Generative A.I. For Writers: An Unfolding (But Not Inevitable) Nightmare!

by

From Chuck Wendig:

I have seen the sentiment around that generative AI for writers and artists is “inevitable,” which is a message that I think falls right in line with the myth of the starving artist — meaning, they’re two bits of pervasive folklore put forth by the Powers That Be, because it rewards and enriches those powers. To put a finer point on it, it’s ****ing capitalism. It’s capitalist propaganda bellowed from the deepest, most cankerous cave of moneyed interests, because if they say it enough times and make it true, then they make more money because we make less money, the end.

Just the same, I’ve seen some actual writers and actual artists start to… really take this to heart. They are taking on the inevitability of Gen AI sure as a broken-hulled boat takes on water — but that boat doesn’t have to sink, and nor does AI have to be inevitable. I do think it is inevitable that Moneyed Interests will continue to push AI as a catch-all solution to problems that don’t exist, and they won’t just let that bone go — but I do think, just like crypto and NFTs and what-have-you, that the actual value of Gen AI and the inclusion of Gen AI is far, far from confirmed prophecy.

So, this is a post talking about what we are, I anticipate, likely to see regarding artificial intelligence and both our writing lives and our writing careers. Note: none of this is good, but again, none of this needs to be inevitable, either, and I feel like blah blah blah, forewarned is forearmed.

Real quick, a quick sum-up of where we’re at with Gen AI in art and writing (and arguably music and game design and pretty much everything else):

a) It is built entirely on stolen work, colonizing the efforts of human creators, milling everything into artbarf and content slurry — and it is worth reminding too that it is not the AI that has stolen our work but rather, the creators of the AI who literally directed their artbarf robots to build themselves out of pilfered material.

b) It is environmentally damaging, increasingly so, guzzling water like a man in the desert and contributing overmuch to carbon emissions — see this article here, from Yale. Immigrants crossing borders are dying of thirst, but meanwhile, we’re feeding a half-a-liter of water to the machines just to ask it a couple-few dozen questions (which it will probably get wrong).

c) It continues to chew at the beams and struts of our information fidelity, and in those holes and in the inevitable collapse, mis- and disinformation will flourish like an invasive species.

With those three things in mind, it is fair to say, I think, that use of AI in writing and in the arts is unethical at present until the problems of stolen material, environmental damage and information erosion are addressed and solved. There’s a fourth thing, one that arguably is too true of everything we touch, which is that Gen AI exists largely to make Rich People Richer, and does nothing for everyone further down the ladder. (This is a much harder problem to solve because, well, welcome to the water in which we swim.) It serves companies. It does not serve people. It doesn’t help writers or artists or the audience. It’s there to make stuff fast, cheap, easy.

And, to opine a bit here, even outside the ethics of this, I also think use of Gen AI in this way is supremely lazy and completely betrays the entire point of making art and telling stories in the first ****ing place. It’s not helping us make the work better and get paid more. It’s relegating art and writing to a hobby only, while simpering incel chimps press buttons and get their rocks off by having the AI make images and stories of whatever mediocre garbage is passing through their minds at any given moment.

But, but, but —

Again, I don’t think this is inevitable.

Here I’m really going to switch gears and talk more explicitly about Gen AI in writing, and the problems it presents beyond the lack of ethics and the fact it’s really just there for lazy people who actually like the idea of writing more than they actually want to write. (Ironically, some people want to be a writer without doing work, but AI doesn’t fix that for them — they’re still not writing jack s****, they’re just zapping the Fancy Autocorrect Robot and making it s*** it out words for them. The software is the writer, not them.)

So, for me there are two key problems with Gen AI in writing —

1) It sucks.

It really just sucks. It’s not good. It can make the shape of the thing you want it to write (article, story, blog post, review) but then it fills it with half-assed hallucinations. Gen AI isn’t here to get things right, it’s here to make things look right, which is a very different thing. AI is vibes only. You don’t get an article — you get an article-shaped thing that’s just a really, really advanced version of Lorem Ipsum.

Gen AI isn’t true artificial intelligence. It isn’t “thinking” per se about input and output. It’s just barfing up the raw-throated bile of effervescent copypasta. It’s just a program tapping the predictive words button. And it knows to do this because, again, it’s stolen a whole lot of material to feed to its Judas Engine. So what it’s outputting is a broth steeped from tens of thousands of illicitly-yoinked human-created pieces of writing.

