Art need not worry

With Sora 2 recently released, YouTuber Casey Neistat posted a video “SORA: the all Ai TikTok Clone. will slop end creativity?”¹ in which he observes a trend:

1. Initially, movies required a lot of money, equipment, people, etc., leading to relatively few movies being created.
2. As production got cheaper and people became more talented, we saw a rise in media production with more movies and cable TV.
3. Then in the internet age, services like YouTube and more recently TikTok have allowed anyone to publish their own videos with just a digital or phone camera.
4. Finally, last week, OpenAI released Sora 2, which is both an improved video generation model and a social media. It allows anyone to make a video, even if they don’t have a good camera, CGI or video editing software, or any skills required for an interesting video besides an interesting idea.

In the first three stages, decreased barriers to publishing video have led to more content being produced. With plastic funnels, he explains that with more content being produced, there’s also more good-quality content being produced, even if most of the content is poor quality. Therefore, with video generation models allowing for the mass production of content—AI slop—he hints at a potential expectation: an increase in AI slop could lead to an increase in good-quality content, the majority of which would be AI-generated. But reasonably, he finds that expectation dubious. He concludes, “are we starting to turn a corner where slop starts to overwhelm or displace actual creativity?”

I agree that it seems doubtful that AI-generated slop would constitute a large portion of good quality media in the future, but I disagree with his wording, which implies that AI-generated slop and “actual creativity” are necessarily disjoint.

Why wouldn’t AI overwhelm the content we consume?

We still watch movies and TV shows, and we hold them to a higher standard than videos posted on social media, but the majority of content we consume is undeniably from social media. Should the trend above continue, that would imply that in the future, the majority of content we consume will be AI-generated.

Yet for some reason, I really doubt this, but it’s not obvious why. After all, AI slop has already begun to creep into our feeds: Italian brainrot was likely the first major instance of AI slop competing with traditional content. Content does not have to be real or authentic for us to enjoy it, especially if it’s not implied that the situation is real. Books, plays, and movies have long often been about fictional tales—“staged,” if you will—and we enjoy these stories for what they are, as an escape from reality. People dislike even the sight of AI, but that might just be a natural human reaction to something new and unfamiliar. Photography was initially rejected as a form of art, so one could argue that AI-generated content is just another medium too new to be accepted as art.

I think that this sense of doubt is because we made one assumption about AI in the trend above: that AI will allow for creating more content than ever before. Is that really true?

Cameras produce more slop than AI ever will

The thing is, AI is really just a step back from the content creation processes we have now. If we focus on cameras as an example, cameras retain two strengths over AI:

1. Cameras “generate” a wider variety of content.
2. Cameras generate content faster than generation models.

By nature of how generation models work, everything they generate is limited by the training data they consume. AI is just an imitation of reality; it can never surpass it. Without further prompting from the human to recombine existing concepts, it will never generate something that it has never seen before. Even if there were some artificial superintelligence that could live the world the way we do, observing and learning from every moment and experience to continuously train its model, the diversity of content it generates is always at most on par with what cameras can capture. Cameras will always be one step ahead of AI.

Secondly, while people may not see it this way, cameras are the true slop machine. Cameras are everywhere around the world, recording content with little to no artistic value—think surveillance footage and Zoom recordings. They’re slop: they’re easy to create in mass, have little human touch, and in turn, the content collectively looks like gruel. Generation models need to burn countless watt-hours of energy, crunching numbers on expensive hardware, just to generate one video. In contrast, cameras are now mass-produced, comparatively more energy efficient, and can produce content in real time.

Of course, these video recordings aren’t what people think of when they contrast AI with human-created content because that’s not what they see in their feeds. That makes sense. I’ve been continuously pumping random videos from my phone—slop—onto my YouTube channel as video storage, and they (fairly) receive only a few dozen views at most. But videos and other works of art with few “views” (or the equivalent measure) exist everywhere, as off-tune humming lost into the moment, graffiti etched on a desk, or a conversation on a long forgotten Discord server. Slop has always existed.

AI slop faces the same challenges as camera slop

I take issue with Casey framing AI slop as not creative. Like cameras, AI is just a tool, contrary to what AI companies want their investors to think.

It is just as easy to film a river minding its own business as it is to generate a video of a river. Of course, you’d need to be physically near a river to film it, and you’d need to wait a few minutes for servers to generate your video. But the end result is mostly the same.

Now you have a video of a river, perhaps fake, perhaps real. You can upload it to any social media, but if you don’t already have a large following or aren’t randomly selected by the algorithm, it’s unlikely the video will see any views. After all, who cares? It’s just a river. Even if it’s a famous river, countless tourists probably have already filmed the same river and posted it to social media the same way you did. A video of a river on its own is slop.

In order for slop to become good quality content, you’d need to put more into it. For example, some options include:

• Curation. You could’ve taken many videos of the same river, but people would rather see the best shot.
• A personal connection to your audience. Your family members will be more interested in your videos of rivers than strangers on the internet, even if they’re AI generated. People are more likely to watch videos from creators they recognize and movies from actors and directors they like.
• A story. Even the cheapest videos on YouTube Shorts etc. are just a person retelling a story from Wikipedia, and they will receive many more views than a video of the river.
• Editing. You’d need a video editor, and to make something as uninteresting as a river into an interesting video, the art ends up being in the editing rather than the river itself.
• Originality. The river suddenly turning into lava would be vastly more interesting than a normal river.

For the last point, it would be incredibly easy to create that type of video with AI. The standards of uniqueness are different for AI because by making slop generation so accessible, you’d need to find new ways to stand out. I’d imagine for Sora 2, this might take the form of finding ways to circumvent OpenAI’s guidelines, for example, and I’d argue that that would be a form of creative expression.

Just because AI makes it so easy to churn out slop does not necessarily mean it will lead to an explosion in AI-generated content that will overwhelm the human content we enjoy today. Having to make AI slop stand out above the rest makes it share many of the existing problems we have today with getting views on social media. After all, as Casey points out, one barrier for video production is distribution, and Sora 2, as yet another social media, does not lower that barrier.

AI will still affect traditional art

Since AI is not going to take over the world, the traditional forms of art we enjoy today should be completely unaffected, right? Of course not.

While we often like to see a human touch in the works of art we see and hear, that’s not true all the time. For example, when you enter a hotel room, you usually do not care too much about the paintings in the room, who painted them, what they depict, the artist’s message. While the artist may have painted their work as art, a hotel guest staying in the room will not interact with the painting as art. These situations, where art is used solely for its medium, such as interior decoration or background music at a restaurant, would be the perfect victims of AI-generated media. After all, even humans already create slop: take the corporate artstyle and tutorial background music as examples. Therefore, I believe that any artists who rely on making a living on producing art for these audiences will likely see commissions drop as businesses cheap out and switch to using AI.

That said, art is never going away. Art has historically always been a symptom of a thriving society, when some of its population no longer need to worry about staying alive. The ways people circumvent censorship, like with “unalive” and 🍉, are a testament to how humans will always find a way to express themselves, and generative AI will not take that away from us.