Writers Strike - AI demands unlikely to succeed

Cafe Society
121

I think this is a place where I differ with @Banquet_Bear. He’s a lot more confident than I am that AI won’t eventually be able to produce full written works, in much the same way that they can currently produce full visual works. I suspect that in a few years they’ll be able to produce books in the Daisy Meadows series, and in a decade will be able to churn out a Terry Brooks novel, and within a human generation will be able to produce CSI Miami scripts.

But I’m glad that helps to clarify what I’m saying.

122

Wise? I’m not sure that’s the word I’d go for but as for making the world better, I think it is reasonable to have some degree of protection, for a certain period of time, for the original work you create. It doesn’t seem morally right to create an exact copy and pass it off as original and profit from it.
If a person knows that the item of worth they’ve created can be identified as theirs and they get rewarded for it, then they are likely to be motivated to create more.
More worthy art in the world seems like a “better” world to me.

123

So…Dan Brown next week then?

124

I would argue that we as humans create by copying (and building on that) in one sense or another. As a creative, I sure as hell do. I don’t want to admit it, but I usually can dig down deep and find what inspired a certain photo of mine to me.

I don’t get the idea that Chat GPT can’t create a story with a coherent beginning, middle, and end. It has for me. Not script-length stuff, but short stories. It’s quite good at that, in fact. It’s absolutely coherent. It’s just a bit trite most of the time. But if you converse with it, ask it to spice things up, introduce unexpected elements, etc., it does a decent job. I have absolutely no reservations that in the near future (10-20 years) it will be able to create broadcast quality scripts.

125

…this thread is about script-length stuff. It’s about the writers strike. No: ChatGPT cannot produce a twenty-two-minute sitcom with a clear beginning, middle and end. It barely can get 2-3 pages (about three minutes of dialogue) before it goes off the rails.

And what it does produce is obviously derivative of existing work. It starts to repeat itself. It doesn’t have an imagination, so the plot beats and dialogue are generic and expected.

Yeah, it will get better. And yeah, you can have a human “pilot” that guides the AI step-by-step to shape a better story. But if you are going to do that: then why not simply have a person write the script in the first place? What value is the AI bringing to the table? What does it add to the production that makes it better?

I mean, the answer to the question is yes. Yes the WGA will stand with the other guilds. They’ve been clear about that. Just as the other guilds have been clear in the support of the strike.

The Screen Actors Guild board have called for strike authorization ahead of their contract negotiations.

The Directors Guild contracts expire at the end of June. Yes, AI will be part of the negotiations for them as well as the other guilds because AI is an existential threat to the industry.

Strawman. Its less to do with “playing around a bit in ChatGPT 3.5”, and almost entirely about knowing what actually happens in the writers room. There is a reason why Ford shut down Argo AI, Rivian is limiting development to L2 and L3, Waymo and GM’s cruise are limited by geography. We were hearing “AI models are improving at an astonishing rate” years ago, so where is my self-driving car?

The reality is the human brain is actually something really fantastic. And that humans, while flawed, are still pretty fucking great. AI isn’t magic. And even after a decade of development, the best autonomous cars are likely to be out-driven by Uncle Bob in the hilly narrow streets of Wellington.

So forgive me for being skeptical. When the AI can actually write a 20 page filmable script then perhaps we can talk.

Laws already exist that protect intellectual property from being used in ways that the owner of the IP don’t want it to be used. Arguably (and we will see how the cases currently at the courts pan out) using IP “as training data for AI writers” is already covered by them.

What the WGA is asking for here as part of the negotiations is that MBA-covered material not be used to train AI. This is what is on the table at the moment. That’s relevant here.

Not really. AI is like fully autonomous cars, or pivot to video, shiny new trends that everyone is keen to jump on, but ultimately won’t deliver. It can do some basic things really well. But it can’t do everything a writer can do during a television production.

126

I agree. I think similar reasoning applies here, something like:

My concern is that, as artists are unable to control the use of their creations by AIs, they’ll be compelled to limit the distribution of their art. As corporations find cost savings in AI, there will be fewer people who can afford to create worthy art. Less worthy art in the world seems like a worse world to me.