It also isn’t good at sustaining anything with continuity. Continuity is really important for writing — in an article, in an essay, and especially in longer-form material. When we talk about Chekhov’s Gun, that’s a shorthand that means the pieces of narrative information we use early are just the start of the trail of breadcrumbs that will carry us through the story. The gun appears early and must be used later — but that’s true of so much inside our work. We introduce things that are important, that have continuity throughout the work, that appear again and again and form a kind of constellation of narrative information — and that information comes in the form of themes, motifs, motivations, descriptions, tension-building plot points, and so on. AI has literally no understanding of that. Because it doesn’t understand anything. It just sees a pile of stuff and attempts to ape the shape and colors of that stuff. Gen AI artbarf can show you a house in image, but it has no idea what building a house means, it doesn’t know what’s behind the walls or how bricks are laid or how ****ing molecules and atoms form together to make everything — it just horks up the architectural hairball on command, like a cat with the Clapper in its stomach.

CLAP CLAP GIVE ME A VICTORIAN MANSE MADE OF CANDY HA HA LOOK MOTHER, LOOK, THE GOOD KITTY VOMITED FOR ME A CANDY HOUSE, I MADE ART, I AM AN ARTIST, MOTHER, PRAISE ME

Anyway. What I’m saying is–

AI doesn’t know s*** and can’t sustain s***.

And here the retort is often, “Well, sure, but this is what it can do now, imagine what it can do in a year or two.” And that mayyyyy be true, but I have a gut feeling that — particularly when it comes to writing — it has some very hard limits. It can never really go beyond the fact it is Fancy Autocorrect. Because it does not truly think, it will always be janky. It will never sustain information for long. It will always lie. It may be able to fake shorter pieces, but I also think that, like humans spotting Terminators, we will develop a keen eye to be able to spot this bulls*** with an increasingly refined Uncanny Valley detector in our guts.

2) The second problem is that it can’t be copyrighted. That’s a real problem, a true vulnerability, though one that hasn’t been entirely tested legally, yet — what if you push the AI-Do-My-Work-I-Suck-And-Am-Lazy button and it spits out a 5,000-word short story but then you change like, every 100th word? What does that mean for its copyrightability? I don’t know because I am a stupid person and not a lawyer, but I do suspect that it remains a very real weak spot in its defenses.

Link to the rest at Chuck Wendig

PG notes that, although he disagrees with Chuck about a variety of contentions in the OP, nobody ever doubts what Chuck’s opinion is about a great many things.

Click to Tweet/Email/Share This Post

18 thoughts on “Generative A.I. For Writers: An Unfolding (But Not Inevitable) Nightmare!”

  1. Two points, if anybody cares:

    1- There is a lot more to the “AI” field than LLMs and more to LLMs than chatbots or content assemblers. There arevall sorts of neural network software for control systems, machine vision, data analysis and inference. Which is why “AI” is going everywhere.

    2- Everywhere mean everything humans do. (Even things you don’t discuss in polite company. 😉 )
    One example os arming, and not just the million dollar combines for corporate farms, but also this for small family farms:

    https://m.youtube.com/watch?v=3M3s-VJ9r2g&pp=ygUQY2JzIG5ld3MgZmFybSBhaQ%3D%3D

    One reason there will be no place for the uninvited.

    Dismissing “AI” because you don’t get the fringe and generally meaningless fringe cases leaves you exposed to its near term use cases. Best to be on the lookout.

  2. It’s not helping us make the work better and get paid more.

    Consumers don’t care about you or your pay. They will buy what they like. If they like AI works, they will buy them. Some authors will compete and do well. Others won’t.

    • Reminds me of the guys who refused to use word processors, prefering typewriters, manuals ideally. Or the ones who chose to stick with handwriting. To each their own.

      But trying to turn preference into virtue has never stopped change.

  3. It begs to be said: Like a great many “big deal” and would-be “big-deal” writers who see themselves as greater than the great unwashed and complain constantly about “*ing capitalism,” ol’ Chuck offers his products for sale rather than giving them away in the true spirit of socialism. Just sayin’.

    • Worth noting is that Mr. Wendig wrote media tie-in novels for the Disney Star Wars Expanded Universe, and that the past tense is there because he managed to make himself obnoxious enough on Twitter that he got fired, which is quite an accomplishment when your opinions are leftist.

  4. If I was to remake “Invasion of the Body Snatchers,” I would have the clones identified by their unironic use of “colonizer.”

    Chuck’s post is unreadable. His rage and accusations and insults blinds us to the fact he doesn’t have many arguments.

    So far as I can tell, AI doesn’t spit out copyrighted material (at least, I haven’t seen it). The except would be people who use a specific image in their prompts, like “Nighthawks but with Disney characters.”

