Skip to main content

On Sale: GamesAssetsToolsTabletopComics
Indie game storeFree gamesFun gamesHorror games
Game developmentAssetsComics
SalesBundles
Jobs
TagsGame Engines

The line between AI assisted and AI generated. For such a line those two would have to be on the same scale and increasing something would make it switch over from one state to the other.

For walking and running that is speed and the line is when both feet leave the ground.

What is the something that changes for AI involvement? I guess it is decision making and sample size. But it depends on the context. If a photographer takes 1000 pictures and selects 1, did he create the photo? Or was it the camera and the objects that were photographed? Does an orchestra conductor create music?

Those two are generally considered art. But if you prompt a gen AI system to make 1000 pictures after your direction and select 1, it currently is not considered art, and it is debated if it counts as creation. And the selected image will be AI generated, even if you change something, because an image is a big sample size.

For code creation you have templates and text book examples, discussion threads and existing code to copy from. Was it you, that made the code, if you copied it and modified it? If your sample size or your own deciscion making for the modification are  sufficient to not violate copyright, yeah, you made it.

So, for code generation, I would put that line at the point where you let the AI do basically the game mechanics, instead of parts, like algorithms and functions. Also, you do not ask AI to give you 1000 functions and chose one. If it works, it works, if it does not, you make it work. 

And for image generation, that line is shifted. If you ask an AI to draw the hands on your stick figure or you have the AI make the stick figure and correct the hands, both those big samples are AI generated. 

As for the rules on Itch, if you do not have AI artworks and merely AI "assisted" code, but not large portions of AI code, I would just leave the disclosure blank for the time being, if possible. That disclosure feature has a lot of issues, both in the declaration and how the feature is useable for players. I have yet to see someone lamenting about how bad AI code games are, which makes me think, that players do not care about the code. They care about story and art works. But Itch lumps all "AI" together with their "no-ai" filter. 

Thank you. Yes, I'm very involved in the coding process. Even when AI leads me down a wild rabbit trail when trying to debug one of its own problems it's still really educational. Heck, I remember trying to debug collision between CharacterBody2D and Area2D nodes in GODOT. AI had me using _on_area_entered(area) when I really needed _on_body_entered(body).
Before it was all over I had print commands set up to spam Output with the ID, collision layer, and mask layer of every bullet being generated. Absolutely everything was lined up but hits weren't registering until my gut told me that particular function might be the culprit. Looked it up, sure enough, just a GODOT quirk. Learned so much that day.

(+1)

The thing is, in coding you have library calls. This is not present in art. You do not paint a thing and instead of a hand put a reference to the hand-maker with the parameter of 5 fingers. Reusing existing things is kinda the whole point of programming.

But if you rehash art, people get angry. For various reasons. Does not matter if you do it by hand or with extra steps of using a large language model. There will also be a line between copying/rehashing and actual learning how to do a thing. If AI crosses that line, it will be hard to argue that it plagiarizes by traning data. It will depend only on the results. A human can also plagiarize. But a human looking at art works will not be accused of rehashing those in memory to be able to reproduce those works.

As long as those systems put out the wrong number of fingers, they are quite on the other side of that proposed line. Or in case of code, your example with the wrong function. The training data probably had examples of those in a different context, but the AI did not grasp what you actually needed, it only gave you a thing that is probably similar to the answer by looks alone.

I hadn't thought of it that way. If recycling and refactoring is already such a crucial part of the coding process, then having AI generate code snippets and generic function templates instead of grabbing them from design bibles, sample projects, or code repositories isn't really that much different in the long run, is it?
Thanks! I'll get back to working on my project. I hope to have a playable tech demo soon.

For code snippets, yes. If you ask the prompt to give you a match-3 game, that would be a bit different. But essentially, code you write is built on other code. You use Godot. The things you do with it, would not be as easy, without Godot existing. There are people doing certain things from scratch (in relative terms. They still use operating system calls and whatnot).

In coding there is also this concept of hierarchy of languages. How abstract is the code from the machine code. The continuation of that concept is using natural language. When the pseudo code is the code. The prompt you use would be the code. But since they use training data of the usual abstract languages, that approach is flawed at the moment. As you saw with your example.

In the arts, there are not that many abstraction layers and "library" calls. There are some. Especially when doing it digitally. And the artists also have issues with techniques like "tracing". Or "referencing". That is just a nice word to describe that one artist copies how a thing looks by looking or tracing the outlines from other art. What you would do in coding by looking at examples.

Anyway, that is my opinion why AI is not as frowned upon in coding as it is in "art".