Can we just ban AI content on IFComp?

Competitions IFComp
101

In the past, they have been able to disqualify entries based on voting irregularities such as multiple accounts voting from the same IP (someone having different accounts vote from the same computer) especially when coupled with weird trends like an account voting just for the minimum of five games and giving four zeros and one 10 or a high score.

If it’s just discussed verbally there’s likely nothing they can do as long as the judges behave like legitimate judges and aren’t obviously stuffing the ballot box.

Procedural Text Generation has always been part of IF. The simplest version of it is a phrase that can randomly vary itself like a mad-lib.

say "You [one of]open[or]crack open[or]page through[at random] the leaves of the [one of]ancient[or]dusty[or]mysterious[cycling] [one of]book[or]tome[or]grimmerie[as decreasingly likely outcomes]."

More complicated versions can vary the response more intelligently based on world-state or variables, but all these rules and algorithms are outputting authored text and not calling out to a separate AI tool to improvise responses using words and phrases the author did not write.

I beta tested Versu and while they would sometimes refer to what it was doing as “artificial intelligence” it was all programmed within the engine which was at least partially based on code and algorithms from their partner, Linden Labs. It was mostly designed to give NPCs agency but the author had to make it all work. NPCs were able to leave and enter scenes randomly, or possibly if they got enough points weighting them toward being “offended” they might storm out of the room. They could carry out actions and even redirect the conversation flow toward a subject they “wanted” to discuss. It was basically a robust version of a conversation system based on a lot of Emily Short’s continuing work. It still only did what the author wanted.

If I tell ChatGPT to implement an Inform 7 game for me about pirates and I enter that into a competition, skipping 90% of the coding process, that’s not fair to someone who put in the hours doing the entire thing themselves.

I’m not saying there isn’t value in companies using an LLM or AI to automate certain coding tasks as part of their business, but enterprise use is different than a contest offering prizes ostensibly rewarding individuals’ IF skills. It’s like one student using a calculator on a basic math test to get an A when other students don’t get to.

16 Likes
102

I’m kind of worried that we’re getting carried away here. What this reminds me of – and I’m 100% certain that you don’t mean it that way! – is people complaining about games having an easy mode, because players should just ‘get good’ and ‘put some damn effort into it’. And that, as we know, is fairly blatant gatekeeping.

One of the things I learned from recent posts to this thread is that there have been communities, especially the Twine community I understand, that got flooded with horrible help requests by people whose AI generated code didn’t work. That clearly makes AI generated code bad for the community. It’s something I wasn’t aware of, but it makes perfect sense.

On the other hand, I’ve also had times when I was really struggling to get the exact right Linux terminal command to do something, and nothing I found on the internet was exactly right for my situation (and I spent a fair amount of time searching!); and asking an LLM gave me the right answer immediately. It would be really harsh to then stand next to me and tell me to get good. I was just trying to fix some small thing on my computer; I did not want to become a bash expert, nor did I have the time to become that.

There are some circumstances where LLMs empower people to do things they would not be realistically able to do otherwise. There are many other circumstances where LLMs harm their users (and unfortunately also their non-users). I’m pretty convinced that our overlapping communities will have to draw some lines in the sand to protect ourselves. But I also think we don’t want to lash out at each other, and especially not at newcomers!

20 Likes
103

I am dying to know why you seem to be deliberately ignoring the point in what reads like a bad-faith effort, especially after @pieartsy has spelled out where this line is twice in this thread. You’re really digging a hole for yourself and your argument here.

7 Likes
104

This is definitely the case - certainly a part of my wariness here is the demonstrated bad faith and sharp dealing of the big GenAI companies in general, and some of the people bringing it into the IF space like what happened in ParserComp in specific.

This feels like it’s just restating the difference between using an LLM to write code versus using it to create in-game assets, which I don’t think anyone’s confused about? At least if this example is meant to present an edge case of some kind, I’m not seeing it.

I feel like it’s more like a teacher assigning their class to spend thirty minutes writing a short paper, and then ten kids turn in handwritten ~300 word essays and two submit 50-page monsters, and now the teacher needs to figure out how they’re going to grade these things (both in terms of what rubric to use, and how they’re going to find the time).

