AI in competitions

[mathbrush (Brian Rushton)]

I think it’s important to see if AI would be useful if it were ethical. “Unethical” and “low quality” aren’t always the same thing, and it can be messy to confuse the two (like when people find out an actor did something bad and say, “well I never liked his movies”, implying bad actors are always bad people).

In this case, I think generative ai use for games is both. The prose given here is definitely serviceable and is probably 3/5 writing wise.

But building a game on this prose would be rough. There are a lot of extraneous details here, like scurrying sounds and shadows. This invites players to “listen to scurrying” and “search shadows” and so on. If the author implements every reasonable action, it would take up a ton of code. But since ai wrote this, those things serve no purpose but window dressing. So now the player has spent a lot of actions trying to interact with the world and guessing what is important to further the story and the game, and failing.

So the author has spent a ton of work and so has the player and no one is happy.

Or, the author doesn’t implement everything, and it’s buggy.

If you look at popular parser games, they tend to write in a way to eliminate extraneous scenery objects. Wizard Sniffer is very barebones. Ryan Veeder is similar, if you read his design philosophy.

Other people make lush games but everything ties in. Maybe there’s a vampire in the attic, so we see bat themed things downstairs and a coffin in the basement and so on.

Like most ai generated text, what’s here looks great but doesn’t hold up to inspection. It might actually be human written as a test for us, but if it is I don’t think it’s useful anyway. The difference is a human can learn and grow. This text kind of reminds me of Horrors of Rylvania.

A final thing is that all of the descriptions are almost the same size. It’s nice to vary those like the way a bio author varies paragraphs.

[DeusIrae (Mike Russo)]

Yup - everything’s the same length, there are tons of chandeliers, the lights are all flickering… it’s not too hard to see the markers. But again, it can be easy to think someone who’s a weaker writer or ESL could be using generative AI, since some of the markers can be similar. And given that as folks have said the consequences for violating rules and community norms can be appropriately significant, taking transparency seriously and not trying to get cute with rules-lawyering about what really counts as generative AI seems like by far the best approach. Heck, we’re just talking about disclosure here, so if an author is doing something nonstandard, they can just sum it up in the blurb and give details in the credits.

[EJoyce (Emery Joyce)]

@Dannii

Were someone to use a genAI system trained solely on ethically sourced data it would probably be much better received

We don’t even need to resort to hypotheticals here; no one had ethical complaints about Robert Goodwin’s two chatbot games or You Will Thank Me As Fast As You Thank A Werewolf (which is trained entirely on the creator’s own writing; hard to get more ethical than that). I don’t think anyone really liked You Will Thank Me, but I remember the reaction as more “huh, interesting experiment, didn’t really work” than outrage that it was in IFComp at all.

[DeusIrae (Mike Russo)]

Yeah, those are perfect examples - they said exactly how they used their tools, and it was fine (as you say, …Werewolf is probably the worst-reviewed of BJ Best’s games, but that goes to the practical rather than ethical issues with LLMs!)

[Joey]

@8bitAG

Haunted House room descriptions

Entrance Hall
You stand in the grand entrance hall, where the air is thick with dust and the scent of decay. The flickering light from a tarnished chandelier casts long, dancing shadows on the cracked marble floor. Cobwebs drape from the high ceiling, and an ancient grandfather clock ticks ominously in the corner, its hands frozen at midnight. The silence is broken only by the distant sound of a door creaking somewhere deep within the mansion.

Library
The library is a cavernous room lined with towering bookshelves, each crammed with dusty tomes and forgotten manuscripts. A large, ornate fireplace dominates one wall, its hearth cold and empty. The only light comes from a single, flickering candle on a mahogany desk, casting eerie shadows that seem to move of their own accord. The air is thick with the musty smell of old books and the faint, unsettling scent of something long dead.
Dining Room
You enter the dining room, where a long, polished table stretches out before you, set with tarnished silverware and cracked porcelain plates. The remnants of a feast, now rotting and covered in mold, sit untouched. A grand chandelier hangs above, its crystals tinkling softly in an unseen breeze. The walls are adorned with faded portraits whose eyes seem to follow your every move, and the air is filled with the faint, ghostly strains of a long-forgotten waltz.
Master Bedroom
The master bedroom is a scene of faded opulence, with a grand four-poster bed draped in tattered, moth-eaten curtains. The once-luxurious carpet is now threadbare and stained, and the wallpaper peels in long, curling strips. A large, cracked mirror hangs above a vanity, reflecting the room in a distorted, unsettling way. The air is cold and still, and you can’t shake the feeling that you’re being watched.

