Are there any competitions that ban generative AI? For some reason I was under the impression that IFComp didn’t allow it, but that is incorrect as can be seen on http://blog.ifcomp.org/. I guess my only option is to not play or judge the entries that voluntarily admit to using it, although I don’t recall such admissions in previous competitions. I could at least see the rationale for allowing it if billions of venture capital dollars were flowing in. There are already competitions for generative art for those interested in such things. What if an AI game wins? Will it always have an asterisk next to it, especially if the various lawsuits against OpenAI get decided against them? Will it encourage more and more AI entries until banning AI from IFComp becomes a ridiculous proposition, to the point that IFComp is in fact an AI comp and it will be up to some noble minority to create a new non AI competition? Does IFComp need to become like the rest of the Internet which is rapidly becoming infested with AI generated filler? Even a mildly conservative approach to copyright (or ethics) would dictate waiting until some of the dust settles. Or maybe the dust will never settle and AI will remain a highly divisive issue, in which case the mildly conservative approach is still to avoid it. Anyways, that is my polemical rant for the day. I’m sure tomorrow I will feel that generative AI is the greatest invention of the millennium.
I know all of the jams the Neo-Interactives run (like the Neo-Twiny Jam) prohibit use of any “AI” tools in entries, but in general different events have been adopting different rules and trying to figure things out – the IF Comp disclosure rule is new this year, I’m pretty sure.
(We’ve had a million and one threads about these issues, of course, but personally I’m also unenthused about the use of AI-generated art and LLM output in amateur IF).
How would a non-AI rule be policed? Do you think you’d be able to tell if an author had used generative AI tools to help them with their text?
IFComp 2024 authors were emailed today about this new requirement. I answered the questions about generative AI immediately. I have no generative AI used in my game, not even cover art.
I think there may in future need to be an extra IFComp rule to explicitly cover cases where generative AI has been used but not declared by the author as required.
But it is an evolving situation. And I think the IFComp organisers are managing things as best as they can.
Of course, that’s easy. You just give the text to ChatGPT and ask it about… oh wait.
On the evidence of the various games and other bits of text I’ve read that have used AI, I’d say the answer is generally “yes” – if you’re familiar with what the output tends to look like, it’s not too hard to spot. There are potentially edge cases, though, like where ESL folks are trying to use computer assistance to tighten up their grammar, that can lead to false positives, so I think the transparency policy is very helpful for avoiding misunderstandings, accusations, and other bad situations.
And sure, this generally relies on people being honest in self-reporting whether they used the tools, but hopefully community norms are solid enough that that won’t be an issue – and beyond that I suspect the likelihood that failing to disclose would be detected, and would lead to quite a lot of anger, also makes it more likely transparency will work as intended. But we’ll of course need to see how things play out!
The Neo-Interactives’ events ban the use of generative AI. I am one of the most vocal members about instating its banning in our jams, and pushed hard for the rule we currently have in place.
Most of the time, it isn’t too difficult to tell when it’s been used for generative art (which is the majority of use cases in submissions for our game jams), since it tends to make particular mistakes and aberrations an artist wouldn’t typically (I am an artist, though even to a layman, some lady with six fingers or windows that strain reality in deformed frames tend to be obvious tells.)
People are typically forthcoming when questioned, or it’s provable with a quick back search- I will sometimes dip into their internet presence and see if they’ve also used it on previous projects, or if on the game page/embedded in the game there’s a mention of its use, since that can help make it a bit clearer. Asking to see in progress documents/shots would also help, but we haven’t had to go there yet.
Ultimately, we rule by majority vote- and if the person is evasive and dodgy about questions, or we have serious room to doubt that they’re being honest with us, we reserve the right to remove their work from our events, since it’s very clearly stated in the rules. We’re a small venue, that doesn’t involve any monetary or material prizes, so that makes it a lot easier for moderation purposes.
We have had some trouble with problematic users repeatedly spamming their AI slop into the jam repeatedly, moments after it’s been removed, over and over- until we switched over to our current model of disqualifying pieces and letting them be in that suspended limbo until after the jam submission window closes.
So far I haven’t seen any games that use ai that were fiery enjoyable, except a small few. I thought Lunium from last IFcomp used so but it didn’t at all, and I thought The Kuolema from Spring Thing used exclusively AI images but apparently used mostly stock images and only a tiny bit of ai which is now optional.
So the ai art-using games that I liked didn’t have any ai or just not much of it in general.
The three games I’ve played that have ai text (confirmed by authors) have been pretty repetitive and have had low interactivity (the choice game was a book with “turn next page” functionality, one parser game had only one “true walkthrough” implemented while everything else was implemented, and the other parser game was just a walking tour with no interactivity.
Because of those experiences, I now expect ai-using games to be low quality and not place highly in the competition.