I just watched the third episode of The Diplomat, which is full of sparkling dialogue with surprising metaphors and anecdotes. I just read another teacher’s list of Chat-GPT-created jokes around exams, including such gems as " Why did the state testing room get so hot? Because all the students were ‘testing’ their patience!" and “Why did the English test go on a diet? It wanted to lose some unnecessary words!” I really don’t want a world in which artists can’t afford to create art because they can’t share in the profits from their art, so we get ChatGPT writing The Diplomat.

Again, I recognize that current laws don’t cover this, just as laws in the fifteenth century didn’t cover the printing press. I won’t sneer at folks who balk at my proposals that they’re “afraid of change,” because that sort of reductionist nonsense is insulting and fatal to conversation; but I do think we should consider how we adapt to change, and we should be frank enough to consider what changes are appropriate in our legal frameworks in addition to our technological frameworks.

The most straightforward fix, I think, is to require owners large-language models to secure permission for incorporating works that are under copyright. This will go a long way toward spreading the wealth and returning control to the creators. If the models really are that great, then their owners should pay for the intellectual property that’s making them great.

127

The Da Vinci Code unironically posits that Ariel from The Little Mermaid is part of a millennia-old conspiracy. Are you telling me it wasn’t written by ChatGPT?

128

…I think we have to see what happens after this round of lawsuits play out before I could comfortably say this is the case. In the Getty lawsuit they’ve got AI generated images that literally show their copyright watermark.

Sudowrite that I talk about upthread, based on GPT-3, admits right there on its home page that the tool will plagiarize if given the right prompt. That’s just asking for trouble. In the EU Clearview AI got fined 20 million Euros in part because “Individuals’ rights not respected” when peoples photos were used as part of the dataset. And this happened a few days ago.

A lot of this is unsettled. The issue here is the “act now apologize later” nature of “disruption.” Its “we will do this thing, and then once we’ve done it the system will have to adapt to our thing”, not the other way around. So they built and “trained” these language learning models first without asking anyone’s permission, and now they are trying to monetize it.

Which is where we are at now. The “asking for forgiveness” part. I’m not particularly in the mood to forgive.

129

Do rewriters get residuals?

130

It isn’t like the source of the images is secret. They come from laion:

And laion comes from Common Crawl:

Both of which are free to anyone and everyone.

131

…“rewriters” don’t exist in terms of the current WGA agreement. This is new territory. The WGA position in the current negotiations is that AI can’t write or rewrite literary material, that AI can’t be used a source material, that it can’t be used to train AI for any MBA covered projects. So as far as they are concerned, “rewriters” aren’t even on the table.

If scripts get rewritten (and this happens quite often in film) then the guild rules dictate how they are credited and the compensation. In the television writers room it isn’t uncommon for the showrunner to rewrite substantial portions of the script but the credit still goes to the original writer. (this can change depending on the showrunner though)

As for residuals: this is another thing that is on the table for these negotiations. Since the advent of streaming, residuals have all but dried up. The studios won’t release the numbers, and they have outright rejected the current WGA proposal. The language learning model thing is important here, but its important to note that this is only one of many things that are up for negotiation.

132

…I didn’t claim the source was “secret.” And that isn’t he basis of the Getty lawsuit.

Anyone and everyone aren’t free to use Getty images outside of the scope of the Getty licence.

133

Which in no way changes the fact that the data has been and continues to be free to download by anyone. My point was, StablityAI didn’t do the scraping, and nobody has to guess what images are in there because they can look for themselves.

134

…and none of this has anything to do with the lawsuit. Its a non sequitur. It’s the kind of thing that people think would be a useful argument in court, but more often than not the courts see right through it.

135

So Stability AI not doing the image scraping has nothing to do with Getty’s suit that Stability AI did the image scraping? It is straight from your link:

“Getty Images claims Stability AI ‘unlawfully’ scraped millions of images from its site.”

136

BTW, you.mentioned European laws. FWIW, LAION is German.