Cellar
The cellar is a dark, damp space, filled with the smell of mold and decay. The walls are lined with shelves holding dusty jars of preserved horrors, and the floor is slick with moisture. A single, bare bulb hangs from the ceiling, casting a weak, yellow light that barely penetrates the gloom. The sound of dripping water echoes through the space, and you can hear the faint scurrying of unseen creatures in the shadows.

It’s got the tell-tale repetitiveness, cliché, and lack of verve. Count all the “flickering” light, “decay”, and “thick with” dust. Good writing should get better the more you read of it, not undermine earlier descriptions through exhaustive sameness. I was able to generate the same phrases and imagery with the first prompt I thought of:

ChatGPT prompt 'Generate text adventure room descriptions for an Entrance Hall, Library, Dining Room, Master Bedroom and Cellar in a haunted house.'

Entrance Hall

You step into the Entrance Hall, the grand gateway to the haunted manor. The air is thick with dust, and the floorboards creak underfoot, sending echoes into the vast space. Tall, arched windows line the walls, their panes obscured by thick, ancient curtains that barely sway in the cold draft that seeps through. An elaborate chandelier, heavy with cobwebs, hangs precariously from the ceiling, casting ghostly shadows across the cracked marble floor. A grand staircase ascends into darkness, while doorways on either side lead to other parts of the house. The scent of decaying wood and something metallic lingers in the air, unsettling your nerves.

Library

You push open the heavy door to the Library, and it groans loudly as if in protest. Inside, the room is cloaked in darkness, save for the faint, flickering light of a dying fire in the fireplace. Shelves stretch from floor to ceiling, packed with dusty tomes and strange artifacts, many of which seem to hum with a sinister energy. A thick layer of dust coats everything, undisturbed by time or hand. A large, ornate desk sits at the center, its surface cluttered with yellowed papers, quills, and a peculiar-looking amulet. The air is thick with the smell of old books and something far less pleasant—a faint, acrid odor that stings your nostrils. You sense the presence of eyes upon you, though the room is empty.

Dining Room

The Dining Room is eerily silent, the long, polished table set for a feast that never came. Tall candelabras stand as silent sentinels at either end of the table, their candles melted down to stubs long ago. The fine china is still laid out, though it is cracked and stained, and the silverware is tarnished beyond recognition. Faded, once-opulent wallpaper peels from the walls, and the heavy drapes are drawn tight, blocking out all light. A large portrait hangs crookedly above the fireplace, depicting a stern-looking man whose eyes seem to follow you no matter where you stand. The air here is thick, almost suffocating, and carries a faint, rotten scent as if the room itself is decaying.

Master Bedroom

The Master Bedroom is a suffocating space, heavy with the weight of forgotten memories. The grand four-poster bed dominates the room, its velvet curtains moth-eaten and sagging. The bedding is rumpled and stained, as though someone left in a hurry—or never left at all. A large, cracked mirror hangs on the wall opposite the bed, reflecting the dim light from a flickering candle on the bedside table. The wallpaper is faded and torn, revealing the bare, rotting wood beneath. An old wardrobe stands open, its contents strewn across the floor, and a chill breeze seems to seep from within it, carrying with it the faintest sound of whispers. The air here is cold and damp, and every breath feels like a struggle.

Cellar

The cellar door creaks open with a sound that seems to reverberate through the entire house, as if waking something ancient and long-forgotten. The stairs leading down are steep and narrow, each step groaning underfoot as you descend into darkness. The air grows colder and heavier with each step, carrying the pungent smell of damp earth and mold. The cellar itself is a labyrinth of shadowy alcoves and dusty, forgotten relics. Old barrels and crates are stacked haphazardly, some overturned, their contents spilled across the dirt floor. A single, bare light bulb dangles from the low ceiling, flickering erratically, casting sinister shadows that dance around the room. In the far corner, you hear the faint drip of water, though the source is hidden in the darkness. The walls feel like they’re closing in, and the air is thick with a sense of malevolence.

[Draconis (Daniel Stelzer)]

My understanding of the IFComp’s current policy is that you should credit an LLM for something if it’s creative, and if you would credit a human for that same task. Brainstorming ideas until something clicks generally doesn’t require credit—I’ve asked for help with that on the forums plenty of times, without crediting each individual person who responded, though I usually thank the forum as a whole—but making cover art or writing the text of the game does.

