AI in competitions

Speaking as someone who has both posted stories he wrote himself and stories he had an LLM write and then edited into something more coherent than the LLM’s raw output, I have always labelled the latter as such and would consider it dishonest not to do so. Requiring disclosure just strikes me as mostly a reminder that people should be honest. Not sure there really is a way to enforce that honesty beyond an honor policy though, especially with how it’s becoming harder to distinguish LLM output from technically sound, if uninspired human output, especially in small bits.

And yeah, intentionality and quality judgement is what distinguishes the best human writers from the best LLMs at present, and perhaps a curated data set containing just the writings and collected letters and notes of the greatest writers instead of everything on the internet indexed by Google or Bingwouldn’t be enough to make up the difference… but that’s comparing LLMs to great writers. If we compare LLMs to more middling human writers, the kind who have binged thousands of hours of junk television and read dozens or hundreds of pulp quality novels and short stories, but have barely touched the classics… they might have a better understanding of what makes a fantasy novel a fantasy novel than a current gen LLM does, but no idea what makes The Lord of the Rings a great fantasy novel, and no idea how to write their own stories with intention or how to decide when a draftis of acceptable quality and when they need to rewrite something from scratch. Perhaps not exactly the same, but I don’t think LLMs using statistical patterns is all that different from the mediocre writer rehashing stuff they’ve read before without understanding what does or doesn’t work for the tropes they’re using, couldn’t spot a plothole if it bit them, and doesn’t know how to make characters that are people who react to events instead of automatons that do whatever the plot demands… And honestly, I’m not sure I can confidently place current LLMs as low as the 50th percentile of human writers for quality… Almost certainly below the 90th percentile of Sturgeon’s law, but there’s a lot of writing out there that isn’t even grammatically sound, and even the absolute toy LLMs can manage that.

Okay, I just have to get this straight.

Who here actively opposes the disclosure agreement? Not any bans, just the disclosure.

And if so, how come?

7 Likes

Yes, I’m opposed to mandatory disclosures for the reasons I described above.

Tldr: the reasons for this aren’t set out well enough, and the issue isn’t severe enough yet, IMO. Others disagree.

For me, the the rule could affect my future comp entries.

If anyone really cares about how I'd change my habits, expand...

Mandatory disclosures would discourage me from using any tools labelled AI because I get little control over the nature of the disclosure (though the 4-point disclosure model is a start, I guess).

I’d probably switch to non-AI tools to avoid being harshly criticized or overlooked, which I guess is a victory to people opposed to the use of AI tools.

I think my work would be of poorer quality because I only use AI tools when I believe they are the best thing for the job. Despite my misgivings to AI output, some useful tools are branded AI or use AI in some way, and those fall under disclosure requirements, despite what some people in this thread have said.


I have no issue with voluntary disclosures. Also, disclosures aren’t the same as credits and attributions. Personally, I credit most tools and resources services that I use, but I would not mention whether they are AI-based if given the choice.
I also wouldn’t emphasize that information to help the audience prejudge my work.

2 Likes

I support the disclosure but think it was inappropriate to introduce this change so late in the timetable. It should have been done much earlier or announced for next year.

5 Likes

I am not against the disclosure agreement in general, but I have some concerns.

Was used to generate text used in this game (use of spell check/autocomplete excluded)

If someone writes a passage of text and throws it into chatgpt to fix up the spelling or extend the story, does it fall under the excluded categories above? Furthermore, tools such as Grammarly utilize AI technology. Does that count as using AI?

To be clear, I do not plan to do any of the above. I’m just thinking this is something which needs to be clarified.

Was used to generate my game’s cover art

This can be tricky. I (and probably others) use public domain assets to create cover art, but I’ve increasingly been seeing AI art flooding public domain art resources. I have increasingly been using AI-art checkers to filter out the AI stuff as it’s just getting harder to tell nowadays.

Painful truth. AI art technology has been improving at the speed of light. Soon, it might be impossible to tell.

If someone uses AI art, but honestly wasn’t aware of it, how would IFcomp handle cases like that? Let’s say they downloaded it off a public domain image resource, and it wasn’t marked as AI?

4 Likes

Misinformation from marketing trends does not dictate whether a tool uses neural networks in a way that would warrant disclosure.

I really doubt hiding the generated nature of a work would do much to boost ratings at all, anyways. Generated prose and artwork are not as good as users seem to think.

Either way, I think it’s more ethical to the audience for you to disclose the nature of generative “AI” in your works, policy or no policy, so that players can make an informed decision of how to spend their time and choose their experiences. Your audience is not your enemy.

