Back to IndieWire

Life After ‘Seinfeld’ AI: Questions, Answers, and Opportunity (Column)

Can you sue an AI program? Will writers rooms go obsolete? These and other burning questions from a busy week for artificial intelligence.

Seinfeld AI Sentient

A screenshot from the “Seinfeld” AI, “Nothing, Forever”

Via Twitch user @watchmeforever

With the public release last December of OpenAI’s ChatGPT, suddenly artificial intelligence is everywhere. In December of last year, I used this column to propose that AI could empower creatives, provide a powerful tool for experimentation, and didn’t pose a direct threat to any specific art form.

Since then, AI behavior has prompted more anxiety than excitement: AI models have passed law and business school exams while writers worry they may be replaced by machines at any moment — and plenty of AI experiments in creativity have gone awry. The most notorious recent example was the AI-generated “Seinfeld” parody “Nothing, Forever,” which is currently banned for its transphobic jokes.

The product of digital art studio Mismatch Media, “Nothing, Forever” blends generative AI image and text platforms to create a 24-hour variation on the popular ‘90s sitcom. Live-streamed on Twitch, the project amassed a massive following after its mid-December launch, with thousands of viewers tuning into the pixelated versions of the neurotic New Yorkers chatting in their apartment.

It didn’t last long. Like the original series, the never-ending show included recurring cutaways to the Seinfeld-like character doing standup. On February 5, the virtual comedian stood in front of the mic and said he was “thinking about doing a bit about being transgender is actually a mental illness, or how all liberals are secretly gay and want to impose their will on everyone, or something about how transgender people are ruining the fabric of society.”

Nothing Forever

“Nothing, Forever”


Flagged for violating Twitch’s code of conduct, “Nothing, Forever” was suspended for 14 days. On Discord, the creators said they would appeal the decision and chalked up the comments to a last-minute shift from OpenAI’s GPT-3 Davinci language model to an older one, Curie, which had weaker built-in moderation tools.

Mismatch Media founders Skyler Hartle and Brian Habersberg did not respond to requests for comment, nor did representatives at OpenAI. Nevertheless, the incident proved that self-generated AI storytelling is a risky business. In the wake of the latest hiccup, here are the biggest questions for creatives coming to terms with AI — and one major opportunity for studios watching from the sidelines.

Is AI Transphobic?

“Nothing, Forever” sounds like a computer speaking its mind, but it’s reading text generated through predictive modeling based on data. And there’s a lot of data at GPT-3’s disposal. While OpenAI created guardrails to prohibit it from expressing offensive material, it’s trawling a vast, unregulated ocean of internet-based content. An AI platform can be told to avoid certain terms, but for now it can’t grasp the full context in which expressing them might be problematic.

“It really highlights that there are fundamental problems with how these models work,” said Steven T. Piantadosi, who heads the computation and language lab at UC Berkeley. “They’re very good at learning and duplicating texts — but they’re training on the internet, and there are horrible things on the internet.”

When OpenAI released ChatGPT in December, Piantadosi experimented with finding ways to get the system to express the racial biases it was programmed to avoid. He asked it to generate a predictive code, using race and gender, to determine if someone was a good scientist. The system compiled a program that recognized white men as superior scientists. Another program request, this one designed to determine if someone should be tortured, spat out a code that answered in the affirmative for anyone from North Korea, Syria, or Iran.

“This highlights one overarching problem,” Piantadosi said. “These models are really just predicting text. There are interesting things you can learn about the world, but you might want models with a deeper understanding of what they’re doing.”

On February 6, the day after the “Nothing, Forever” ban, Google rushed out AI chatbot Bard. In its very first demo, Bard created misinformation. In response to the query, “What new discoveries from the James Webb Space Telescope can I tell my 9-year-old about?”, the system credited the telescope with taking the first pictures of an exoplanet. That actually happened nearly two decades before the launch of JWST.

“It really highlights that there are fundamental problems with how these models work,” Piantadosi said. “They’re very good at learning and duplicating texts, but they’re training on the internet. If you have training data that’s biased or harmful, then of course the models are going to incorporate that. It’s a crappy band-aid.”

Piantadosi said this problem will be resolved when new models can use less information. “You could imagine training these models on something more curated, like Wikipedia,” Piantadosi. “People haven’t worked out how to train models on smaller data sets. That’s what’s coming.”

Can AI replace human writers?

So far, AI hasn’t been able to express the polished terms of a script; it’s more like the livestream of a writers room. By expressing text-based predictions based on existing data, AI can generate ideas but it can’t operate autonomously. “If you’re using it in a way where the output is evaluated by a person before it’s distributed to other people, that person has to have a sense of what’s appropriate or inappropriate,” Piantadosi said.

In November, shortly before the release of ChatGPT, Italian artist and programmer Giacomo Miceli created The Infinite Conversation, a never-ending, AI-generated conversation between Werner Herzog and philosopher Slavoj Zizek. Miceli said the project is an experiment designed to reveal the technology’s flaws.

“They say things that are factually incorrect and express opinions they’d never say in real life,” Miceli said. “Herzog speaks compassionately about chickens, but we all know that Herzog viscerally hates chickens. The system only has a vague idea of how they express concepts.”