Similarly, I would credit a human differently for “this person wrote half the text of the game” versus “this person translated my English into French” versus “this person edited the text I’d written and helped it flow better”, so I would credit an LLM differently for each of those.

It’s not really—to my read, at least—taking an ethical stance on the use of LLMs, it’s just calling for attribution to be given if an LLM is having creative input on the work. Inform 7 translating your code into Inform 6 then compiling into Z-machine assembly isn’t really having creative input, nor is a web browser rendering HTML and CSS into something pretty; the creative work lies in writing the input to those tools, since we can be confident that the same input will produce approximately the same output.

But the point of GenAI systems like ChatGPT is that the output is not predictable from the input; it’s meant to be (or at least come across as) creative and original.

[8bitAG (Strident)]

@Joey

It’s got the tell-tale repetitiveness, cliché, and lack of verve. Count all the “flickering” light, “decay”, and “thick with” dust.

Well, the text it generates is very similar to the sort of text adventures from the 1980s and 1990s I play. To be fair to Copilot, the repetition was largely eliminated with a follow-up of “Can you make the descriptions more different to each other?” (Which I won’t post here). It’s probably a better “writer” than me, so it’s a good job I don’t enter competitions.

[mathbrush (Brian Rushton)]

Maybe you should enter! I got into if because I was repeatedly told I was a bad writer and wanted to ‘prove myself’. And my first few games when reviewed agreed I was bad writer, but I was able to place pretty high from the strength of mechanics alone. Id like to think I’ve improved since then but my point is in not the only one; there’s a lot of great games with bad writing in IFcomp, so you could join!

[dgtziea]

AI cover art doesn’t bother me nearly as much as AI text would. And for now, I’d try to judge and review disclosed AI entries–even text ones–“fairly”. Though the idea of spending too many hours of my life writing down critical feedback for LLM AI text doesn’t seem all that appealing. But I’ll try to judge it like anything else. Also not as confident about my own abilities to always spot AI output as other people are, as we look more towards the future. I do foresee that being a problem.

I do hope that as a community, we don’t put too much emphasis on cover art. The cover art should be more extraneous, and not treated as part of the entry.

I don’t want people to overly worry about their entry’s reception if they have bad or no cover art, and I especially don’t want people without art skills to worry about it! Royal-free photo? Fine! Just the title in a fancy font? Great! Nothing at all? Please write a good blurb, but I’ll still play it! And I hope other people won’t judge books by their lack of cover as well, and still try them out.

But if AI cover art is something we really don’t want in comps, then I’d argue we should also try to de-emphasize the need for “good cover art” overall. Because then there isn’t a skill gap issue there for people to look towards AI to solve.

There’s also a way that websites can emphasize and de-emphasize cover art vs text, and you can see it in how headlines and thumbnails are organized and have evolved on lots of article-based sites out there. And that also plays into how much cover art is needed for a comp: the comp websites.

I seemed to recall Spring Thing had a really nice way of handling no-cover-art entries. Looked it up, and yeah, it looks like they let you add text to display in a little text box? GREAT! Kudos to Aaron, or whomever designed that! We’ve seen an increase in cover art over the years, but you can see several in 2017: Spring Thing Festival of Interactive Fiction. I think it looks great, and no-cover-art entries stand up well, just in terms of visual at-a-glance appeal. And this where the history of the website comes into play now: before 2015, the spring thing site was much simpler, and there was no cover art at all.

The IFComp site meanwhile just moves all the text over if there’s no cover art. Maybe not quite as nice, but okay. How much does that effect what gets played? itch.io, I noticed during ParserComp, really expects cover art. You just get a little no image:( with a sad emoji. That seems like a value judgment, doesn’t it? Other game hosting platforms fully expect cover art. I don’t want that value judgment to seep into IF.

(okay that last part doesn’t have much to do with AI. But I wrote it already so…)

[vivdunstan]

@dgtziea

I do hope that as a community, we don’t put too much emphasis on cover art. The cover art should be more extraneous, and not treated as part of the entry.

I’ve found the cover art seeming expectations really off putting in the past. It’s always been the most problematic aspect for me in preparing an entry for a competition. And it hasn’t honestly helped me when umpteen people have said it’s easy, just do something simple. Though more constructive suggestions help more.

And while I’m not a fan of normalisation of generative AI use even for cover art, I can fully understand why some authors would want to use it. Especially given how I perceive an expectation for good cover art.

This time around I managed to produce my own cover art, drawing on my iPad. It was helped that my game really suited the rather rudimentary style of art I produced!

