On December 14, 2022, a new season of Seinfeld debuted on the streaming site Twitch, airing continuously, 24 hours a day, every day for a few months. Ok, not really. The AI generated Seinfeld parody “Nothing, Forever,” did run 24 hours a day until February 6, 2023, but the show was completely procedurally generated via artificial intelligence. While the show was pretty well received, with even the official Seinfeld twitter account linking to the Twitch channel, “Nothing, Forever” opens up a series of new questions regarding the intersection of copyright law and artificial intelligence. Most importantly, does Seinfeld, Jerry, the actors, NBC, or any other person or entity with rights to Seinfeld the show have a copyright claim against “Nothing, Forever?”
What’s the deal with artificial intelligence anyway? As best understood by us pea-brains here at SeinfeldLaw, AI feeds on whatever information is available to it, let’s just round up to everything on the internet, and by some sort of mechanized learning system can then generate content based on the information it has gathered and processed. When a person asks AI to deliver standup in the style of Jerry Seinfeld, or create episodes of Seinfeld based on situation prompts given to it, the AI digests the relevant information within its network and generates content based on that information and what is asked of it. It’s sort of like that scene where Kramer mimics Jerry’s standup. Just like Kramer takes what he knows about Jerry’s comedy and mannerisms to put on an impression of him, the computer is simulating content based on the information it’s been given.
So what are the laws about artificial intelligence and copyrights? Well, it’s sort of an open questions. So much so, that The Congressional Research Service released a May 11, 2023 memo addressing this very topic to help lawmakers craft legislation to govern this new area of law. One of the biggest questions the memo discusses is whether AI infringes on existing copyrighted works by “generating outputs that resemble those existing works,” too closely.
To prove any copyright infringement, a plaintiff must show that the defendant 1) had access to the work and 2) created something that is “substantially similar.” Unlike with people, who may have created something totally independent of the copyrighted work even if it looks the same, artificial intelligence must be fed information in order to produce anything at all. Moreover, the machine will likely have some kind of record of the materials that it was trained on. So, proving prong 1 of the analysis for AI should be fairly straightforward by simply just figuring out the content that its network has access to.
But with regards to the second prong, what does “substantially similar” even mean?. Courts have generally given vague definitions, such as that the “concept and feel” or “overall look” must be substantially similar. In more legalese, other courts have said that “the ordinary reasonable person would fail to differentiate between the two works.” The “qualitative and quantitative significance” standard is a further prism by which to analyze. But what each of these analyses include, as we discuss in our piece on “The Soup Nazi,” is that there has to be something creative to be a copyright. Just an idea, or a style, isn’t going to be enough. In others you can’t copyright the idea of a nasally sounding comedian telling “you ever notice” jokes interspersed with situational comedy about the day to day goings on of four New Yorkers. Copyright infringement is only going to happen when, and this is where it gets tricky, it starts to look like the original, or feel like the original, or be close enough to the original where your average person might think that Jerry Seinfeld had something to do with “Nothing, Forever,” even though in reality he doesn’t.
So how likely is it that artificial intelligence will violate copyright law. The folks over at OpenAI say not too likely. “Well contrasted AI systems generally do not regenerate, in any nontrivial portion, unaltered date from any particular work in their training corpus.” Or in human terms, if it happened, which it probably won’t, it’s an accident, and we owe you one.
But a recently filed lawsuit filed against Stable Diffusion, an open source artificial intelligence that generates images from text prompts, claims that that just isn’t true. The suit alleges that all AI generated images are copyright infringements if the AI network includes copyrighted information, because it is just then regurgitating the copyrighted information into something else. Sure, it might look a little different, but it’s pistons are grounded in copyrighted work. It’s sort of like the argument between Jerry and Puddy about “the move.”
So where does this all leave us? As far as “Nothing, Forever,” is concerned, it’s unlikely the creators are liable for any copyright infringement. While the show bears some resemblance to Seinfeld, the names are different, the voices are completely distinct and the characters & set pieces look generic enough to be distinguished from their real world counterparts. Going forward though, as AI content becomes more fine tuned, both visually and audibly, it’s not hard to imagine a computer generating a Seinfeldian type show that looks and sounds indistinguishable from the original. Undoubtedly, Congress will craft new legislation to address this intersection of artificial intelligence and copyright law, and until then courts will have their dockets filled up interpreting how existing law applies to these new questions. For now, it’s a fascinating new area of law that we have only just barely scratched the surface of here.
*Editors Note: Right of Publicity laws are another angle in creators bringing potential legal suits for AI content that too closely resembles themselves