The Slop Singularity

When fake approaches the limits of reality

Most people who are critical of AI are concerned with immediate, first-order effects, like a personal loss of their own economic bargaining power, rather than what generative AI will do to the systems we collectively understand to be the internet. Those second-order effects are more profound than any fear of competition, because AI will not succeed due to quality, but quantity.

The Slop Singularity approaches.

As long as it is cheaper to make AI content than real content (relative to expected returns), users will make it constantly and without limit.

As time goes on, the total amount of AI-generated content on the internet will approach 100 percent of the whole.

If it takes three hours to make a real video, but only three minutes to generate a fake one, users will make infinite fake ones, even if they are terrible in quality, and get only a fraction of the views of the real thing. Real content would have to be orders of magnitude more viewed to make AI “not worth it.” And since the algorithm only cares about engagement, it will serve many AI videos at random to viewers. And, of course, as time goes on, the real piece of content, even if much better, gets fewer views because there are 10,000 other pieces of AI content (and growing) to show the viewers that might be interested in the original, real video.

AI content need not replace the human-made; it needs merely to bury it. Since the trend for technology in general is to get cheaper and better with time, you can expect the (at least hopeful) profitability of AI content to remain, even if marginal, as the slop becomes overwhelming and each piece of content gets even more worthless. Keep in mind that while the creative among you can think of many ways AI could be used to make good art or improve workflow, that is not what AI bros have used it for. As with the nature of what is easy, the result is the swiftest path, the path of volume. Using AI a little to “speed up” certain things will be the same result as traditional art in that you cannot keep up with the volume of careless crap. You must join the slop stampede or be trampled.

You can add to this certain practices of tech companies, such as summarizing searches using AI so that individual websites lose all their traffic, and the end result is an internet where it is impossible to be real, impossible to find other real people, and impossible to meet costs when posting real, human-made content.

The internet will be dead because no living person will be on it.

The destruction of the internet attention economy is only half the problem. The other half is that this singularity of slop will degrade the LLMs themselves through recursion. Last I heard, an estimated 50% of webpages are now generated by LLMs. If Gemeni summarizes a web search for you, and that search includes summaries by AI, then that information will contain some amount of “hallucinations” – the noise that results from the extrapolation of limited data. Those hallucinations then become part of the next round of data, and over time, it becomes impossible to determine what is real and what is not. Any fact or event could have been conjured up by an AI, then “calcified” by more content being generated with referencing hallucination.

Even if a user knows this, think of the work he has to do to verify or falsify any piece of information generated by a Google search. Every article could be generated by AI based on other AI, which hallucinated fake citations and got published in a real journal because the humans relied on AI to verify things…You get how it goes.

Search will be dead.

Social media will be useless.

What happens next?

The future is always uncertain, but to go forward, we will have to go back. That means going directly to individual websites, generally recommended by other trusted individuals and groups. Private message boards, moderated by humans and full of people with long-standing cred, will have to come back from the alpha version of the internet to be the main method of real interaction between serious humans looking to learn new things.

In the arts, we’ll all have to be jazz musicians, in a way. Creating something live and in person, and sharing that moment with other people, will be the way to ensure that music is not only human but has human meaning, since it will be a shared social experience. The perfect studio albums of the 2000s and onward will have to take a back seat to imperfect, but far more unique, live experiences. Improvised music has been my own focus since 2023.

Visual artists will have to cultivate their own networks, but may find that the physical product, the real drawing or painting on real paper or canvas, will be the only product discerning people are willing to pay money for. Theater may make a comeback as regular people quickly become tired of always wondering whether something is real or not.

As for literature…I don’t have much to say. I wrote a book live. It wasn’t really a popular series, but viewers had asked me to do it for years, and I did it. A book, unlike other pieces of art, is generally a solitary experience and, by design, comes in a replicable form. The fact is, we will desperately need publishers to do the vetting for us in the coming years. We can no longer rely on Amazon to feed us the books we want. The problem is that traditional publishing is not in a good state artistically or ideologically. But that could mean room for new players to get their real, human-made, and original art into physical bookstores, which themselves will be the first filter for normal readers. If you can get it in a store, it’s got a very good chance of being real, rather than something crapped out by an LLM for a (hopefully) quick buck.

There is also the possibility of the Slop Singularity being slowed by costs, not of the generative software per se, but the tech companies’ ability to host the infinite deluge of crap. YouTube allows users to upload for free. That might go away because it’s not economical to host and deliver a 4k video to the tiny number of viewers that might look at each piece of slop. YouTube will have to create an unfortunate buy-in for content makers in the form of hosting fees for each video. The ability for a nobody to throw up a viral video will be diminished, but I don’t see a way to avoid the economic realities of the situation. Websites also generally cost money to host, so there may be similar disincentives as the net as a whole becomes saturated.

Or media sites could choose to vertically integrate and cut out the creators (or prompters) entirely. Spotify and Amazon are already testing this, as I understand. YouTube could as well. Perhaps they will generate the AI for the user based on his tastes directly, then the people scrambling to flood the sites with slop will be overwhelmed; no amount of volume will be enough, and the entire AI jig will be up.

But, by that time, the damage might be done. We’ll have a collapse. Maybe it will be a good thing, and you can catch me playing some improvised music for a real audience at the local café or park.

I am an independent artist and musician. You can get my books by joining my Patreon, and you can listen to my current music on YouTube or buy my albums at BandCamp.