But I have honestly found producing cover art really off putting in the past.

[dgtziea]

@vivdunstan

I’ve found the cover art seeming expectations really off putting in the past. It’s always been the most problematic aspect for me in preparing an entry for a competition.

yeah, exactly! Exactly my worry, and exactly why I understand why even someone who isn’t a huge AI enthusiast might still look to those tools for cover art help.

There were some IFComp blurb + cover art review threads a couple years back, and those always mildly concerned me. The idea with those was that blurbs and cover art were what players used to decide what to play, so “finding your audience” was important. It is, and the threads were always meant to be fun, but you know, think too much about “audience” and suddenly you’re thinking in a marketing mindset, instead of just… IF as a creative outlet.

And well, if the cover art’s worthy of getting critiqued and talked about, then aren’t the expectations rising too much for our text-focused medium?

[alyshkalia (Tabitha (ey/em or they/them))]

Re: cover art, I quite enjoy making my own using royalty-free or public domain images and a nice font. If authors posted on the forum that they were seeking simple cover art of this nature, you’d likely have multiple people who’d be happy to make something for you!

[EJoyce (Emery Joyce)]

@dgtziea

itch.io, I noticed during ParserComp, really expects cover art. You just get a little no image:( with a sad emoji. That seems like a value judgment, doesn’t it?

That’s a recent change, I think—you used to get a grey background with a star on it, and I’m pretty sure that was true as of the point when I finally uploaded cover art for Someone Else’s Story three or four months ago. I don’t like the change but I don’t think it can have influenced IF culture all that much in that short amount of time.

[DeusIrae (Mike Russo)]

I also have no visual design sense so making cover images does stress me out too, I admit - fortunately my wife does have have some of those skills and is willing to help out! But I think if you look at high-placing games in recent Comps, it’s pretty clear there’s not much of a bar you need your cover to clear to do well; public domain picture or generic background with a slightly in-theme font for the title seems to work just fine for many authors (paying attention to your blurb is really good advice though!)

[Encorm (N. Cormier)]

@dgtziea

It is, and the threads were always meant to be fun, but you know, think too much about “audience” and suddenly you’re thinking in a marketing mindset, instead of just… IF as a creative outlet.

Eh, I don’t really mind this in the context of a competition, especially one as big as IFComp. When there’s 70+ entries people are always going to latch onto something to figure out what to play first. With that many entries you’re competing for people’s time and attention just as much as their vote – it’s just the nature of the beast. If you don’t want to think about your audience at all, why enter a ranked competition?

I haven’t seen much discussion of cover art or blurbs outside competitions, let alone criticism based on presence or quality.

EDIT: @EJoyce just reminded me that our extremely well-regarded, XYZZY-winning game has absolutely F-tier cover art. It was made at 3am the night before the deadline out of free clip art. And nobody’s ever said a word about it!

[HanonO (Hanon Ondricek)]

I think the major point of contention is IF in the context of a competition or commercially: If an AI generated your plot and much of your prose, why would we reward an author for what might be considered a largely machine-generated work?

I enjoyed Colossal Cursehead which was in the Really Bad IF Jam. It literally mashes up three games in ways that are interesting. It’s by no means a playable game but as an experiment that wasn’t asking for payment commercially and making no bones that it sourced its text from three games is the way to do it.

There’s already a slippery slope with AI “tools” - I think @Encorm has it right:

@Encorm

Part of the problem here is that “AI” is a hot buzzword right now and is getting slapped on everything

@Encorm

Photoshop’s AI tools, some of which have been around for years under different names and only got rebranded to “AI” recently to capitalize on the zeitgeist. These are fundamentally different from generative AI so I think it’s important to keep track of the difference when we’re discussing this.

There have always been algorithmic filters in art programs to randomly generate a water texture or a sky or VHS “noise” in an image. I agree that there has always been that line of “transformative” work. If I take a painted sky from a Thomas Kincaid painting, flip it sideways and zoom in on one section, re-tune the colors so they’re completely different so it’s unrecognizable and use it as a background for another image and my title as a logo, is that sus?

The Cragne Manor logo is actually an example of this. During the creation of that game as part of our inspirational and development pow-wow in the Slack I created a collage public domain photos of New England buildings, flipping some of them upside down and re-assembling like a puzzle. Ryan liked it and actually used that as source art to create an entirely new illustration for the logo almost like a woodcut. I believe he also wanted to get away from just using a photo montage.

Here are some of the reference photos.