Also, it should go without saying that the consequences of being discovered trying to suppress or lie about this information could have serious social consequences. Think of it like a dev being discovered trying to intentionally hide content warnings or quietly delete bad ratings/reviews/comments.

If an anti-AI player finds out, post-playthrough, that you used generative “AI”, and that you suppressed this information, they’re not gonna just chuckle and say “Aha, dang, you got meeeee lol”. I don’t really know how you imagine that working out.

Yeah, but that’s still not happening in the next 12 months (I guess I’m placing bets now lol), and when it does happen, then competition policies will be adjusted to something else. The current policy is for the time until that moment arrives.

13 Likes

Misinformation from marketing trends does not dictate whether a tool uses neural networks in a way that would warrant disclosure.

The competition organizers and audience decide that, which is the problem.

Someone in this thread said they believe AI-powered text-to-speech would fall under the disclosure. However, I would absolutely disclose that under a mandatory rule in 2024.

In practice, in 2021, I cited the name of the service, but didn’t mention its branding (deep learning at the time, now it’s AI). I didn’t mention it because I didn’t think anyone would care at the time…and nobody did.

That’s not a defense, it’s a recognition of the changing community stance.

Think of it like a dev being discovered trying to intentionally hide content warnings or quietly delete bad ratings/reviews/comments.

I don’t support mandatory content warnings, either, even though I read them for other games and I sometimes include them in my work. If IF Comp organizers demanded specific content warnings or specific ways of displaying them, I probably wouldn’t submit work to the competition a second time.

Where my exact boundary is…I don’t know.

Nobody here is in a position to delete comments anywhere AFAICT so that’s irrelevant.

I really doubt hiding the generated nature of a work would do much to boost ratings at all, anyways. Generated prose and artwork are not as good as users seem to think.

I don’t really know how you imagine that working out.

It’s not about ratings – it’s about being dismissed at face value. If someone is upset that I’ve used an AI tool, or feel that I didn’t have adequate content warnings, or have any other issue, they’re free to give my work the lowest rating possible, whether that’s in IF Comp, on IFDB, or otherwise.

It’s only deceptive if audiences are entitled to know certain things, which I don’t agree with, and which is why I oppose mandatory disclosure rules.

Simply put, I’d rather pay the price by receiving low ratings than by giving up freedom in how I present my work.

(And losing freedom in how I present my work would stop me from using some tools, from including some content, or submitting to some events. Again, this would be a victory for people opposed to AI’s use, and one I’d respect even if I don’t like it. Just like how I respect low ratings when I get them.)


I’m muting this thread since I’m thinking and writing too much about an event I’m not even entering this year. The last thing I’m going to say is I intend to comply with any organizer decisions going forward, in case that’s not totally clear.

1 Like

Which, I would imagine is why you’d want to disclose that your prizewinning blockbuster of of an interactive fiction was generated by a machine. At that horizion, what’s the point in humans having contests at all? If the AI can do it better and just churn out literature so we don’t have to, what’s the point in humans entering a contest?

If you had a co-author helping you write something, you’d credit them. If you have an AI co-authoring, you should credit it.

Stupid Hanon metaphor: [1]


  1. If you build a pie-making machine to win a pie-baking contest, shouldn’t you actually enter a contest about making machines instead of pies? ↩︎

8 Likes

We’ll cross that bridge when we come to it. :innocent:

I write trashy fantasy gamebooks (including my upcoming IFcomp entry). So I’m at greater risk as that’s the kind of stuff AI will come for first. :upside_down_face:

If we use public domain elements to create cover art, do we need to credit the exact sources? I did mention that I used CC0/Pixabay art assets for my Spring Thing cover art. I’m also concerned that writers who are less discerning about art may unknowingly use AI art assets they assumed were legitimately CC0 human artworks.

100% agreed. I never had any arguments there.

Full Disclosure: I did not use AI generated art or text in the Spring Thing, or my upcoming IFcomp entry. I check all art assets I use as far as possible to ensure that they are not AI generated.

1 Like

This was supposed to be an analogy to how this happens in other dev spaces (like YouTube comments being removed on game trailers). I wasn’t citing something happening in the IF scene. I was saying it would similar to a response to that kind of action.

