AI in competitions

Competitions
61

The nature of a competition is that if cover art is removed, something else will take its place – blurbs, probably. If you remove the blurb, things will get judged initially on tags and titles. You could keep going until all games are given a number ID and a link and no other context, but why?

As someone who’s entered IFComp before and is (hopefully) entering again this year, I’m not terribly worried about the cover art affecting anything. The entries live and die on the content of the actual games, and if you have a good game then good cover art will, at best, get it looked at sooner. The strong reviewing culture around the Comp plus the long window for judging means that good games will get traction based on word of mouth eventually.

There’s always time for more in game development – more polish, more rewrites, more bug-fixing, more little details. If someone (me) is up at 3am the night before the deadline working on any aspect of the game then that’s a time management issue, regardless of what you’re doing? At least we weren’t removing game breaking bugs at 3am! That was Lady Thalia 2

6 Likes
62

Regarding cover art… ultimately it’s the game that sets the ranking. In the last IFComp, 1st place had very simple (but bold, clear and iconic) artwork that even a non-artist could have made by hand:

image

72nd place was clearly AI-generated:

image

I rest my case.

15 Likes
63

What do you mean you’ve never heard of 128-piece chess where all the pieces are black and some squares aren’t squares?

12 Likes
64

In short: why should I bother reading it if you haven’t bothered writing it?

18 Likes
65

I do have to chuckle a little bit about the scorn heaped on AI cover art, given that one reviewer (who shall remain nameless) said of Galaxy Jones in Spring Thing 2023:

I’m not usually one to harp on graphics in my text games, much less out of game artwork – but holy heck, the cover art for Galaxy Jones makes an impression…

Nevertheless, I’ve sworn off the use of AI graphics in my games.

4 Likes
66

lol I will take the L on that one! This is why I think it’s helpful to separate out the questions on legality, ethics, and consequences from those about quality; my sense is that though it takes some work, it is possible to get something reasonably good looking out of the art generators, while your best case scenario with prose is to barely achieve boring adequacy.

(Though I will say that the Galaxy Jones banner-drop did even more to get me in the space-adventure mood than did the cover art, that was awesome)

7 Likes
67

I certainly was not fishing for compliments; my comments were because I generally find it interesting that people are so certain they would be able to spot AI-assisted text; but I do appreciate your kind words. (I think the constraints of the platforms I write for, in terms of memory limitations, have more impact on the succinctness of the descriptions.)

I’m probably just poor at spotting LLM-generated AI text. I’ve discussed this elsewhere, but because I have aphantasia there is no associated imagery when I read or write. I skip over any descriptive passages in books and only concentrate on action or dialogue. I guess I am always looking at them from a very functional point of view and I’m not really emotionally connecting with prose.

4 Likes
68

Good question! Some of the most beloved IF games, such as Bronze, According to Cain, or Blue Lacuna use stock images with text applied. If you can do that, I’d say you’re golden. There are many free image editors (Paint.NET, Gimp, etc.) that can do it, and some online photo editors as well.

4 Likes
69

This illustrates the benefits and dangers of using AI to write room descriptions. The golden rule in IF is to not put anything in the room description that you don’t want the player to examine. All of these florid details look good in the description, but you’ll have to hand-write or generate a response when the player looks at, smells, or listens. I personally consider these “responses to investigation” to be one of the best parts of IF, so using AI to punch up your room descriptions confers an according higher sense of responsibility to make a description for everything in the room description. It’s not gonna look good if your room has something as intriguing as “jars of preserved horrors” in it but the game goes on to say “I don’t see any horrors here.” if you X HORRORS.

6 Likes
70

I’d like to address some of the questions about what counts as AI or plagiarism.

I think people consider things that give the game artistic merit to be the things they don’t want to see AI write. This means descriptive text and images - the things that are visible to the player of the game.

I don’t think anyone considers grammar services or Inform’s compilation tools to be AI since they don’t result in something “newly created” being added to the game. I’m sure anyone can understand why using a spellchecker on a sentence that you yourself wrote still means that you wrote the “creative substance” of the sentence, because you can still see the hand-typed sentence that was there before it got spellchecked, whereas you did not contribute anything to the substance of an AI-generated image because your hand-typed prompt isn’t visible in the image.

What people who are against AI ultimately care about in plagiarism is not the idea that a creative work could have been influenced by another, but that those prior works are not given credit and it is not possible to avoid accidental plagiarism to as high a level that handwritten or hand-drawn media can.

If you notice that your story is sounding like Lord of the Rings, you can either change the story so that it’s less derivative, or you can post a dedication to J. R. R. Tolkien in your credits. If you AI-generate a story that was inspired by Lord of the Rings, both you and the AI won’t realize it and so you won’t be able to see that there is anything that needs to be made more distinct from LOTR, and you won’t be able to credit Tolkien either because you didn’t know the AI had his works in mind. Anti-AI people hate this “black box” method of generation where there is no due being paid to prior writers or artists despite their works being used as training data for the AI.