This is a version of my collage that another author put through some generative tool (“Dreamtime?”) and you can see it actually is made up of sourced photographs since you can see faces and creatures and puppies and…grocery store produce sections in the image. I love the craziness of this one, but it would be understandably shunned by many people on here.

Long story short - too late: I can appreciate that IFComp is making a rule that AI generation tools must be credited or cited as you would an author or another work. It doesn’t necessarily mean the work is sullied because of it, but it lets the people who are fiercely averse avoid those works - and preferably not hate-vote 0 without playing the game because grudge-voting is against the rules.

[dgtziea]

@EJoyce

That’s a recent change, I think—you used to get a grey background with a star on it

Thanks for the context! The change won’t effect IF culture yet I agree. The grey star background does sound familiar, thinking back. I’ve played older itch io hosted competitions before though and those never really stuck out to me the way the new defaults immediately did, so maybe that follows my point about website choices and cover art.

@Encorm

EDIT: @EJoyce just reminded me that our extremely well-regarded, XYZZY-winning game has absolutely F-tier cover art. It was made at 3am the night before the deadline out of free clip art. And nobody’s ever said a word about it!

I think that’s great cover art!

@Encorm

I haven’t seen much discussion of cover art or blurbs outside competitions…

Isn’t that my point about how our comp websites can effect how much cover art matters though? Games outside competitions don’t need cover art. And our comp websites don’t need to have a place for cover art at all, right? That might be radical; I think it should be an option. But I’m just reflecting on how our community thinks about cover art standards. Maybe not just in reviews or forum threads, but just in what gets played at all, which is more where comp website layouts come in.

And should authors feel the need to rush at 3AM the night before to put something together at all?

[HanonO (Hanon Ondricek)]

@dgtziea

But I’m just reflecting on how our community thinks about cover art standards.

Cover Art is an advertising tool, just like a blurb. It gets attention, which is what you want in a Comp.

A good cover might not reflect the actual quality of the game, but it can attract eyes. it’s frosting on the cake. Do you want one with beautiful custom packaging, or do you want a plain flat sheet cake in a box that just has a sticker with the word “CAKE” on it in Comic Sans?

Back in the day, I bought games from Radio Shack that were a cassette tape with a dot-matrix printed label in a ziplock bag. Cuz that’s all we had back in the d-[coughs graveyard dust].

[Joey]

@8bitAG

It’s probably a better “writer” than me

I beg to differ, you are a superior writer because you are writing with a purpose. The writing in a text adventure of the sort you make doesn’t just provide atmosphere. It conveys useful information to the player (about what can be interacted with) and it can also reveal character. Generative AI prose tends to produce a lot more descriptions about unimportant scenery than necessary, and (because it doesn’t know your characters) it can’t reveal character. This aspect of writing is something you’re quite good at: the primary strength of your room descriptions is in their capacity to reveal something about the protagonist:

Fewer aspects of the world are described than in the generated prose, but it’s both more functional and characterful. Because the scant items that are described in your games are backed up by the world model, they have a greater reality and weight than empty generated scenery.

[EJoyce (Emery Joyce)]

That’s a Spring Thing game; as you note, Spring Thing does not require cover art and has a perfectly good alternate solution for games that don’t have it, so I did not in any way have to scramble to put together a shitty jpeg-artifacted clipart cover at the last minute, I just did it because… IDK, I was kind of in a fugue state at that point. I may have had a migraine? (That either happened with LT1 or LT2 or both, I forget.) I was not thinking clearly, in any case.

Anyway, every game I’ve made for which someone else wasn’t making the cover art has had art that is either shitty, minimalist, or nonexistent and as far as I can tell it has never affected critical response to the game or the game’s placement in competitions. The only time I’ve seen strong negative reactions to cover art has been when it’s AI-generated, which is something people have a moral objection to that no one has to art that is simply bad. I honestly think people who feel like cover art is required and needs to be good and who use AI because they feel like they need it in order to “keep up” are perceiving a problem that isn’t really there and “solving” it in a way that creates a different problem — although the response to the AI cover art may accidentally reinforce the perception that people care about your cover art.

Edit: Upon further reflection I’ve remembered that I did in fact have migraines in the home stretch of both LT1 and LT2, and with LT1 in particular the emotional effects of the migraine made me very, very weird about the game and unable to stop messing with it (which was a vicious cycle because staring at the computer screen made the migraine symptoms worse). So yeah, I was extremely not in my right mind at the time and I would not recommend drawing conclusions about the pressures of the IF community based on stuff I did when my neurological issues were screwing with me.