The Slop Machine (Essay) (Patreon)

I’ve always been a storyteller. My first incomplete works date from middle school, and I have a number of short stories and poems I wrote in high school that I actually finished. The drive to write and create has always been there for me. Because of this, I wanted to go to college and study creative writing, but that’s not what I studied in school. Instead, heeding the practical advice I’d never make a living at it, and thus embracing a lifetime of regret that I wasn’t following my dreams, I studied computers, the other thing I liked in high school.

I have since come to terms with and found a way to manage my regrets that I didn’t study writing in college, but I always considered the two paths divergent. One was focusing on fantastical creations ungrounded by reality while the other was grounded in logical calculation. Coding is a world of aligning inputs and data so they are ordered and layered in a logical structure that lets you complete the task at hand. Writing too has its logic and conventions, but a good writer knows when to divert from the expected, subvert expectations, and delight the reader in a way they weren’t expecting. In my best writing, I surprise myself with what really ends up happening.

These worlds have, in the last few years, suddenly converged in a way that takes language and subverts it not toward the end of telling stories, but toward the end of performing calculations in a way that moves from logical and precision towards fantasy fiction. Large Language Models trained on huge data sets can approximate amazing results based on almost input you give them. I use the word approximate, because that is exactly what these systems are doing. Accuracy has never been guaranteed with these models, and research suggests that this technique may never give accurate results. To get accurate results, you need to train an expert system, and that is a completely different but related type of AI that requires far more money and work to be put into it to create true artificial general intelligence. So far, no one has achieved this, and many researchers are not sure we would ever be able to do that.

There are many types of calculations in mathematics and engineering where a sufficiently accurate approximate result is sufficient for practical applications. The people who are promoting Generative AI are absolutely pretending that the approximations created by it are good enough for everyone, and they’ve released it to the world without even telling most people that the results are approximations of real answers to the questions being asked of the AI. The consequence of such fantastical use of so much computer power on the public is a series of increasingly dangerous situations.

The most useful case for Gen. AI, and I say this with great sarcasm, is when you already know half the answer. There, you can much more easily evaluate the result and determine if it’s correct or not. However, why the fuck do I want a machine that is only really useful if I know half the answer to the questions I’m asking it? What is the fucking point? What happens if I don’t know the answer at all? It’s going to confidently tell me that’s correct, and who knows if that’s good or not? And if I blindly trust the AI, well it might just poison you. If you think things have improved since 2023 when that article was posted, they’ve not. AI results don’t look better when you get into health information.

Now, AI proponents will of course think I’m just talking surface level nonsense, that my tech skills are out of date, and that I’m just being a hater. “You can trust the AI!” they’ll exclaim gleefully. Except, if you go to professional conferences, like I did back in November, where they tell software developers how great these tools are, they also tell you how important it is you verify the outputs generated from AI. I sat through an eight-hour workshop about agentic AI agents in the software development world, and one of the key points brought up repeatedly is you need to verify what it’s doing. You turn this shit loose on your code base and just accept its inputs without question, you’re going to end up with garbage.

This slop is flooding the internet, to the point it’s increasingly harder to know what is real and useful without sitting down and actually dissecting what you’re seeing and reading carefully. That’s not saving people time in creating things, it’s just moving the burden of verification from the creator to the consumer. Yes, ethical creators are still going to spend time researching and making good stuff, but now we have to assume that most things we encounter, if we don’t know who created it , are suspect. Media literacy is incredibly important, and instead of finding a way to build a better world where you don’t need to always be on guard, we’ve created slop. Worse, big companies are falling all over themselves to get us this slop faster and easier, because that’s what the people who own the companies that create the slop think we all want. That’s certainly what they economically need to have happen to recover the massive investments in this slop they’ve made.

They’ve managed to bring fantastical creations ungrounded in reality to computer science and the masses. The people peddling this AI slop aren’t telling people what these approximations of reality really are. Worse, by the very nature of how they work, the results of any AI query is nondeterministic, meaning it will generate different answers to the same exact question. Adding guardrails will never eliminate this problem.

Do I want to engage with it? No! Do I want to use it at work? No! Do I want to use it in my writing? Absolutely fucking not. I like writing because it’s a fun fantasy I get to create, and I absolutely want no fucking slop in the process. I want this shit as far away from me as I can get it, and if I could bury it in the backyard, I would. I like having my own thoughts and if that requires me staring at a blank page, I much prefer doing that than having it butting into my creative process telling me what I could type next. The people trying to sell us AI products are trying to constantly give us tools that put the AI right there front and center on the blank page so we don’t have to think. It’s almost as if the people marketing these products are afraid we’ll think for ourselves.

Well, I like thinking for myself, and I want to continue thinking for myself without offloading any part of the process to corporate America. Thus, the writing of this essay was done by me at a keyboard without the use of any AI. Take that generative AI.