examples, and instructions. It’s a fuzzy boundary.
…curious to see how ChatGPT would write a feature length script, I gave it the following prompt:
“I’m writing a movie about a guy named John Smith. John is a retired assassin. He used to get paid to kill people. But then he got tired of the life. He gave it all up so that he could live the quiet life with his dog. Then one day, someone kills his dog. This is a feature length movie, so it will need to be over 90 pages long with a clear beginning, middle and end. Please write me the first ten pages of this script.”
I copied and pasted the output into my scriptwriting software and it came to 18 pages: so about 20 minutes onscreen. I won’t share it here, but I will share how it ended:
So two things here: firstly, it ended the story here. We didn’t get to 90 pages. Not even close. It started to repeat itself fairly early on. Our protagonist met his rival female assassin fairly early on. They fight, join forces, and then go after the bad guy.
At one stage John asks “Who sent you? Why are you after me?”
And the woman (who in the script, is called “WOMAN”) replies “It doesn’t matter anymore. You can’t stop what’s coming. You’re in too deep.”
Thats it. Thats the entire motivation of the bad guys. The bad guy is called “MASTERMIND.”
And secondly: the story fails to kill the dog.
Max the dog lives to the end.
Now I don’t have a problem with that. I would have saved the dog as well. I HATE it when they kill the cat or the dog.
But considering the prompt: the dog was supposed to die. I purposefully didn’t write this as the “motivating incident” in the prompt because I wanted to see if the AI would use that as part of John’s motivations. Ultimately, nobody is motivated by anything in the script. Boy meets girl. Boy fights girl. Boy teams up with girl. Boy and girl defeat the bad guy. End script.
I won’t judge the dialogue because the dialogue is always pretty crap, and it’s crap here.
It does have a clear beginning, middle and end. Except it isn’t feature length: the whole thing is over in twenty minutes.
I decided to try the prompt again to see if I could get closer to 90 pages. So I removed the line about “give me the first 10 pages” and instead just asked it to write the script.
I only get six pages this time. The female lead gets a name here: SARAH. And so does the villain: VINCENT. So not feature length. Only six minutes long. About as long as the ad breaks that play during a sitcom.
And the dog dies. But I’m not entirely sure when the dog dies. It is shortly after the dog gets issued its battle harness. Then after it dies, it comes back to life again to help deal the killing blow to the bad guy.
I’m aware that the free version of ChatGPT might be playing a part here, It might be limiting the length of the script on purpose, however it would be helpful to tell me this rather than just ending the story prematurely. A dedicated script-writing AI would obviously perform better.
But we can see from this example what the immediate limitations would be on just a language learning model. It would need to be connected to some sort of database. It needs to be connected to a timeline. It needs to understand what “dead” and “alive” means, so that all the dead characters don’t suddenly turn into zombie hoards in the final act. And it would still require significant input from human writers at all stages of the project to keep everything on track.
From the way I see it: these are inherent problems with the learning model. The actual dialogue in dedicated AI Writers I think is probably the easiest part to improve, and I think we will see that over the next few months. But thats the easy part. As it is, it can’t break a story. And it can’t maintain internal consistency. It can’t get into the nitty-gritty of plot and character development outside of the broadest strokes.
There are so many problems with this response, but this is the first one, and I’ll deal with it and be done.
I never said that the labor that went into something determines its value. I said something vaguely related to that–and what I said is what I cited. Your changing what I said to something stupid, and then chiding me for not citing the stupid thing I didn’t say, is exactly why I don’t think there’s productive conversation to be had with you.
At this point, I’ll let the rest of the misdirection go, trusting that readers will take your responses with the appropriate amount of salt.
Long thread and forgive if I missed things.
For humans how derivative can something be before it infringes on intellectual property rights?
How significantly different must it be?
Let’s face it - most of what is created is not new creative ideas. Most is not Breaking Bad. Most is as close to another successful show but with some change that makes it not exactly the same.
The big hits with new ideas are infrequent. Very valuable, like grand slams or at least home runs. But a manager may be happy with steady production of just getting on base. Another episode of the basic formula of the show. Filler eps maybe at first. Keystone bits and the rubric, character arcs, created by people. Until further advances.
Can writers win pushing that step off? Doubt it. Should they be able to?
New truly creative ideas that work are unlikely to come from this form of AI and so what? There will be further advances.