As for the rest of your response, I still don’t agree with you at all, but you’re certainly consistent, so I don’t have anything else to say to you. :woman_shrugging:

3 Likes

I’m gonna be honest with you, as long as proper contents warnings are displayed, you’re fine. I don’t think anyone wants someone to be triggered by something that could have been easily avoided by simply saying “hey, this contains [xxx triggering thing].” So this example kind of falls flat to me.

Look. If you didn’t write the work, I’d like to know that. I think the concept of having a “ghostwriter” in general is deceptive, whether it’s another person or it’s AI. It’s like saying, “oh look XXX made this” but they didn’t, and I’m praising them / criticising work of theirs, and it wasn’t even them.

7 Likes

Repeating what others have said:

  1. The rules are changing a few weeks from the deadline. That is almost universally not cool in any type of competition.

  2. I believe if there is any rule about disclosure of any type of computer assisted development, regardless of whether or not it is deemed “AI” or not, it should apply to every tool. If I used Grammarly to correct my sentence structure, should I not confess (in fact, from what I can tell only spell checking is allowed, grammar checking is not listed on the form as allowed).

  3. IFComp has not sufficiently defined “Generative AI” or listed specific tools that would require disclosure. I suspect that is due to #1.

  4. The form is just a series of check boxes, last time I looked. I didn’t even see anywhere for the author to explain the extent of their use of AI.

But most of all:

  1. This means people will prejudge works of IF without even playing them. This means regardless of how good a work of IF might be, if someone used “generative AI” (again, not thoroughly defined) for even a tiny fraction of it, some people will just outright not play it.

I am in general not opposed to disclosures of the use of authoring tools, I just think the way it is being done now is hasty and unfair.

-v

6 Likes

But… we all already do this, for a variety of reasons. We all make decisions about what works we’ll engage with or not based on exterior factors like the blurb, the content warnings, what event it’s a part of, etc. It’s okay – good, even – for people to use this kind of information to make informed decisions about what works they engage with.

10 Likes

It’s simple. CREDIT YOUR ARTWORK. If you used your kid’s pre-K doodle, credit your kid. If you use your friend’s photograph, credit your friend. If you used Canva to add text/elements to that photo, credit Canva. If you used a Botticelli painting, credit old Sandro. If you used Midjourney, credit it. It’s standard procedure. There’s nothing weird about disclosing how you made your art, and the rules of most comps and of IFDB require you to disclose where it came from.

As far as AI-generated text goes, I’m not sure why anyone wants to do that. Isn’t the point of writing IF to write IF? It would probably be fairly obvious if you did that, and your game probably wouldn’t be great. There’s a huge difference between getting ChatGPT to write your descriptions and using spellcheck and we all know it. But again, if you used a poem or an epigraph in a game, anything that you didn’t write, you should credit that. I don’t see why AI-generated text is any different. The rules and conventions are already in place.

I agree with @alyshkalia here. Nearly everybody picks games according to their whims. Great cover art matters. Great blurbs matter. Genre matters. If I see that your game is a dating sim, I’ll probably skip it because I think that’s boring. There’s absolutely nothing about this rule that is any different from requiring CWs-- some people won’t play games with violence, or sex, etc, because they object to that. Players deserve that information and that choice. Why is this any different?

17 Likes

Looking it up, I think that Grammarly does not use AI. It’s apparently allowed in many university courses where using AI can lead to misconduct. But, I might be wrong. I’ve never been to university.

1 Like

Re: Grammarly, they have their own page about how it works:

-Wade

3 Likes

Citazione

I have no problem in credit the AI engine I used to generate my art cover.
On the other hand, I find it embarrassing to have a prominent label pointing to the AI image as if it were something that could harm someone.

3 Likes

I think Grammarly has an AI writing tool. (LINK) But I’m not an expert and do not use Grammarly anyway. Someone correct me if I’m wrong on this.

I created art under the terms of the CC0 and Pixabay license, which do not require attribution, but I will write up a full list of credits for all stock art assets used in my game if I do participate in IFcomp. Using “if” because I’m behind schedule and may not be able to finish my entry on time.

Honestly, I am not promoting the use of AI and I am not opposed to the disclosure agreement in general. If I gave that impression, I apologize.

1 Like

Oh, right. My bad. :slight_smile:

1 Like

I don’t really have a dog in this race since my WIP doesn’t use AI (and besides it won’t be ready in time this year :frowning: ), but here’s an idea.

Maybe the rule should be that authors have to disclose ALL the tools they used to build their game. That way AI tool use is disclosed for those who don’t want to play games that use AI but not in a stigmatizing way since it will be included among a list that includes text editors and outlining software. :person_shrugging:

5 Likes