11 Likes
105

So, I’m currently somewhat ill – a flu or something, I’m not sure. Anyway, I was lying in bed last night, totally unable to sleep, and my flu-addled brain just kept returning to this thread with a hundred variations on the thought that LLMs will destroy everything that is good. I would have been really smart of me to not return to this thread today!

I was not really smart. PM me with suggestions for the most wholesome IF Comp game I could play this evening. :grinning_face_with_smiling_eyes: (Thanks, I got some nice recommendations! You can still PM me, of course, but I should be covered.)

11 Likes
106

Conflating one thing with another totally different thing because they happen to have the same name is a rhetorical move, and a very clumsy one.

7 Likes
107

I think it’s worth reiterating that there are also plenty of reasons people don’t want LLM-generated entries in the competition entirely separate from the ethical and environmental ones, which wouldn’t be mitigated even if someone invented a LLM trained entirely on public-domain and permissively-licensed material that runs on an average laptop.

The proximate cause of Spring Thing banning LLM-generated entries was the problems it was causing for the community: reviewers not wanting to participate in the competition if it meant reading LLM-generated text. That’s why Spring Thing’s ban applies to images and text but not to code—because LLM-generated code wasn’t causing a problem for reviewers.

Given how a decent proportion of IFComp entries are using AI-generated cover art, and that’s turning people away, banning that seems like a clear and unambiguous positive: the impact on LLM-liking authors is minimal, and the impact on the community is good.

Is the smaller fraction of entries with LLM-generated text also causing this sort of problem? I wasn’t convinced it was, before, but this thread now points to “yes”; even when they’re clearly flagged, some reviewers are saying they’re tired of reviewing these entries, while others are saying they’re avoiding the competition entirely because of it.

We can debate the ethics of edge cases all month if we want, but from my point of view as a forum moderator, the more important question is: are the drawbacks to the community outweighing the benefits? And signs currently point to yes. There are debates to be had about the ethics of laws and government, after all, but at the same time, speed limits make driving safer for everyone.

22 Likes
108

I genuinely don’t understand why people even use AI for cover art. Yes, the art is very detailed and ‘polished’ in a sense, but it looks so, so terrible; it gives a negative impression of the entry it’s attached to almost immediately. And the more ‘ambitious’ the cover art (eg, trying to replicate historical styles or do highly rendered illustration), the worse it tends to look. In basically every case I think I would be more positively inclined towards the game if you just used simple word art or a public-domain illustration you took time to choose with care.

22 Likes
109

While that has been true in the past, and was one of the reasons the Twine Discord banned AI-generated code, the field is constantly moving. Recent tests show that LLM’s across the board are much better at Twine formats than they were, mostly because they can simply have the entire documentation of a format uploaded along with your question/code.

They still make mistakes with Twine formats, or use out of date functions, or mistake your intent, but that’s true of every language when you ask for niche solutions (the less your particular issue has been seen in the training data, the less well they do).

1 Like
110

I should say that this does happen sometimes, and particularly happened in Spring Thing this year, without (I believe) any real intent. One of the entries (which was good and polished) got several votes sent in with email addresses from the same university as the author. All those people voted for that game for best of show and didn’t vote on anything else (whereas some people had votes for every single game for ribbons). One of those voters even emailed me saying they were a coworker of the person and wanted to see if the votes went through. My guess is that they (quite reasonably) told friends about the competition and that those friends wanted to support them.

Fortunately, I didn’t have to take any action because all of those votes added up were still only enough for third place, and only the top two places count. That’s why I think it’s good to have rules where you have to vote for more than just 1 game (IFComp does 5). I’ll have to remember to include that as a rule for Spring Thing 2026.

(I also included a long comment about AI here but it’s just rehashing both things I’ve said a lot and others have already said on this forum, so I’m trimming this down to the newer info that is relevant).

