Hi! I am working on a website akin to steamdb but for categorizing games on their AI use in the following
Categories:
- Handcrafted: The game has been confirmed untouched by GenAI
- Tainted: Gen AI has been used in any aspect of this game’s creation. Any usage of the technology puts you in this category.
- Slop: Use of Gen AI is outwardly visible by players
- Fraud: Developer lied about AI involvement in their products.
How the system would work:
It’d be a comunity driven, honor based effort. Users would summit reports and a group of trusted moderators would validate them.
Why I’m making this post:
I am looking for possible questions, problems, challenges and ideas you guys might come up with. I have a “Questions and answers” section down bellow with what I’d cover up myself.
Questions and answers
What counts as GenAI?
Large Language Models (LLMs) such as ChatGPT, Claude, Deepseek, etc. Image generation models line Nano Banana, Stable diffusion, etc.
Why take such a harsh stance on GenAI?
AI use has been irrevocably tied to the problems it creates. It’s being used as a tool to centralize power away from artists and creators towards a select few people with too much money on their hands.
Using AI is analogous to accepting all of the valid concerns against it and endorsing the problems it’s wedged themselves in.
What counts as AI usage?
Any personal use by you or anyone you are in charge of on the project. It’s the duty of the person in charge to ensure that your work stays human.
“I use local models I’d be allowed in handcrafted”
You’re still furthering the use of a technology designed to steal and plagiarize. While local models address some of the enviromental problems of GenAI it’s still an active endorsement of the theft done to train such models.
“I barely used GenAI for my game”
You made the choice to use GenAI on your game. You might have not been aware of the consequences of such choice but it’s still a choice you made.
“My game has been tagged tainted even though I made it all”
This is a community effort, I’m sorry if you’ve been falsely targeted/tagged. There’s always going to be bad actors. Contact the website at support@website.com and we’ll do our best to rectify the error.
“Why shouldn’t I just not disclose that I used GenAI”
No one’s forcing you to disclose the use of such technology but it’s something customers care about letting your customers make informed choices is how you build a reputation of quality and honesty.
“Why shouldn’t I just lie about my GenAI use?”
https://dictionary.cambridge.org/dictionary/english/fraud
Appart from that we have a special category reserved for this fraud. This is a website built around an honor system of reputation, do not tarnish it with abuse.
nerdspeak
And with that the mess of ideas and html I’ve been messing with is out. My progress is slow but I’m making steps. The tech stack is based on the human-centric language zig, using a jsx-like framework named ziex, which I’ve been thoroughly enjoying. I plan on making the website fully open source but there’s very little to showcase just yet. I’m just here to bounce ideas with the open web.
I have a suggestion: be explicit about music this (models such as Suno, Udio, etc.) and general audio (models as Whisper, etc.)
Re mistagged games, I think you may be opening yourself up to lawsuits, especially if this project succeeds in driving sales away from slop. Just look at Reddit as a cautionary tale of what community moderation can lead to: power tripping and astroturfing.
I’m all for something like this, but practically speaking I think it’s more effective to have store fronts force AI disclosure on store pages (and add filtering functionality). If a developer lies about it, they may be liable for false advertising. It’s also just more useful for consumers to have this information at the point of purchase (a browser extension could pull this off)
Itch.io recently added an AI disclosure category to all games, and I think it’s a good start.
Mind expanding on the lawsuit part? So, a game gets incorrectly tagged and the maker sues over contacting first?
Edit: I might have to think up ways of making clear that the liability lays on the reporter, not the platform.
This idea is basically a dedicated product review website, which is nothing new, but AFAIK there are precautions you need to take as the site operator to avoid liability for false/defamatory statements posted by users.
I’m not a lawyer, so maybe talk to one before you launch lol
Where does “I did a web search and found an answer on an AI generated source” land? I feel like the Internet is increasingly infected, and finding actual answers to stuff is hard.
Yeah, that’s just not enforceable with how much they shove it.
This is a good idea but I don’t think the current design adds much over existing categorization. There are already ways to find no-ai/handcrafted games. And complete slop is (at least for now) fairly obvious to spot. Imo the opportunity to refine the scale is all in the middle which you’ve collapsed into a single category.
