Call me a Luddite: I’m fed up with predictive large language models. I’m exhausted with it. I gave the stuff the good old college try. I looked at what it could do, how to program it. I started way before it was a Big Deal. Like many people today, back five years ago I wanted to be on the rising wave of a new technology. I wanted to know what I needed to know to continue working in a tech field, and the projects at the time seemed exciting. Who wouldn’t want an AI to answer the same, simple help desk questions so you didn’t have to?
Things I Wished It Would Do For Me
I wrote a Twitter Bot to post random Optimistic Science Fiction Story Ideas. I did this because I wanted more story ideas, myself. It ran for a few years. In the end, it just underscored for me that more ideas does not equate to more stories. The prompts didn’t inspire me, and when Twitter changed their API and broke the bot, I let it stay broken.
I started developing, and then stopped because I realized the icky moral implications, a bot to gently teach randos who sent me direct messages about treating women like people instead of some sort of sexual vending machine.
For years, I had tried to engage the many, many men who would send private messages aggressively flirting. I’ve written before about how… well, bad they were. They’d ask bland questions, reveal nothing about themselves, and then get indignant the minute they realized this wasn’t going to lead to a free sex fantasy chat. Wouldn’t it be great to automate my responses? I could help them, maybe, and save another woman trouble. Hell, I didn’t even need a large model. They were so predictable, I could use a parser such as they used for old text adventures. I got through three levels of exchange in a weekend of furious typing.
But, ultimately, it’s not my job to be an unpaid, untrained therapist to the masses. How would I even identify the best practices? Worse, what if my earnest bot actually convinced them I was interested in them, that their crappy hit-on strategy worked?
Garbage In; Garbage Out
My irritated remark every time someone asks me about AI is “It’s neither Artificial, nor intelligent.”
Not artificial: It is trained on human actions. It is a copyist. It does not really generate anything other than juxtaposition. Not artificial, merely automated.
Not intelligent: it doesn’t know what it’s generating. Doesn’t know what a sentence is, really, or even what grammar is. It is following rules and probabilities and flinging stuff together until it sticks.
When I started learning to code, a common phrase I heard in classes was: Garbage in; garbage out. Meaning that computers could not produce meaningful data from trash. You get what you put in. Your database lookup isn’t going to find the descrepency in accounts receivable if you didn’t input all the receipts.
The slop that gets created is an approximation of the pool of resources fed into the machine, and that’s all.
A few years ago, a school near here tried to train a model to predict which applications represented students who would actually enroll, meaning they could weed out the hopeless causes and have fewer applications for the faculty to read. The first test removed all the minority students, citing that minorities were less likely to be accepted. They abandoned the test in chagrin. They would just have to read their applications and hope they got less biased.
I Got Scammed
Here’s the thing that got me to write this blog post: I thought I could tell AI-generated stories from real ones, then I became an editor and got scammed. TWICE.
When in the past, I had been presented with AI-generated stories labelled as such, I easily saw their flaws. When they weren’t labelled, I thought I was dealing with a precious newbie writer who didn’t quite grok tension or scene yet, but was trying. There was a rudimentary plot, and I could guide them into improving.
It took another editor contacting me about someone who had scammed them to get me to run an AI-checker on the piece. When the first one came back 100% AI, I ran a second and a third. I then went to another editor and asked their advice. They pointed out red flags I had overlooked, including the odd request not to use our usual Paypal method for payment. “Paypal has fraud protection,” my friend explained.
Then we ran AI-checkers on all of our pieces, to be sure, and found another one. It had the same red flags: the author simply said “sure whatever” to my edit suggestions, the piece was a little too perfectly aligned with our written guidelines, and they had odd requests around payment.
The final straw was when confronted, the scammers either were silent, tacitly admitting guilt and moving on, or tried to fight, pushing it back on me, “Prove that it was AI!” A real author, when asked to talk about their process composing a piece, would have something ready at hand to say. Or at least, “Work with me, how do I prove this was my work?”
And yes, I am concerned that I’m giving future scammers some advice here. That’s why I won’t say more.
The real losers here are the newbie writers, the ones who aren’t quite there. As an editor, I want to accept stories that need a little love to be presentable. I want to help the newbie! But now, I know I can’t tell if it’s unskilled or Memorex. I can imagine many editors wanting to stick just with known writers, friends and acquaintances… closing one of the few open doors to a beginner in our field.
What Does It All Mean?
Is this the literal death of the author? If I can be fooled, well, can’t anyone?
I foresee in the very near future that the job of churning out work-for-hire or ghostwritten memoir will be taken by AI. All those “I have an idea for a book, you write it and we’ll split the profits!” will skip the unwilling middleman. Likely, formulaic works, like pulp romance will be hit first: The kind where they have the first sex scene by page 21 and the second by page 45. You know, Harlequin.
That doesn’t mean people won’t be writing romance, for low and high brow, and that there won’t be a heavy load of activity on Archive of Our Own tailoring specific kinks to specific fandoms. But we write those things for the joy of the thing itself, not for the money.
I worry about the money. For most authors, churning out a work-for-hire is the most obtainable way of earning money with our skill. It’s not stable, but it feels good to earn from writing as opposed to your ability to maintain a cheerful demeanor over a cash register when the fifth person in a row asks you why the burgers cost so much like you, personally, set the prices, or have any pull with the corporate decision makers.
The thing is: only those with the means can create art for the joy of it. If machines are making art, what do the artists do?
Ultimately, if we want art in a world that won’t pay artists for their craft, we need to find a way for artists to live. Universal basic income, anyone?
