Last week, I wrote an article partly defending the use of artificial intelligence (AI) in certain scenarios. This week, I want to talk about an AI use case that is not only clearly bad, but actively repugnant to me: AI-generated videos.

For context, on Oct. 1, OpenAI released a new social media app called Sora, powered by their new video generation model, Sora 2. This model is unique for being significantly more realistic than earlier versions and can generate audio with video. The app is similar to TikTok, except that it allows users to generate new short-form vertical videos using the video model. It also has a feature called “Cameos,” where you can create videos of yourself or of your friends. OpenAI currently requires an invite code in order to join, as they are trying to keep the amount of computing power required to generate these videos in check.

There are a few dimensions along which I think this app is terrible. First, I would say that I am broadly against short-form vertical video platforms with an infinite algorithmic feed. I think they are highly addictive and not a great way to spend time. Almost everyone I know who uses or has used these platforms, including myself, has expressed that they feel like they are wasting time while using them and that they have trouble stopping. This means that even when ignoring the ethical concerns, OpenAI didn’t need to go out and make another app to try to suck up all our attention.

Putting aside the problematic nature of this sort of social platform, I think the AI-generated nature makes it even worse. Sora’s AI capabilities introduce a dystopian vision of a world where algorithms tune AI videos to create the most perfectly stimulating and engaging content possible for a specific user in real time. This is not what Sora is doing at the moment, as all the videos are made from human-generated prompts, but I don’t think it’s too much of a leap to see how we could get there. To his credit, Sam Altman, the CEO of OpenAI, recognizes this risk, explicitly saying in his blog post about the product, “It is easy to imagine the degenerate case of AI video generation that ends up with us all being sucked into an RL-optimized slop feed.” He then goes on to talk about the care that they are putting into ensuring that does not happen. I am glad that Altman seems cognizant of the risks, but I guess I just do not see the upside of even the most mindfully crafted AI content. Even if they avoid the worst-case scenario, I have trouble seeing how Sora could truly enrich someone’s life.

The other big problem with the AI generation is the tremendous amount of computing power and electricity it costs. First, this has a bad impact on the environment. Video generation is vastly more compute and electricity-intensive than text generation, and I would argue its possible benefits are dubious. It’s clear to me how large language models (LLMs) can be used to accomplish real work and add actual value to the world, for instance, through writing code. The people who use LLMs to write code are likely to report that their lives have been in some way improved by using the LLM in their work. However, with video, it’s much less clear what the benefits are. I don’t think we need to or should be harming the environment and accelerating climate change to pump out more and more synthetic AI slop videos.

I don’t think Sora makes much sense, even from the perspective of the AI believer. Just a week before Sora came out, Sam Altman wrote a blog post in response to Nvidia’s investment in OpenAI, which OpenAI was using to purchase more graphics processing units to build more datacenters. In the post, Altman claims that we are close to major breakthroughs in AI, citing that we may be able to cure cancer or provide customized tutoring to everyone on Earth if AI companies have enough computing power; because of this, Altman says there’s an imperative to build more computing power so that we don’t have to choose between those goals. If Altman really believed we were on the cusp of using AI to cure cancer, I don’t understand why he’d allocate a bunch of computing resources to the AI slop app.

I should be clear: I have used the app and watched a fair number of videos. I am not worried because the videos are bad. I am worried because at least some of them were good. By “good,” I mean engaging, interesting, funny or realistic. For all intents and purposes, OpenAI has succeeded in creating a machine that can take a text description and generate a TikTok video matching that description. This is a major technical advancement, which I think is to be commended, but I am worried precisely because I think these AI videos have the possibility of outcompeting human-made content. The thing that is so addictive about platforms like TikTok is the endless novelty. Sora dangerously expands the novelty horizon. Every idea can get turned into reality, almost instantly. The AI slop is threatening precisely because it’s good.

I don’t know where things go from here. OpenAI is not the only company working on this technology. Meta launched a similar social media product called “Vibes” around the same time as Sora, and Google is working on its own series of AI video generation models called “Veo.” It seems like the cat is out of the bag. The current versions of these models are the worst they’ll ever be, and right now, I think they’re good enough to cause major societal impact. I haven’t even touched on the concerns about the necessary loss of trust in video content that comes along with models like this, which I think are substantial. It is an open question how much longer video evidence will be admissible in courts, for instance. I also haven’t touched on the copyright violations that ran rampant on Sora the first day it was released. We will have a lot of issues to sort through with this technology.  I just hope that we can keep our priorities straight and remember the value of human creativity.