I think most people who are willing to accept some ai use would benefit from a more nuanced approach. For instance, many people might be ok with some minor AI assistance in the programming process, but draw a hard line at ai assets. Your scale doesn’t serve those users.
I’m not arguing with your very valid points on the ethics and result of AI, just pointing out that to add utility past the current options you might need more granularity.
Also a major gray area is the use of libraries that may include ai code. For example, most games are now made with engines like unity, unreal, or godot. It’s almost certain that unity and unreal now include ai code. So can a game dev never use an engine again without falling into the tainted category? Do they have to write their own quaternary code for 3d translations, sprite blitter, texture mapper etc from scratch? What about using the windows API for file handling? That’s likely got ai code in it now. I guess to be considered handcrafted games need to be freebsd only now? Or ship with their own operating system?
I think we’d be better served by a scale like this:
0 - fully human made 0.5 - used a game engine / libraries / os SDK etc which likely includes AI generated code but is itself not considered total slop 1 - used some minor AI assisted coding techniques (e.g. code completions / merged an ai big fix after thorough review, etc) 2 - used extensive but human directed programming, assets all human made. Shows evidence of care put into design, graphics, story, etc 3 - vibe coded or close to it, but all assets still human made 4 - some ai assets, human directed work 5 - obviously totally vibe coded with generated assetsIt’s a conscious choice. I want to design the index to encourage people working towards the handcrafted stage. Any kind of AI usage immediately putting you on either slop or tainted drives the point home.
Ah I misunderstood the purpose. I thought it was to provide the community with a guide they could use to decide which games to support rather than a tool to condemn developers for their choices. Sorry for the misunderstanding.
For the nuance, I’m setting up a labeling system for the users to decide for themselves.
The second side of the comment about using libraries that coopt AI code. I think LLM use falls into a personal category. You can’t control that someone down your tech stack used it but you can certainly not use LLMs.
If you find out that your tech stack is tainted but your workflow is human asking for a rewrite is unfeasible. I’d argue that the next project should use human-centric tools but that is something for later down the stack.
I would actually argue that there is some validity to scrutinizing code supply chain. For example a dev could choose to a human-crafted library in c++ or use a vibe-coded “port” to rust or whatever because that’s the language they’re more familiar with. Those choices are worth investigating.
For that I’ve got a personal example. I have a professional project made in laravel a PHP framework. I made the choice to use it before the AI craze, I am maintaining the project but I didn’t choose for the tech stack to embrace AI like it did.
The fact that these projects take long to make does bind you to the choices of someone else makes that you can’t control.
But you did nail it in the head, it’s the focus on personal choice that I’m aiming at.
categories
Treat it more like stages of cancer.
You should definitely include AI gen voice and music.
Missed those. Will add. What are the common music gen services? Suno?
For voices I only know eleven labs
For ”music" gen, Suno, Kapwing, ElevenMusic, MusicGPT, Flexclip, Udio, AIVA, FlowMusic, Soundraw, Boomy and Soundful are the big ones right now, although I’m sure there are many more.
For voice, ElevenLabs, MurfAI and WellSaid are the ones I’m aware of.
No, I think all that you will achieve is to normalise the use of gen-AI. “It’s only a 1/5 so it’s just a little bit of gen-AI, that’s basically fine.”
It’s a bit like making a scale for vegans to tell them how much meat is in their meal.
I make an exception for something like Dragonsword, personally. They’re a small team from Korea who got fucked by the publisher, sued, and won the rights to their gacha game. They admit that some dialogue was run through a translator. Arguably no different than google translate, especially now.
Shit, my mom’s GPS doesn’t work anymore without Gemini having access to the car’s radio.
Granted, the amount of slopped mods…
Then your mother needs to not use that gps. I am assuming that is the one built into the car, they were an obvious issue from day 1
Alas, it’s aftermarket and is the only thing that works via android auto now, it seems. It’s idiotic.
You ask for an address and it tell you about the things at the address. If you instead type in an address, it’ll fuckin’ act lile a GPS, but that’s really hard to do without pulling over.
