Can we just ban AI content on IFComp?

Competitions IFComp
121

Before I go any further, I do think in the spirit of this competition which historically has showcased small/single developer, handmade games, AI generated cover art shouldn’t be used (and definitely not wholesale AI gen text which is what is being mainly judged.)

Moving past that and why people do want to use it though, AI images have been rapidly improving at a scary rate. Just a few years ago I’d laugh at how anyone could mistake AI images for reality with the 15 fingers, unattached limbs, and generally really wonky proportions/settings/colours etc, but now, I sometimes have to really look, especially if there is someone artistically inclined generating the pictures and then fixing any of the worst AI glitches. Sometimes, the main giveaway is just bad lighting/shadows (AI still seems to often really struggle with that.) I almost got caught out using a small black and white line art image last year that was labled “not AI” on a stock site in a photo manip, but after starting to work with it, noticed a few iffy details and- yep, when run through a detector showed it likely was AI generated which was frustrating and meant I needed to pull it out and start again.

I’ve seen two instances now of where people have claimed to have handmade artwork and gotten away with it for a while before anyone actually pulled them up on the blatant AI use, and they were even getting some praise for the art as well. (They both had hallmarks of AI, but you needed to know how to spot them, and more than that I understand people not anticipating that they may just be being lied to). One at least did appear to be using the AI art and pretending it was their own for an ego boost (which I’ve never understood, but it used to happen with plagerism for praise for work that wasn’t theirs, and now I’m seeing it with AI gen work passed off as handmade). I’ve also actually being seeing a lot of mislabled stock of late. Even people trying to avoid AI can face an uphill battle and you have to be extra, extra careful and check all your images, and if in doubt, run them through a tester like hivemind. From some discussions I’ve had, some people find any kind of photomanipulation intimidating (but I really do recommend anyone wanting to make their own covers to give it a go!)

It’s also AI is quick and easy. I mean to find an image you’ve often got to sift through a whole heap on stock sites which takes time, make sure as best you can that the image isn’t stolen, and the CC is as it should be for use, crop and add your title at a minimum. (And the amount of work can go up from there if you want to combine images, colour correct etc.) With AI you type in a prompt, maybe ask for a refinement or two, and it spits out what you asked for, looking somewhat like a comissioned piece so it’s very quick and easy.

Anyway, I can see why some see the appeal in AI generated cover art- being quick, simple, easy, and arguably prettier than just a title on a screen for those that can’t find a stock image to suit. Over at another forums, I posted a list of some of the websites that offer free to use stock as a resource list quite a while back (and could also include some free to use image editors, there’s even one that has a fair amount of photoshop’s functionality called Photopea that is free to use). Maybe running up to the next IFComp, something like that could be posted to help give people the resources to make their own if they wish? Especially given choosing to use AI cover art will likely have a fair few people just skip the game from the sounds of things so next year the inclination to use it may be lower.

4 Likes
122

I imagine the IFComp organizers will come up with a solid policy, but personally, my line is that LLMs should not be used for creative purposes in making a game. In other words, if you’re expecting the output to be more or less the same every time you press the button, that’s okay—Grammarly should flag “he are coming” the same way every time I open a document, and Google Translate should say “bonjour” every time I ask it to translate “hello”.

(Of course, proper translation and editing work involves plenty of creativity, but I don’t expect that out of Google Translate—I expect Google Translate to give me the literal words in a fairly stilted way that lets me figure out what they say.)

But if the goal is to put in a prompt and have a LLM create something new and creative based on it, that’s going too far (in my personal opinion). That’s the point at which human creativity is being replaced.

17 Likes
123

It doesn’t help that “AI” is seen as a buzz word in the wider community (outside IF space) and is being slapped on a lot of things that have been around for a lot longer and are not GENERATIVE AI, because AI is cool and cutting edge, and buy into the hype that our program that totally has this new fangled technology :stuck_out_tongue: . It is a pet peeve of mine and muddies the waters about exactly what the types of AIs are and if they are generative or non-generative, and what they’re being trained on.

And then there’s degrees of generative AI. Like I’d personally put a story generated by ChatGPT from a prompt, into a different category to someone using grammarly to clean up their text they have hand written a bit, or google translate to translate a few words here and there, even though both I believe do use gen AI, just to different extents and with a different focus. (One to replace the writer, the other to try and improve human work.) It can be argued over things like when grammarly rewrites entire sentences for people, if that’s still truly their own work, but this does likely start getting into shades of grey with where opinions lie on its use.

I agree this is probably a good dividing line as any if we’re trying to draw one that is relatively clearly defined.