17 Likes
111

Hi all. I registered just to post in this thread. I’ve never participated in IFComp before and have no interest in authoring, but for years I’ve considered playing and judging - and this year I finally took the steps to see what that entails. And I have to say: immediately I was put off by the mere presence of obviously AI generated entries among the lot. This is something I naturally assumed IFComp would have ruled against already - five years ago AI Dungeon showed the way the winds were blowing, and watching platform after platform impose bans to protect the integrity of their community, I figured that IFComp would be at the forefront of handling this.

As a result of seeing the AI slop, I simply decided against playing this year. I had also prepared to recruit friends to help out, because group discussion about games is a fun way to boost interest and enhance the experience. But I won’t be doing that either. (And not just for me - I already know my friend group is opposed to this rising tide of garbage, so I wouldn’t even want to waste their time either. I assure you, they have the same objections.) This isn’t a threat, it’s just an explanation of the multiplicative effect of the choices being made here: if people stop sharing IFComp on social media and pushing for folks to get involved, it will see drastically reduced participation and outreach.

I need to be very clear that this isn’t something you can “debate” people out of. I see a lot of people going to bat for LLM tech in here, telling people they’re wrong about how it works, or what it can do, etc. But even if the environmental costs were reduced to 0, and it gained perfect Twine or Inform support, etc. that wouldn’t matter in the least to the central issue people hold - which was spelled out in the initial post. Bluntly, AI generation is against the spirit of IFComp, at least as far as me and many others believe. If that’s not the case - if the mods agree that it’s as valid as the human creations - then anyone like me who disagrees will silently disappear, and the remaining may enjoy the reduced community they’ve created without our whining.

One thing pointed out before in the “AI Slop” blog post is that there is no recourse for me, a (potential) judge, to push back against this. If I summarily give all AI entries a 0, I haven’t done my 2 hours duty, and moreover I’ll be spotted for obvious review-bombing which is also against the rules. If I avoid playing AI games at all, then their final scores will be dominated by people who either don’t care or are outright supportive of this kind of spam. If I don’t participate at all, at least my time wasn’t wasted, but nobody hears my reasoning for why I’m not playing. IFComp is bleeding judges without any direct statement or understanding as to why. (EDIT: I see in the posts above that there’s a survey… maybe I should have looked harder haha. well, I’ll be putting my opinions there, if this thread runs aground)

My final response here is that I see a big post about the difficulty of weeding out AI entries, the balancing act mods must make, that people can lie, etc etc and I will also say to this person: that sucks, but… I also don’t care. I’m sorry. As a lay judge checking this out for the first time, the fact that nothing was done at all to stop this - no matter how “hard” it would have been to do so - to me is akin to IFComp saying they welcome these entries. Is that unfair? Is that placing undue burden on the admins? Probably! But such “there is no perfect answer” objections fall on deaf ears as the Internet slowly becomes overrun by AI slop everywhere, and people like me begin to seek out refuges doing something against the encroaching wave of such spam.

36 Likes
112

Oh one more thing, and sorry for double-post: ironically, I found “Penny Nichols” to be the least offensive of the AI-disclaimed entries, specifically because it did something a bit novel (kind of a hack to turn a chatbot into an IF parser). That’s cute, maybe a banana contender even, though there is prior art in “AI Dungeon”. For IFComp you can do this exactly once, and any further entries of the sort I think should be canned immediately for bringing nothing novel to the table.

Of the other entries, none seem to be doing anything useful with the AI tools, just the usual we now see every day: making “professional” (ick) cover art, content-aware filler for room descriptions, etc. I’m honestly saddened by the one entry that used AI to “fix” their non-native writing, because that betrays a lack of confidence in the author but also the reader. The assumption is that we would see their human flaws and think “nah, I’d rather have the LLM smoothed over version”, which is depressing…

16 Likes
113

Right, I saw this earlier in the thread.

Has this kind of thing been an issue in the past at IFComp? There is some friction to participating as a judge in bad faith (creating an account; rating five games). In any case, it seems a bit early to worry about vote manipulation in this year’s results?

114

Thank you for registering to make this post. I’d quote parts of it to emphatically agree, but the truth is I nodded along to everything you wrote, so I’d just be quoting the whole thing.

Sincerely, thanks.