How so? The categories of AI disclosure seem fine to me.
As far as I know it’s only hypothetical at this point, but I feel like there’s room for an additional category of game that actually embeds an AI kernel in the game itself to do stuff like “procedural generation but better” or “chatbot-driven NPC interaction” at runtime in order to provide flexibility/breadth that’s intractable to do by hand, and that that’s materially different than having it be vibecoded or use pre-generated AI slop assets.
You just know that AI NPCs are going to be abused soo hard.
What do you mean by “abused?” I was imagining the user running a model locally, so doing whatever they wanted with it would be just as valid as doing whatever they wanted with Ollama etc…
If you mean “abused” in the sense of “murdering Nazeem for condescending to you about the Cloud District,” then sure, and that’s valid gameplay.
If you mean “abused” in the sense of “jailbreak the NPC to do your vibecoding for you,” then no, I wouldn’t expect that to be possible unless the developer was stupid enough to give their game free cloud AI access.
Well, you’ve seen the sister driving game tho?
This? No, I hadn’t until just now.
…So yeah, that’s exactly the sort of AI use I was thinking about, although I’m weirded out by this in the Steam game AI content disclosure:
The sister’s voice is generated in real time using a local model.
Connects to 3rd-Party Service for AI Content Generation: Gemini API, ChatGPT for Global / Qwen, Byteplus for China
If the sister’s voice is local, what’s the third-party service for?
They planned local model but they’re bleeding cash atm handing out free inference.
LOL, interesting.
Removed by mod
Nice idea. I have some questions.
Suppose an old game is released for a modern system. The game itself predates genAI and is untouched, but the bundled emulator contains generated code. Would this constitute a game containing slop?
If the answer to the above is “no”, what about the case where the publisher only distributes a binary blob from which the original game can’t be extracted?
Also, if a game is made with an LLM, but the generated stuff is subsequently cut out and replaced with handcrafted data, would you still consider it to have slop?
Finally, what would be the status of a came where all code, art, and music is handmade, but the devs employ LLMs to find bugs?
Thanks for the questions!
- The first question is the one that has the most nuance. I believe originally it’d simply be considered handcrafted but the introduction of a vibe coded emulator WOULD move it from handcrafted to tainted. I’m working on sufficient labels to be able to work out these nuances out.
- I said yes, skipping
- You still made the choice to use genAI, placeholder assets were a thing before and they still can be in the future. Errors happen with placeholder assets, specially those made to pass as real assets. Here we have precedent, Alters developers used GenAI and mistakenly (or not, we can’t verify) left some AI assets in the release version. What tainted remarks is the usage of LLMs at all, doesn’t matter the circunstance.
- LLM usage, tainted.
Ps.: I’ve added an extra phrase to the tainted description.
Second ps.: There might be a method of redemption for a project to move from tainted to handcrafted but it’d be easily verified. For example, Dave the diver has an LLM usage notice where they did the translations with AI. Should they re-do the translation work with actual translators (or let the community translate the work themselves) there might be a case towards bumping from tainted to hand-crafted.
I’ll be honest, I feel fine tainted being the “standard” and handcrafted being a the selective seal of approval. I’ll think things over. The chance for redemption creates subjective cases that give the people to say “why them but not me”.
OK, that’s reasonable. My mind tends to go for edge cases. Thanks for clarifying.
Your questions were great to bounce against. It led to the introduction of a new edge case: redeemed
It doesn’t erase previous GenAI usage but it shows the author’s effort to go a better way
Redeemed might not be the best word for it, “an active effort” seems very open to interpretation/bias?
I do however like the concept of a dev deciding (or being forced by backlash) to ‘lift their game’
Personally, none.
“Slop” is independent of LLMs. Judging how sloppy they are is best done by playing a game, and reviewing it normally.
In other words, I am actively disinterested in a board to filter games by factors that aren’t even in the game. Especially when that information is so sparse.
And I think judging them by development ethics is such a rabbit hole, it should include factors way outside the realm of AI usage, like… well, everything AAA devs have been up to.