5 Likes
124

Tangent, but: I don’t think the IFTF should implement a rule to ban people from using Grammarly but if I could, personally, make everyone stop using it I would. Please stop using Grammarly, these ‘grammar correction’ models are actually turning all your writing into a finely homogenized slurry. They can’t tell unusual usage, creative turns of phrase, off-beat vocal rhythms, or in-character reported speech from ‘grammatical mistakes’. They’re optimized for writing emails that instantly put the reader into a deep slumber, not fiction that’s meant to hold my ADHD-ass attention.

I think most people using these kinds of apps are doing so because they have low confidence in their writing from a formal standpoint of writing ‘correctly’, which means they don’t know how to push back on the 10-20% or so of suggestions that these apps made that are actually quite bad in the context of creative writing. Grammarly isn’t exactly snake oil, but it’s kind of like… it’s one of those products that’s largely marketed on people’s underlying insecurities.

Edit: Also, to be clear, absolutely not your fault if you’re lacking confidence when starting out or feel like you need help with ‘grammar’; I think a lot of the way people are taught to write and a lot of people’s relationship to writing gets warped by this sort of ‘prescriptive grammarian industrial complex’. No one is obligated to just start doing this and feel 100% sure about everything from the word go. But I think Grammarly is actively bad for your development as a writer, even if it results in the work you do publish early on having fewer terrible comma splices in it.

36 Likes
125

@Sequitur May I show my writing classes this post? This is just a perfect articulation of why Grammarly is the devil.

But I also don’t think Grammarly should be banned from IFComp. As a person whose professional writing has been used for LLM training and, along with my wife, might very well be getting settlement money from the Anthropic Settlement, I hope all of these Gen AI companies implode in an internecine economic armageddon. But your mileage may vary.

If we were using Robert’s Rules here, at this point I would Call the Question and put a proposal on the floor to ban creative use of Gen AI from any game entered in IFComp. “Creative” would include any part of the writing process. Spellcheck, grammar check, and translation would be allowed. From an artistic perspective, I’m not sure if it should include coding and hope people who know more will continue to weigh in.

Anyway, I feel like there’s a clear consensus building around several key points. Is the survey the place where this matter will be decided?

14 Likes
126

Can it please be pinned to the beta testing topic that when asking for testers you should say if the game uses generative AI? Debugging someone’s generated content (which may be more work than actually generating the content) seems like something some people may not want to do.

14 Likes
127

I would like to know whether ‘brainstorming’ or ‘coming up with names’ counts as ‘creative’ in this definition.

4 Likes
128

I would say yes. Ted Chiang, in his takedown of AI, specifically notes that brainstorming is one essential place where writing happens: “An artist needs to have control of every aspect of a painting. A writer needs to have control over every sentence in a novel.” Names might seem like a more innocent use-case, but I would argue that names need to be extremely carefully chosen, and the process shouldn’t be short-circuited by a “good enough” list generated by an LLM.

Of course, others may disagree. This is just my initial thought.

10 Likes
129

IMO, it only makes sense for a rule to ban certain types of content from actually appearing in submissions—e.g., disallowing text and images produced by genAI tools. Putting aside my personal feelings about the use of genAI for this purpose, I think telling authors how they can and can’t brainstorm would be outside the scope of what competition rules are for.

10 Likes
130

My immediate thought is that it counts as a creative contribution if it’s the sort of thing you would credit a human for. Writing game text yes, brainstorming a list of names no.

But that’s less because I think LLM brainstorming is a good idea, and more because it seems like an easy guideline to understand.

8 Likes
131

If the goal of the ban includes quality control, I’d recommend banning the use of generative AI for writing code as well (with possible carve-outs for using AI tools for debugging or as an API reference). Snippets blindly copy-pasted from Claude et al. are sure to contain bugs or unexpected interactions.

3 Likes
132

Sure, you might also be interested in this post by my good friend and fellow writer Laura Michet.

10 Likes
133

Possibly given you’re asking the AI to write you ideas/outlines/whatever based on prompts you’re giving it, but there’s absolutely no way you’d be able to police it. Only the author would really know for sure what their inspirations were behind a story. I think it’d be pretty useless even trying to ban something like that given it’s often difficult enough to prove if AI has been used in passages of writing. Trying to work out if someone’s just used established tropes or got an AI to write an outline I think would be all but impossible.

@Sequitur I personally don’t use Grammarly. I know a lot of people that do though. I did try it out for a while just to see what all the fuss was about and found the aspect of it in the web browser that seemed to want to obsess over what was in emails odd. (And on top of that, I don’t love that you’ve got a generative AI program getting access to potentially personal or restricted info in emails on work systems). But then again I am a native English speaker and my emails are readible, and generally I remember not to send off emails when someone’s managed to make me really upset without walking away for a cool down first. It is kind of crappy if they’re aiming this at the general public to make everything homogenous, rather than just to increase readability or for when you really need a second opinion on the tone of an email. But they need to get people either hyped or insecure somehow to sell those subscriptions :roll_eyes:

5 Likes
134

Yeah the Grammarly ads are really shocking to me because it’s like, it’s using the same playbook that’s been used to sell hair loss cures and weight loss products and certain skincare products and so on and so forth forever but for the concept of “writing emails good”.

