Slop comes for everything you love
I consider myself a lapsed member of the interactive fiction community. Writing and read IF was more or less my education in interactive narrative and where I got my start as a writer of stories meant to be read on the computer. I made several of my closest friends within that community, and I have very fond memories of submitting to the yearly Interactive Fiction Competition.
But I haven't kept up with it for several years; in the last couple years, a combination of intermittent health issues, being actively busy, and mounting burnout have kept me from engaging with a lot of seasonal events, including ifcomp.
This year, I have more free time (not out of choice), and so I wanted to check it out once more, play the entries, and submit a ballot! Go back to my roots. Have fun.
Cue the horror strings.
Out of a few dozen entries, I count 12 that explicitly say they used generative AI in some way or another – whether to generate a cover image, in-game assets, or actual text. These are just the more egregious examples of using it for cover images.
The cover image aspect of it is kind of dumb and instantly disqualifying to the authors in question – it obviously looks bad, but also cover image quality was just never really an issue with the ifcomp; historically most people used a public-domain image, a photo they took with their own two hands, or a simple word mark as their cover.
Using a GAN image instead lives at the intersection of tryhardism and aesthetic incompetence; you're trying to look more 'professional' in a field where professionalism is not really the goal, and in the process you're putting out something that looks gross and low-effort.
But the thing that's really crazy making to me is the games that use LLMs for text, including one particular entry that, to quote:
This adventure has been fine tuned and optimized to permit a player to paste the prompt into either the free version of Claude.Ai or ChatGPT and have a play and entire adventure.
Instead of pre-selected "choices", players tell the GM "Investigate the smoking cabinet" or "Consult the legal department on ...". The GM, within the well known limitations of AI, will improvise a response and move the story along.
To play: past the contents of the attached file into either Claude.ai or ChatGPT. Feel free to try it out on other LLM platforms, but your mileage will vary.
I realize that the IF community has a certain trauma around gatekeeping of what 'counts as IF' but I don't think a prompt you're meant to paste into chatGPT 'counts as IF' in any meaningful sense.
At this point this basically means I am not going to be playing or judging any games this year either, and I think it curtails me trying to return to involvement in the IF community, because there's now kind of a toxic dynamic at play if you're someone who has ethical and aesthetical objections to AI slop:
Theoretically, you could rate the AI entries at a 0; whether you bother to 'play' them at first (I'm not sure you can even play that last one in the expected sense and I would not want to touch chatGPT to do it either), the fact that you're summarily nuking them is a pretty obvious violation of the judge rules. I don't want to participate in a way that will be read by comp organizers as bad faith.
Alternatively, you can play and rate only entries that don't use AI. This would seem to be 'fine' but it creates a dynamic where the only people willing to play and rate the AI entries are people who are not going to object to them on grounds that they are AI, and thus they're getting judged by a different standard than everyone else's work. This means that the final result of the competition is at risk of legitimizing AI use or worse, making someone who put out real work feel bad that they placed behind someone who put out slop.
And I want to be clear, I would rate the worse, most amateurish piece of real human writing above the 'best' piece of slop.
Consider the example above of the 'game' that you 'play' on chatGPT. How many people are going to play and rate that who aren't already habitual chatGPT users?
Ultimately I think that slop is always creating a gresham's dynamic – bad-faith submissions crowd out good ones. The field of AI-generated entries seems to have grown significantly since last year.
These days I am not really active or connected in IF, which means I'm not effectively capable of pushback, so I'm kind of left just expressing my severe disappointment. For me, what IFComp and the broader IF community represents is genuine creative exploration that's unconstrained by commercial considerations, by trend-chasing, by trying to meet some standard of popularity or mainstream acceptance. It's a safe space for creative weirdos.
To see it become a stage for people who want the outcomes of creative exertion without the actual exertion is really sad to me. It also incorporates all of the pernicious effects of LLM usage into the competition as a matter of course. For me the saddest example is this one entry, where in the author's disclosure they write:
As you may know, English isn’t my native language, and in my previous works that was sometimes a drawback. For this project I focused on a smaller location and used ChatGPT as a writing assistant. I ran almost every sentence through it several times to correct grammar, smooth out phrasing, and find the most context-appropriate terms. It often suggested multiple options, and that really helped me tell my story the way I meant it, even in a language I’m not yet fluent in.
Which essentially amounts to: instead of producing something flawed and human you made something that a lot of people are not going to look at out of principle, and which (having seen plenty of LLM-written text at this point) I know is just going to cap out at 'barely adequate' in quality anyway. It's just sad; we are losing the ability to give people meaningful feedback when that feedback is going to cause them to retreat into embracing LLM usage.
Ultimately this feels like trying to go back to a space I used to be fond of and finding that it might not really be there any more.
I always welcome responses on things I post on this blog, but I want to specifically highlight that I have an open inbox specifically for this purpose at askdarchid [at] pm.me.