Let’s talk about how generative AI fits into our vision.
In short: it doesn’t.
A recent Proof News investigation revealed that Nebula creators’ data has been used without permission to train generative AI models for Apple, NVIDIA, and others. I’m quoted in the article:
“It’s theft,” said Dave Wiskus, the CEO of Nebula, a streaming service partially owned by its creators, some of whom have had their work taken from YouTube to train AI.
“Will this be used to exploit and harm artists? Yes, absolutely,” Wiskus said.
One popular defense of this theft is that the data used is merely the caption data for these videos, as if to agree that stealing a video is wrong but written work is fair game. This is misguided at best and evil at worst, but the most insidious part is that because the data is text, it can be used to generate more than just videos. Our work is being weaponized to harm artists across multiple disciplines.
We also know that consumers are starting to reject AI slop. A recent study at Washington State University showed that AI is a net negative:
“When AI is mentioned, it tends to lower emotional trust, which in turn decreases purchase intentions.”
So if creators don’t like it and consumers don’t like it, why is it everywhere?
Like cryptocurrency before it, generative AI is a tech industry buzz term that dials up the heart rates of venture capitalists and grifters alike. These plagiarism machines have spawned an entire cottage industry powered by stolen goods and operated by talentless hacks. Billions of dollars have been spent on technology designed to steal from creators, replace us, and use our work to replace artists in other fields. Even now, they openly admit that their businesses cannot function without theft. This isn’t okay, and Nebula can’t be a part of it.
Our official position: Nebula will not use any generative AI tools known to be trained on unethically sourced data in the creation of our original content or in the production of our software platform. While we cannot dictate what our creators do, use of such tools will be a factor in how we select creators in the future.
Like so many things, this requires some level of nuance. “Generative AI” can be understood in casual use, but software using artificial intelligence features — even those powered by the large language model (LLM) technology that drives generative AI — is getting harder to avoid.
Moreover, not all of it is evil, and our goal isn’t to simply rage against literal machines. Predictive text and autocorrect are powered by AI. Our automated closed captioning system is powered by AI. While not as accurate as human-written captions like the ones on Nebula Originals or provided by creators, these automated captions provide a level of accessibility on catalog videos that we wouldn’t be able to provide otherwise — a net positive while we work toward a better system. And, notably, the tool only transcribes words that were already written by humans.
There are also interesting creative applications. I’m aware of at least one major studio using generative AI to de-age an actor in an upcoming project. Assuming the photos they used to train the model were ethically sourced, that sounds pretty cool.
So where’s the line?
There are three considerations we make when we look at AI tools:
1. Is this tool powered by data sourced without the consent of the artists or creators of the original work?
It might be hard to know what powers a given system. Apple says that their upcoming “Apple Intelligence” consumer tools don’t use unethically-sourced data; the creator data mentioned in the Proof News investigation was only used for research tools. Trust them or not, Apple has a good record for data privacy, and making a public statement about it at all is unusual. I’m inclined to believe them, and I hope more tech giants start making similar statements.
2. Is the tool intended to replace human creativity?
A good litmus test here: is the tool being used to add things, or is it being used to remove things? Photoshop’s “content-aware fill” is an AI tool. Apple’s new Photos app lets you easily remove people or objects from photos. This feels fundamentally different from typing in a prompt and letting the machine imagine it for you.
3. Does this tool make us better, or does it just make us lazy?
Doing things faster is only useful if what gets done is just as good as if we had taken our time. While our projects often have deadlines, this isn’t a race. We’re here to make great things that wouldn’t exist without us. We’re here to help creators express their creativity. Saving time by churning out AI slop doesn’t serve any goal we have.
It should act as a signal to the world that over the last year multiple unions have gone on strikes predicated primarily on the issue of generative AI being used to either steal from or replace creative workers. While it’s easy to see that Nebula is made up of creators who would logically be concerned about our work being stolen, it’s also important to consider the perspective of our employees. Developers, designers, artists, editors, composers — creative professionals whose jobs are often the target of these tools.
Nebula is a place for stories and projects created by humans. We want those humans to feel secure that we aren’t looking to replace them with robots or sell their work to the plagiarism factory. Their voices will be included in the ongoing, evolving conversation as we define specific policies, guidelines, and approval processes to hold ourselves accountable to one another and to our audience.
ChatGPT will not be invited to the conversation.