As someone with a master’s degree, I obviously consider it to be a priority to cite your intellectual antecedents and to be able to write a sufficiently-new work that does not resemble theirs so much as to count as plagiarism. Writing stuff with AI scams you out of having to be well-read enough that you can come up with stories all by yourself without it being a copy of any one particular story. You might think that getting to read less doesn’t sound like a scam, but it is. There’s a reason that people consider Tolkien’s work to be better than the written-by-a-teenager Eragon, and it all comes down to Tolkien having been so well-read that he could synthesize all the myths and stories he knew and boil them down into a new story that does not overly resemble any one of its parents. I think that’s as good a retort as any as to why AI writing shouldn’t be used in IF. Over time, the community of readers - whether paper or IF - can sniff out the wheat from the chaff.

8 Likes
72

For as long as the AI companies like OpenAI continue to train their LLM models on data scraped from the internet, there will still be massive ethical problems. The only way I can see to use an LLM ethically is to be personally responsible for its training dataset.

Using genAI to write posts on this forum is not only played out, it’s also not permitted.

10 Likes
73

I think a big thing I’m seeing here is people confusing the signifier with the signified, or mixing up correlation with causation—taking something that good games have in common, and assuming it’s what makes it good.

An example in public life is blue check marks on Twitter. Once, their purpose was to verify who people were to prevent impersonation. They became a status symbol that some people sought after and others made fun of, but they represented being a “real person”.

Elon Musk changed it so anyone could buy blue check marks. They were instantly bought by tons of porn bots and trolls, to the point where a blue check mark response is now a sure sign of a scam or an especially stupid take.

It’s the same thing with ai cover art and text. People see that several well regarded games have good cover art. But the good art didn’t buy scores or popularity, necessarily; instead, good games come from authors who care about their final product, invest a lot of time in it, and make sure every checkbox is ticked off, and that includes cover art. It’s just a sign the author cared about every detail. Adding good cover art to a bad game won’t accomplish very much.

Similarly, ai produces grammatically correct sentences that use an educated vocabulary. In the past, that was a sign of good writing, since if someone was practiced enough at writing to use good vocabulary and took enough time to avoid typos, they probably had interesting thoughts and good arguments or plot.

But now with ai you can have glamorous cover art and excellent grammar and still have an awful game.

Edit: I type on my phone a lot and have lots of typos and this has led people to judge me in the past, so I believe what I said above to be true. I’ve used grammarly before a lot on posts to find spelling typos. I don’t use it to rewrite sentences and ignore things outside the errors tab.

More edit: in the end, it could end up just like blue check marks, with art and text giving a reverse significance from before. “Detailed cover art? Smooth, polished sentences? Nah, this game is going to be awful.”

26 Likes
74

Honestly, the notion of cover art for If feels kind of alien to me.

As for the ethical side of things… honestly, I’m not sure how the way LLMs are trained is fundamentally different from the way human writers train…

But yeah, if you can make a mechanically amazing game with awesome puzzles, but suck at narrative, perhaps it’s better to just have a generic excuse plot or even just make the game an abstract puzzler with no story than to tack on an AI story.

1 Like
75

Here’s a list of considerations about generative AI in no particular order, along with some random thoughts…

  1. Ethics of training data. Should authors disclose the algorithms and training data they are using?

  2. Authorship. If a computer cannot be a sole author, why can it be a co-author?

  3. Disclosure. How can you even prove if generative AI was used in the creation of a work? What should be the punishment for violating this? I’m not sure I would advocate for harsh penalties and lifelong bans, especially for 1st time, young, or otherwise not too bright offenders.

  4. Quality. I’m not sure this is a great criteria for resolving this issue. I suppose if the competition got inundated with spam it could be an issue, but that would be true even if good quality spam, if that even makes sense?

  5. Community. How much should other principles be compromised for the sake of community maintenance? Would banning generative AI create too much of a headache for the organizers?

  6. Toolchains. Ie using generative AI for coding, brainstorming, translating, etc. Use for brainstorming is probably not a good consideration for acceptance/rejection, even if it personally rubs me the wrong way. Translation is itself a difficult and creative process, especially when applied to difficult and creative texts, so I don’t really see it as the simple tool others have suggested.

  7. Procedural generation. As connoisseurs of interactive media and as computer programmers, I don’t think anyone here is opposed to procedural generation per se, so the question lies in some other consideration. I’d hate it if someone created an amazing holodeck level game that didn’t violate any of the other considerations and it was banned.