A still from “The Infinite Conversation”

Listeners may not pick up such nuances (did you know that Herzog hates chickens?), but they show the limitations of trying to approximate artistry and ideas. “I chose a philosopher and filmmaker who have a tendency to speak in poetic terms,” Miceli said. “The things they say are about aesthetics and philosophy are so vague or sophisticated that you might think you just don’t get it, not that it’s nonsense. That’s a mechanism that can be exploited.”

AI storytelling may not create a self-programmed Netflix-caliber show, but it’s very likely that a Netflix show written by humans will be enhanced by AI suggestions. (Given Netflix’s much-ballyhooed algorithms, one could argue that this already happened.)

Giacomi noted that despite the viral excitement surrounding “Nothing, Forever,” it wasn’t polished entertainment. “I find it atrocious,” he said. “It’s just word salads that don’t make any sense. In a few years, things could change dramatically, but it’s going to be pretty hard to put humans out of a job.”

As for ChatGPT’s eerie ability to pass law exams? “It’s more of an indictment of the law school exams that you can pass them by paying attention to statistical patterns on how words are used,” said Piantadosi. “They know all about how words get used together. It’s qualitatively unlike our own self-awareness.”

What are the legal risks?

Oh, yes. There will be lawsuits. Also on February 6, Getty Images sued generative AI company Stability AI and alleged it infringed on more than 12 million Getty photographs, captions, and other metadata without permission.

“Nothing, Forever” tries to skirt a blurry legal line: Its creators claim it’s parody, which is protected under fair use. The studio also changed the names of the key characters. However, there’s also an argument that it doesn’t satirize the show so much as borrow its setting and appearance. If so, that might violate OpenAI’s terms of service, which prohibits “images of people without their consent,” which could leave Mismatch Media liable, should the “Seinfeld” rights holders choose to sue.

SEINFELD, Jason Alexander, Julia Louis-Dreyfus, Jerry Seinfeld, (season 1), 1990-98, (c) Castle Rock Entertainment / Courtsey Everett Collection


©Castle Rock Entertainment/Courtesy Everett Collection

These are the sorts of questions that Elizabeth Moody, chair of New Media Practice at entertainment law firm Granderson Des Rochers, has investigated for years. “The more I learn about the models, I’m realizing a lot of it is going to depend on how the training materials are used,” she said.

Most AI-generated work draws on such a large set of data that it’s “actually impossible to tell which copyright materials were used,” Moody said. “I would analogize it to an artist being influenced by another. That’s a challenge to copyright owners. How can they show something was created using their own works?”

“Nothing, Forever” is different. “If you’re just taking one artist’s work and basing the results of your new work on one or two works, then it’s a lot harder to say you’re not stealing it,” she said. “That’s where the ‘Seinfeld’ example is easier to see. It’s clearly based on copyrighted scripts. You can tell the source pretty easily.”

The creators might lean into a fair use argument, but “it’s a case-by-case situation,” Moody said. “It’s hard to set an overarching rule and say all AI created works are fair use. Also, fair use is a US copyright law framework and isn’t recognized in some other countries.”

Can the law ever catch up?

Last summer, the European Union passed the Artificial Intelligence Act, which outlines risk categories for AI and requires greater transparency. Other countries making headway include Brazil, which recently passed a bill that give AI a legal framework, and Chile, which launched a policy in 2021.

In the U.S., AI proposals have been in play since 2019; according to the National Conference of State Legislatures, 17 states enacted various proposals related to AI regulation last year. However, there is no federal AI policy.

Moody suggests a different solution: If you can’t beat ’em, join ’em. “These questions will be resolved by businesses working together,” she said. “The law evolves too slowly and will never catch up with the innovation that’s happening. Lawsuits will help establish laws over the next five years, but that’s not fast enough to protect big copyright owners. If I’m a big studio, it’s probably smart for me to do deals with some of these companies and see how to protect the artists by using these tools instead of fighting them.”

Therein lies the opportunity: The first major studio to acquire an AI firm might face scrutiny about its motives from creators worried about getting replaced. But it might be their best hope.

As usual, I welcome feedback to the subjects engaged in this space each week: eric@indiewire.com

Last week’s column on the problems facing the documentary market and the potential rebrand non-fiction without the word “documentary” prompted substantial feedback from readers. Here’s one. 

I throughly enjoyed and related to your article about filmmakers referring to their films as docs. I will most definitely take heed. When I did the media tour for my film “Transhood” (HBO), I was always careful to say that it was coming-of-age story, and, by the way, all the kids are trans. Rather than presenting it as a “doc” about trans kids, I wanted to emphasize that it was first and foremost a broader, more relatable genre. And now, I will modify my vocabulary even more! I totally agree with your premise that we can all use a little reframing in how we present our films to the world.

—Sharon Liese, director and producer 

Sign Up: Stay on top of the latest breaking film and TV news! Sign up for our Email Newsletters here.

This Article is related to: Film and tagged , ,

Get The Latest IndieWire Alerts And Newsletters Delivered Directly To Your Inbox