    AI uses existing texts much like humans do. The models take it in and blend it into something else. Unless you can show me a paragraph that contains “it was the best of times, it was the worst of times,” I’ll stay skeptical.

    Chuck rages about environmental damage, well, that ship to China sailed a long time ago. Anyone who uses a smartphone knows the scarring incurred in mining the rare earths used in its manufacture. You might as well rage about the rubber in your tires, the fibres in your polyester, or the gas in your car.

    As for copyright, unless you say you’re using AI, who’s to know? Frankly, I wouldn’t use AI to create finished work, anyway. I doubt it’s capable of doing finished work I’d be happy with. Art? Yes. Voice impersonation? Amazing. Images? Absolutely. I expect it to allow you to script a scene and have it “shot” with actors (similar to cut scenes in video games).

    AI like everything in our benighted world, is a tool. An invention. And Chuck, despite his fulminations, offers no solution. Because there is none. A total ban would drive it underground, to be used by bad actors inside and outside of governments and political parties.

  5. Overly emotive,

    If someone is dying of thirst(dehydration??) while crossing the border I would suggest it has more to do with lack of preparation, carrying capacity then how many gallons goes into a data centre x hundred miles away.

  6. Talk about monomania.
    With all the things Generative softwarecan offer writers, he focuses on the least important and uses it to dismiss the entire field.

    Last week, I did a little checking into CLAUDE AI 2.0 and found you can upload entire (first draft) manuscripts (plain txt format) for analysis for plot holes, inconsistencies, character arcs, and other “beta reader” functions.

    The weakness of generative software for story telling is that it is only knows what other stories have done. Well, what better tool to analyze the wordsmithing side of the business? It can’t tell original from derivative but not only isn’t that needed, it is better it doesn’t try, which a human might.

    Yet again the new refrain points the way: “AI won’t take away your job but somebody using it might.”

    Dismiss the tools entirely for what they won’t be used for and you’ll deprive yourself of what they will be used for. Your choice. Just don’t expect everybody else to do the same.
    They won’t.

    • ??? After-action report on how well Claude did at literary analysis / beta reading? I did have fun testing Claude in a different way, but it didn’t occur to me it could be put to that purpose.

      • I’m not ready to try it myself.
        Not well enough to pass judgment on how well it does what it claims to do. Or report on it. Not yet.

        But it was the first thing that came to mind when I heard of its input capability.

        I’m in no rush and I’d rather wait for a *local* SLM version. (Figure 6 months to a year for the second-gen AI PCs.)

        Likewise, when I heard about DALL-E, Midjourney, etc, the second thing that came to mind was “interior line art chapter images”. The tech ain’t there: it needs memory and iteration. Both are announced and coming.

        Sora? 60 second limit? “Book trailers!”
        Voice generators? Audio books! Narrated, not dramatized.

        Using LLM as story generators? Yeah, right. The world doesn’t need more pastiche or thinly disguised fanfic.

        Now, Patterson might find a use for it: feed it his famous chapter summaries against a SLM trained on *his* books and get back a book *he* might have written. But again: memory and iteration are needed so he can play story editor/beta reader.

        Everybody always jumps to the luddite take: technology making humans redundant. And it’s always the wrong take. Humans adapt and come out ahead.

        With *my* background I always go to “how can this improve my workflow”. (Which workflow I would never recommend to anybody else but works for me.) Tech is about tools and tools exist to make human output easier,faster,better. For those *willing* to stop and think about *their* needs and how the tool can help.

        And yes, it brings disruption, but disruption works in mysterious ways. 😉
        Consider this roundabout side effect of “AI” and the media pearl clutching:

        https://www.msn.com/en-us/money/other/how-gen-z-is-becoming-the-toolbelt-generation/ar-BB1kS2wP

        All the whining about how “AI” threatens white collar jobs is driving the smarter zoomers to take a closer look at where the world is going and moving to craft and technician training instead of the diploma mills and a future as baristas and activists/politicians. At a time the country is rebuilding its manufacturing capabilities around the newest tech. And paying serious money for competent workers, up to the six-figure range.

        As I mentioned a while back, the automated factories of the ’30’s and ’40s will have no trouble finding the maintenance mechanics to keep them running. Not at $100k a year. (And likewise have no need or use for the uninvited the activists keep bringing in. That is a looming problem.)

        The wave of “AI” tools is just ramping up and it *will* blow up a bubble just like the dot.com bubble of the ’90’s. But even after bursting, that left the US economy 40% more productive by 2006. There is no reason to expect any less (and ample reason to expectvmore, because of demographics) from the new software tools and robots. (Oh, those AI robots! Whole different story coming.)