76

I’d rather not have IF Comp restrict or require disclosures of AI use unless and until it becomes a blatant issue.

The same goes for pretty much anything else. I don’t think pre-emptive rules and requirements are a good idea. I appreciate when IF Comp organizers take the time to decide when something is acceptable or not as an edge case.

If the people running IF Comp decide AI is a overwhelming issue, though, fair enough.

Personally, I don’t use AI in my recreational writing. I’ve used AI-powered text to speech services in one game.

77

How Is this pre-emptive? I don’t think people were having this discussion prior to the explosion of generative AI, and the actual use of it in games and competitions.The new inclusion of the disclosure requirement strikes me as reactive, though as to what specifically they are reacting to - you’d have to ask them. Evidently some people consider the use of some kinds of generative AI to be problematic in and of itself, without requiring some other problem in addition.

4 Likes
78

Without going into details, there have been incidents where non-disclosure of generative AI use caused issues in the recent past, and there’ve definitely been people asking for a requirement like this (I made that suggestion in the post-Comp survey last year, and I really doubt I’m alone). So yeah I think it’s very much responding to real stuff that’s happening rather than theoretical shadow-boxing.

9 Likes
79

That’s not the sort of AI that IFComp has an issue with, to my understanding.

My rule of thumb is, if you can expect to put in the same input and get more-or-less the same output every time, it’s not the sort of AI that these rules are about. (Even if companies are now trying to slap the “AI” label on everything.)

6 Likes
80

Yes. If I get tired of writing my fortieth “you are in an underground cave” description and I go to Chat-GPT to help me describe one and I copy and paste what it wrote into my game, that’s co-authorship I should disclose.

That’s different then having a conversation with Chat-GPT about “what are some unique ways to describe caves?” and using that as research “hey, I didn’t know the difference between stalactites and stalagmites!” and learning some new vocabulary and then based on what I’ve learned using that to write my own actual prose. That’s similar (and likely not as good at) just googling writings of spelunkers to get a feel for the vocabulary and nomenclature.

I don’t think people are calling for “punishment” for using AI, they’re just saying it needs to be disclosed. Just as you’d disclose source material that wasn’t yours. In Alice Aforethought I lifted prose from Through the Looking Glass to the point that the credits list Lewis Carroll as inspiration and co-author.

That’s one point I keep seeing - nobody is necessarily saying AI generated prose is “bad” it’s just not technically written by the author whose name is on it. Just like an image generator that creates faces or distinct objects had to get the source imagery from somewhere and that somewhere is another real artist or photographer.

Again, it’s not to a banning offense. There is such thing as a derivative work - like fan fiction or remix or homage. It’s just a bit shady to say “Judge my prose and award me a prize” when your creative input was telling a machine to write something based on a prompt and hitting a button. IFComp isn’t saying you can’t enter a game like that, you just need to dot the Is and dross the Ts for attribution.

I posted about this before, but proc-gen has always been a part of IF. Procedural generation involves source material prose bits the author created or curated and the author created the algorithm or routine that creates the text in the game. I’ve build procedural text in games - the coffee shop in RSPM has ambient messages of baristas calling out orders for people to pick up, and that routine grabs a random customer name from a list, an adjective from the list “sweaty/busy/attractive/angry/…” barista and varies a couple other things. I made all those lists of things - I will cop to using resources like a baby-name generator to get ideas for names to include, but those are public resources designed to provide inspirational lists of names - I didn’t use every name on the. list, I picked the ones that were appropriate.

Procedural text can be as simple as "'Sure, let's go,' Bob said [one of]annoyingly[or]encouragingly[or]sadly[or]enthusiastically[in random order]."

6 Likes
81

Based on the replies, I guess I’m unaware of what happened in past events. I played and reviewed a few entries in IF Comp (and Spring Thing) over the past year, but admittedly I didn’t go through all of the entries.

If any organizers think the problem is big enough that requirements are needed to address it…that’s their call.

But, again, if the judges can handle individual issues easily at this point, I’m against disclosure requirements due to popular demand alone. (Since I’m in the minority, though, I guess this is inconsequential.)

I don’t particularly like most AI-generated content, but I also see community enforcement around it as too strong in general.

I won’t go into my reasons too deeply because I know my position is unpopular. Objections to AI content are often related, directly or indirectly, to how people profit from and own their work. I’ll probably never see eye to eye with other people due to particulars of why my writing outside of the interactive fiction community is profitable.

I’ll add that I’m grateful for IF Comp’s allowance of Fair Use/Derivative Works. I was really happy with how the organizers handled things when I said my past entry would fall under this category. Fair Use isn’t directly related to AI-produced content but it’s something I hope to see continue throughout any IP-related debates now or in the future.

4 Likes