137

I read the whole article looking for that quote – “Did Darren actually read to the end of the article to find it?” I wondered–before going back and seeing that it was the tl:dr summary right at the top. It’s a poorly-written tl:dr, as you’ll note if you r. It doesn’t accurately reflect either the article nor the press release on which the article is predicated, where instead of saying, “scraped,” they say “copied and processed.” Although Stability AI might not have done the scraping, they almost certainly copied and processed the images.

If you thought that the lawsuit would fail on such trivial grounds, that’s a bit surprising.

138

…I’m not a lawyer, and I don’t pretend to be one, even on this site. When I said “I think we have to see what happens after this round of lawsuits play out” I meant it. Its just one lawsuit. There are others. And I’m a photographer. I’ve watched Getty win the most absurd lawsuits in my time. They wouldn’t be putting out statements (and to be clear you were quoting the headline, not the statement from Getty) like that if they didn’t think they had a case.

139

I wanna look a little more at this ChatGPT “joke.” This is a bit of a digression, so if you read it don’t say I didn’t warn you.

Anyway, it’s like a Viceroy butterfly. Viceroys look like Monarch butterflies, but only if you don’t look closely; and they’re completely non-poisonous, unlike monarchs. This looks like a joke, but only if you don’t look closely; and it’s completely non-funny, unlike jokes.

It looks like a joke because it follows a traditional riddle structure:

  1. It starts with a “Why” question.
  2. The subject and the verb are a strange juxtaposition that you wouldn’t encounter in everyday life.
  3. The punchline references elements from the subject and verb in a sentence that sounds reasonable.

It’s lacking any sort of wordplay in the answer that would make it funny, though; and ChatGPT doesn’t add wordplay because it doesn’t know what wordplay is. It doesn’t know anything, of course–this is just one subset of not knowing anything.

So I was trying to think how I would make a similar joke. I started by thinking, “How could a punchline about going on a diet involve wordplay?” and came up with the word “pounds,” which has multiple meanings. It can mean money, or it can mean hits. A punchline could be, “Because he/she/it wanted to shed some extra pounds,” as long as there’s a double meaning to that sentence. That’s close enough to ChatGPT’s attempt that I think it’s comparable.

My first attempt was something like, “Why did the British shop go on a diet?” But the punchline doesn’t work, because I know that shops don’t want to shed pounds, they want to gain them.

Then I thought about boxers. “Why did the beat-up British boxer want to go on a diet?” Again, though, I know that “shedding extra pounds” is a really bizarre way to describe not getting hit, so bizarre that most people wouldn’t get what it meant. So I rejected that as well.

What if I switched it around? Instead of “Why did X go on a diet?” I make it about a diet person wanting to do something else? I tried, “Why did the British diet guru buy something?” That works better, but it’s still pretty confusing: when you’re buying something, you do “shed pounds,” but it’s not what you want to do, so the punchline doesn’t really work.

Spending spree! That’s where you get rid of a lot of money quickly! And I finally got my joke:

Now we have a joke that has a question with an unusual juxtaposition of subject and verb, and a punchline that works for both aspects of the question, based on wordplay.

This is not a great joke. I don’t need you to tell me that; I know that. But it is a joke. It’s a monarch, not a viceroy. If it showed up on your Laffy Taffy wrapper, you wouldn’t be like, “What the fuck was that?”

What’s interesting to me is my thought process:

  • I had to break down what I knew about a specific kind of joke structure.
  • I had to search my brain for diet words that had multiple definitions.
  • I had to think about what I knew about those different meanings, and what I knew about the world (e.g., that shops don’t want to lose money).
  • I had to reflect on idiomatic usage of words and to predict how people would respond to a particular phrase.
  • I had to abandon a structure and reverse it.

In the end, I had a joke, and I can grade it as a C- or D+ joke.

That’s fundamentally different from how Large Language Models work.

It is possible that eventually we’ll get an AI that can make real jokes, not pretend jokes, but I don’t know how that will work, as long as the AI can’t engage in the sort of reflection that any human can engage in. Maybe brute force will eventually enable an AI to make real jokes; but barring a major change in process, it won’t ever be able to engage in reflection.

Incidentally, I propose we use the word “viceroys” to describe the sort of AI products that look like the real thing but are missing a critical element.

140

It puts me in mind of the first AI joke I ever read, “Why is a goldfish like a laser beam?”