What Makes Something AI Slop?

There’s a new term you may have heard of when referring to some of the internet “content” that’s been posted lately. AI Slop. But what does it really mean? I think generally people use this term to describe poor quality content that comes from generative AI tools. AI Slop can be things like those images with “wrong” fingers and garbled text, nonsensical or erroneous answers to questions asked to LLMs, Generic soulless AI generated music, badly written code ridden with security flaws, etc. The list goes on. Now, obviously there are a lot of people using Generative AI tools to create art, music, code, written content, etc, who feel like the quality of what it puts out is adequate or even desirable. Because of this, the concept of AI slop is very polarizing. One person’s slop is another person’s masterpiece. And it also must be said that the concept of “slop” content on the internet didn’t exactly originate with AI. There’s been a growing proliferation of human generated slop content over the past few decades, although one could argue (it’s me, I’m one) that the slop created by human beings is still infinitely more interesting than what’s generated by AI because it still sometimes has some element of lived experience and creativity. A solid example of that to me would be shitposting. It’s truly an art when done tastefully.

The Arguments for the Use of Generative AI tools

So again, there are people on both sides of the fence who strongly believe the other side has completely wrong opinions about the use of AI to create. But for now, let’s focus on the people who feel like it’s a net positive. One of the main arguments I’ve seen for pro-AI usage is that it saves time and money. This frees up creators so that they can focus their efforts and resources on other aspects of creating or running their small business. For example, in the past, a musical artist used to be able to mainly focus on creating music, but when social media became the dominant medium for people to get their music out there, things shifted into requiring everyone to be “content creators”. So instead of just making and releasing music, an artist now has to be active on platforms like TikTok, making dance videos for every song, on X posting edgy content for clout, on Youtube demonstrating their recording and production processes… again, the list goes on. The big problem with all of this effort spent creating ephemeral content is that it diverts energy away from the thing you set out to do in the first place and as a consequence ends up diluting the final creative product. Of course, you can opt out of this social media content rat race, but you do so at the expense of visibility online. So, in comes AI to the rescue. It sounds almost Utopian to be able to going back to focusing on your main objective and letting AI do the rest. Just make a song then tell Midjourney or something to generate some cool album art and you’re done. No spending hours learning image editing software or graphic design skills. No spending the little money you have dealing with difficult to work with Graphic Designers. Life is Good, right? Well… not exactly. At least not yet. At the moment, anything generated by AI has this “look” to it that people can recognize. And unfortunately, this can signal to those people who notice it and care, that you’ve taken shortcuts on your art. Of course, some people don’t care at all or might even think it’s cool that you used it. But for the ones that do care, they can’t help but wonder… If you “cheaped out” on some parts of your product, what else did you cheap out on? The reality is that using generative AI can make you lose credibility as an artist from certain fans, but ultimately, it’s up to you to decide if you wanna cater to those people or not.

People Who Feel Like AI Allows Them to “Finally” Create

Another common argument I’ve seen from a lot of people is that they simply didn’t have the skill level or resources to create the things they wanted to until AI gave them that opportunity. There are also people with certain mobility or cognitive limitations due to disability who feel empowered to create the art they want now that they have access to Generative AI. It’s obviously a great thing to use technology to democratize the process of creating, so I don’t think anyone can really argue against that. But it has to be said that, what makes Art “good” or “interesting” isn’t necessarily how perfect it is. Actually, (at least in my opinion) the human error or randomness associated with Art is exactly what makes it tell a story. I’m sure generative AI will get better at embodying this over time, but at the moment, it outputs things that are very “fine tuned” to have a specific look or feel, which can easily come across as generic or uninteresting. Of course, there are many exceptions to this, especially concerning people who do work to train and tweak custom AI models and do more with the output than just using it as is. I find the technological artistic aspect of AI to be a lot like that of photography. When cameras were first introduced, I’m sure there were people who thought photography would “replace” painting, and in some ways it did for the sake of record keeping. But painting didn’t die… in a lot of ways it just got more and more interesting since artists could now think about new aspects of creating instead of making everything look “realistic”. Also, photography ended up becoming its own art form and I believe the same thing will happen with Generative AI when more artists figure out more novel ways to use it.

The Training Data Problem

So just one of the other elephants in the room is that AI has to be trained on specific data in order to be able to generate things that are similar. One recent example that was pretty big online is the generation of all those Studio Ghibili style images. They kinda flooded the internet with seemingly just about every other person’s profile pic becoming a Ghibili version of themselves. Now, not to ruin the fun… but for the AI tool that generated those images to be able to do that, it had to be trained on artwork created by real human Ghibili artists… that Studio Ghibili owns the rights too. One technique of AI image generation is called Neural Style Transfer. This is where you essentially have two images. One that you want to use as a base for the composition of the final piece and one that you want to take the style of to impart on that first image. An example would be taking a photograph and giving it the style of Van Gogh’s “Starry” night, so you can take any photo and make it look like a Van Gogh painting. Ok, there’s no denying that it’s pretty cool to be able to do that, but at the same time, it kinda sucks that the artist has no say in if you can just infinitely pump out images in a style that they created. Tools like Chat GPT and Midjourney use a more complex ensemble of methods than just Neural Style Transfer, but the base technology still needs input from images to “learn” from and it’s definitely a moral and legal grey area to use artists works to train an algorithm without their consent.

The Dead Internet Theory

One last thing we have to mention in regard to AI slop is the fact that making it easier to generate content also means there will probably be more content generated. The problem with this is it’s probably going to get even harder for human made content to get seen if the internet is flooded with an infinite supply of content that takes little to no effort to create. We’re kind of entering into this feedback loop where platforms have these algorithms that identify and boost the most engaged with content, and by knowing what type of content get’s the most engagement, people can set up bots that just create this kind of content on autopilot. Of course, no one knows what the future looks like (especially because humans are so fickle) but we’re definitely moving in the direction of losing a human connection to content posted online. At this point, what is and isn’t AI is still discernible to some degree, but that’s changing fast. Soon, it might not be possible to tell what’s real or not, what’s human created and what was generated by AI, so what will content posted online even mean? I guess we’re gonna find out soon. It’s going to be an interesting next few years.