But I could see someone that is ESL using ai for translation help and that game scoring highly. I could also see someone using ai to generate ideas, which would be impossible to catch and wouldn’t be as big a deal to me.
But I’d be surprised if someone could do really well in IFcomp with ai text unless the game would have done well without it too. There are so many components to a good IFcomp game besides padding out a wordcount.
Edit: also, all three people who used ai, I felt like I would enjoy their own writing just as much or more. All three seemed like they could make good if on their own.
I would, in general, like to see games disclose whether generative AI was involved and, if so, how and on what. I’d favor comps and jams requiring such disclosure. (I have no problem with a given comp or jam just forbidding inclusion of assets/text made via generative AI, but in the absence of that, I’d favor requiring disclosure.)
It would certainly be an interesting situation, even if my apocalyptic predictions don’t come true.
I support IFComp’s decision to require disclosure, but I would be disappointed to see judges voting down games for the use of AI. I’d rather see people forgoing such games.
From the IFComp Blog post:
Note that Judge’s Rule #7 still applies:
Judges must make a good-faith effort to play, as intended, every game that they submit ratings for. Conversely, a judge who did not or could not make a good-faith effort to play a certain game must not submit a rating for that game. The competition organizers reserve the right to disqualify any ratings that appear to have been submitted under any other circumstances.
Votes are closely monitored, and disqualified if necessary. Like it has always been.
Yeah, I don’t think I’ve seen anyone talk about down-voting, and as Manon says that’d presumably be easy to detect in any case. But disclosure means that folks who don’t think they can, or want to, evaluate AI-generated stuff in the same way as human-generated stuff don’t have to (or find themselves in the awkward position of learning or suspecting a game uses AI stuff only after deciding to play and rate it).
One point I’d like to make: procedural text generation (“procgen/prog-gen”) has always to some extent been a part of interactive fiction. We’re always striving for responsive NPCs and non-robotic responses and not seeing the same exact text twice, and one way to do that is to insert a sense of randomness into the prose, or develop methods where responses can take into account world-state. The author builds routines to procedurally generate non-repeating responses. The Harbourmaster in BoS is loquacious so there’s essentially a very complicated mad-lib assembling stories he tells, but I as the author provided all the data it uses. I don’t think anyone has a problem with procedural text generation where the author builds it and authors the prose - or response-bits as it were.
Emily Short did a lot of work in this respect with projects like Versu where the goal was characters with conversation topics who might take control of a conversation or NPCs might converse among themselves and be responsive to plot developments and previously discussed subjects. Emily and Versu essentially were weighting algorithms that only used in-game/world information and only improvised within the rules it was given.
That’s a cool thing, and the type of thing AI is increasingly good at doing, however if it’s a “learning” model it’s going to do things the author doesn’t expect. It can become biased. It can decide it’s going to insult the player - and that isn’t the author’s fault, is it? The machine doesn’t care.
I think the primary thing most AI-averse folks would like to avoid is some type of entertainment which includes AI interaction which might “phone home” and relay information it learns to some company or database who wishes to exploit it, or uses the information it learns from one person with another. It would be scary if an AI or LLM could learn to phish personal information by having a chatty NPC have a discussion about the player’s childhood pet’s name, which might be a security response to log into their bank.
I wouldn’t want an IF game to call out to an external AI engine for that reason, and tangentially because once I’m interacting with an AI I’m no longer playing the game and reading the prose the author wrote.
I don’t mind if people use an AI in the writing process or for brainstorming, but if you can’t be bothered to write the story and hand it over to a robot, it’s a waste of time - at least for judging purposes in a competition specifically.
This is my stance. I would be very surprised if a primarily-AI-generated work could place in IFComp. The disclosure rules just mean that people who want to avoid it, can. (Which includes me.)
So since every writer submits their entry to the IF Comp without any sort of revenue expectation, I assume there are also a bunch of digital artists here who also offer their services without any revenue expectations as well? I’d love to employ real person rather than AI to generate images, I’d give them 10% of $0 revenue. Or is a blank cover image a better look?
I’ve actually had quite a few people offer to help me with my cover art before (mostly because they thought what I had was bad). If you kindly and in good faith made a thread asking for cover art help, and maybe included a sketch or mockup of what you’d like, there’s a good chance you’d get help.
I have gotten into art recently but only do pencil sketches. I’d be happy to do something for you like I did a recent game: Untitled Relationship Project - Details but otherwise I’d try just making a thread.
You have also websites like Canva which has plenty of free template to create cool covers.
Note: yes, I know Canva now offers AI-generation, but there are more than enough free non-generated elements to create stuff.
Yup - some authors who are also artists might also be interested in swapping art for testing. Or if there’s a game whose cover art you really like, asking the author where they got it can be helpful too.
For a lot of stories, you could make very fitting covers from a royalty free photo like those on Unsplash, with a bit of text added on.
Oh, Manon shared a link to a whole topic filled with resources like that.