7 Likes
135

I’ll certainly be on special lookout for the post-comp feedback survey, if that is really the most important venue for feedback. But while I’m here, I’ll add my 2 cents in the [unlikely] chance it matters.

Setting aside environmental, social, and ethical concerns (I know you really can’t, but let’s just pretend), I don’t think there is inherently an artistic problem with using LLMs to create work, so long as the application is novel. Most of the time, it isn’t. The entire value proposition of an LLM is to generate something you would expect to see. But I’m sure there could be interesting and neat ways to use that expectation to actually defy expectations. You’d probably have to think on it a little while, and it wouldn’t just be pasting a prompt into a box somewhere. Nevertheless, I do think we have something of a duty to stave off poorly executed use of LLM generated text wherever possible, because it’s coming for all of art fast and hard, and this is an independent, small, and pretty well controlled environment where we actually get to have a say. (And also because of all those environmental, social, and ethical concerns). If people want the chance to be experimental and creative with LLMs and IF authoring, let some other event crop up to serve that audience.

That being said, as a player, I really could care less if code specifically is LLM generated or not. I won’t be reading it and I really only care if it works. Probably no one would ever figure out if you used it, especially if you used it do something fairly pedantic that would come out exactly the same if you knew the language better anyway. I think it is only practical to call for a ban on LLM generated text and images.

Also, special shout out to kaemi, glad to see I’m in the company of other energy markets nerds.

12 Likes
136

As are snippets blindly pasted from Stackoverflow, or from forum posters, or from anywhere else. If the concern was only quality (I know it isn’t) then I think AI generated or assisted code, when used by someone who otherwise knows what they are doing, are no worse than most other sources of code help.

There are other reasons to consider AI code to be bad, of course. One of which is creative. While we have a tendency to focus on the text or visuals of games, the code of many IF games is a key part of the experience, especially on a conceptual level (the why, not just the how). If an AI designs the mechanics of your game, or implements them for you, that’s just as much of the creativity given away as if it writes the textual content for you.

There’s probably an extremely hard (or impossible?) to enforce degree of judgement here though. If an AI writes all the code for my game, it’s not really my game and I don’t think it should be in the comp. If I ask an AI to fix a single function that I’m struggling with, then it’s not quite the same.

4 Likes
137

Thinking about the point others have made that the obvious, eye-searing AI cover images in the list of entries may be preventing IFComp reaching a wider audience, it occurs to me that I emailed GamingOnLinux to let them know about this year’s event and never heard back. I’m now wondering if the AI slop is why: they’ve got a pretty clear stance against it.

I don’t buy the idea that popularity is always good and the goal should always be to draw in as many people as possible, but there’s a massive difference between accepting that a niche event with a clear focus is valuable in its own right and running something in a way that’s distasteful to the people who do discover it.

Long story short, I went into this thread thinking “I have reservations because a ban might rule out some hypothetical groundbreaking submission in future” and now my bigger concern is that we’re probably already missing out on groundbreaking submissions (either because potential entrants are put off by the presence of AI submissions, or because it’s limiting the promotion that would have led them to the event in the first place).

24 Likes
138

As a practical matter, if enough participants feel as strongly against generative AI as the consensus in this thread, I agree that alone may be enough reason for a ban.

But I struggle to understand the position of people who see a few games clearly labeled as using gen AI, and then refuse to play any of the majority of games clearly labeled as making no use of gen AI. There are many games that I know I won’t enjoy in any given year of the competition, for any number of reasons (theme, medium, content warnings, etc.) so I just… don’t play them?

Disagreements about what constitutes IF and what entries should be allowed in the competition (complete with threats of boycotts and bold pronouncements that new types of entries will destroy the soul of the artform; drive off newcomers; thrust the competition into disrepute, etc. etc.) are as old as the competition itself. I participated in some of those arguments and have come to regret many of my past intemperate positions. Many very smart people in this thread have put forth arguments for why This Time It’s Completely Different, but I’m just not convinced.

6 Likes
139

I agree. Too many fanatics zealots on both sides of the aisle.

Perhaps forum admins should implement an AI-topic ban (there’s certainly enough historical discussion about this subject to last a lifetime). :wink:

1 Like
140

This is something you have to look outside the community for, where AI has become almost synonymous with low effort and low quality. For these people the presence of AI is a red flag for the quality of the entire competition if they’re not already familiar with the IF scene.

11 Likes