I also want to say, there are other competitions like Spring Thing and Ectocomp, and jams like the ones held by the Neo-Interactives, which have banned AI. I’d love to see you and your friends’ energies there.

7 Likes
115

Another example of one LLM entry putting someone off the whole comp:

Last year, streamer “kaliranya” was enthusiastic about playing the comp games, and went to the trouble of streaming her playthroughs for others to enjoy.

Unfortunately, she was unlucky enough for the random shuffle to select as her first game the one entry with (declared) LLM-generated content (Hebe), and as a result was put off enough to delete her stream and not even consider any of the other 66 entries.

23 Likes
116

The latest posts from people on trying out IFComp for the first time and then not doing it because they saw the AI-generated art and text had gotten me thinking: is this generative AI stuff actually worth “debating” when it’s pushing newcomers out of IFComp?

My previous post on the matter was about judging for IFComp. I’ve spent a few years in the IF space, so participating in this year was a no-brainer decision. I might personally dislike the LLM entries, but the system does theoretically allow me to ignore these games and just focus on the stuff I care about.

But as I see posts from newcomers and returning members talking about how their personal shuffles are giving them terrible AI entries and then giving up, I started thinking this mindset I have is not something new players should be expected to have. Yes, this is an approach many of us judges ended up taking. But considering cases like kaliranya the streamer, I don’t believe new people and outsiders can talk about IFComp without feeling they’re legitimizing something they abhor.

IFComp 2025 has been surprisingly quiet, even if we consider how niche interactive fiction is. I initially chalked it up to people and podcasts being busy, but I think the added factor of “look at all of these AI art covers” made covering these games especially more daunting. Adding to what I already said, I feel like saying the same stuff about what one likes and dislikes about AI while discussing IF with outsiders might be adding too much stress.

I know what I’m thinking about loudly is going beyond the scope of judging and authoring. It’s about other people outside the IF community looking into us. But I think the public perception of IFComp is at stake if there are no rules discussing what should be done about this technology. I thought it would be fine if the competition made people disclose what LLM technology they used in creating their games, but now I think that without any other supporting mechanism is just bad optics.

I think I’ll revise my “I can’t judge these generative AI games in good faith” stance to “We need to do something about how IFComp is perceived by people”.

29 Likes
117

Yeah, I’ve had some discussion of the past couple years of IFComp in fandom spaces, with people who enjoy IF but aren’t very plugged in to the community and maybe don’t pay much attention to anything outside of IFComp, and the general sentiment there has also been to find it off-putting that there are genAI covers and genAI content in games. Of course, fandom is another hobby space with a lot of people who make basically-unmarketable art for the fun of it and the joy of connecting with other people who share your niche interests, so it tends very AI-negative, but still.

17 Likes
118

You jest, but this was a real point in a New York Times article this July:

Most experts acknowledge that a takeover by artificial intelligence is coming for the video game industry within the next five years, and executives have already started preparing to restructure their companies in anticipation. After all, it was one of the first sectors to deploy A.I. programming in the 1980s, with the four ghosts who chase Pac-Man each responding differently to the player’s real-time movements.

6 Likes
119

This is one of the most exhausting threads I’ve ever read. People are throwing around the AI term and everyone’s definition of it seems different. Thus, constant back and forth. I read a “total ban on everything AI”, then I read oh, but using Grammarly is ok.

It would be extremely helpful to know if “BAN AI” includes every whiff of AI that brushes against your creation. Understand that in a few years, simply the act of turning on a computer will be invoking AI in some way.

Is banning AI truly an all or nothing approach?

  • AI = direct interaction with an LLM and outputted

  • AI = indirect interaction with LLM and then parsed back into the game engine

  • AI = code generation

  • AI = code modification, debugging, optimization, restructuring, language conversion

    AI = beta testing using an LLM

  • AI = generate game content

    AI = game content modification, grammar, spell check, language translation, prose and style enhancement

  • AI = image generation

  • AI = image modification, color and style enhancements, background and image eraser, format conversions

5 Likes
120

:melting_face:

People have already discussed this higher in the thread (ex: victor, tabitha.). There are also very easy rubrics to point to re: how spring thing and ectocomp handle it.

9 Likes