        That’s what I’m doing these days, sniffing out the shape of the mid-21st century and beyond. If you’re into SF, you have to. And if I’m right (and the autocrat wars don’t derail it) things are going to get fun.

        So, back to you: where can *you* juice your workflow? What tools would you like to see? Keep an eye out: they are coming. Experiment, be patient and don’t rush. The current generation is just tinker toys compared to the customizable*personal* tools of ’25 and beyond.

        • Toolbelt generation? HGTV, DIY, and This Old House for the win! I want a fabric printer so I can tell a machine what pattern and color I want on my clothes or curtains, and it will print it up for me. Bonus if it’s the correct texture as well.

          The most tedious part of design projects (both digital and offline) is looking for the design assets that match my vision; my mother and I spend more time on that than the production part for textile projects. All the Luddites in the world will not hold back the tidal wave of adapters when such tech finally arrives and is used on those channels. Lines out the door at Jo-Ann Fabrics and Hobby Lobby (they’re next to each other where I live, so it would be epic).

          Now, for me I want the AI for concept art, and for low-level production — think game modules rather than Hollywood special FX. But like I said before, the generators need to be able to save what it generates and build from there.

          I would love it if I could tell Sora to generate music played with ancient instruments (sistrum, lyre, kithara, etc) like Peter Pringle plays on his YouTube channel. There’s a riff he plays on an Anatolian lute (Enkidu And The Woman Shamkat) that sounds like a riff I heard in “Rakset el Fadaa” by Omar Khorshid (an Egyptian guitarist). I can’t hire these guys (Khorshid is dead) but a Sora could help out here.

          But again, this tech has to be something I can use locally, without subscribing to the developer’s cloud. Back to Claude, I had imagined you feeding it one of the public domain books you’ve converted to e-books as a test of its analytical skills. Especially if the book is sufficiently obscure there’s not a lot of commentary about it online.

          • Good point.
            Gutenberg books are in txt format already.

            And this one would be a good test:

            https://www.gutenberg.org/ebooks/50571

            As to home cloth printers, well, there is this:

            https://hackaday.com/2022/11/28/desktop-sized-fully-automatic-loom-is-an-electromechanical-marvel/

            Luddite bait! 😉

            (Note that some existing 3D printers can do cloth, as noted anf lined in the piece.)

            What you want may be coming real.soon.now.

            Oh, and you forgot Mike Rowe. He is quite happy his message is bearing fruit. So is Zeihan.

            Me, too; less woke, more craft is the road forward.

            • Luddite bait!

              I see what you did there … and that loom looks like a nice 1.0 prototype and I’m curious to see what the 2.0 and 3.0 versions will be like. I’ve not seen this site, but Sarah Alvarez’s video (linked from the post) on 3D printing fabrics shows we’re closer to a the sci-fi future we see on Star Trek.

              Some parts of the future look so exciting, if one can cut through all the FUD.

              The PJF book you link to should be a good test. I think I’ve only read his Riverworld novels, which were turned into a Sci-Fi channel movie if I remember correctly, so less suitable for putting Claude through his paces.

              • Oh, my.
                Never mind RIVERWORLD.
                Farmer has better.
                (THE LOVERS, for one.)

                For sheer fun, try the WORLD OF TIERS series.
                (Kickaha is a hoot. So good he went from support to star in the atter books.)

                https://en.m.wikipedia.org/wiki/World_of_Tiers

                Or TWO HAWKS FROM EARTH.

                And GREEN ODYSSEY, which fell out of copyright for failure to renew it when it was required, is a fair sample of his brand of adventure SF.

                Farmer did everything from Tarzan and Doc Savage pastiche, to dead serious speculative sf and everything in between. (Like, JESUS ON MARS. Seriously.)

                Recommended but beware of A FEAST UNKNOWN which is not for everybody. Tarzan meets Doc Savage but with no guardrails. Violent and raunchy. ‘Cause he could pull it off.

                As to the future, if we can beat back authoritarians foreign and especially domestic “saviours of democracy”, the future can get amazing: moon base, asteroid mining (imagine if PSYCHE trully is most heavy metals), home robots, exoskeletons for the elderly, and HUD contact lenses among the easy predictions. 3D printed clothing, too. Download a design and print it exactly to your specs.

                Flying cars is the least of it.
                But only if we get past the crisis of the 20’s.
                Not a sure thing.

          • With music generators, copyrights csn be an issue with some tools. All grant the user unlimited use but some retain the copyright.
            Time will shake them out appropriately.

Comments are closed.