AI in retro games

This includes all mentions of maintainability, lack of optimization, difficulty debugging, etc.

It's evident that AI can produce very low-quality software, especially if you don't know what you are doing. The mentioned GoldenEye XBLA is a good example of that.

But pretending recent AI models haven't reached a point where they can produce software comparable to or better than what most software engineers can produce (while iterating much, much faster), is simply disingenuous. And that makes the argument very weak, because even if it were true, it wouldn't be for long.

True, people without the necessary skills are putting many low-quality projects out there. But this was also true before AI for every new tool released: low-quality graphic romhacks (tile editors), low-quality colorizations (GB Colorizer), etc.

Of course, those very same tools also allowed great projects to be created. So I don't think it's a good argument.

It's true, AI undoubtedly has an environmental cost. I won't pretend I know the details of how many resources AI can consume, but I do know that this mostly applies to AI companies when training frontier models.

In other words, enormous resources are used to create the model, not to use it. Reading the article (and listening to most people talking about AI, really) made me realize how misunderstood this is.

If you still consider using a model created this way unethical, which I would understand, I don't think this argument will stand for long anyway: it's only a matter of time before running capable local models becomes cheap.

This is true and absolutely unacceptable. I'll obviously won't excuse the shitty practices of especially AI companies, but here's something that most people don't really realize: they stole content to accelerate the development of AI, not because it's inherently necessary for the technology.

This is an important distinction worth understanding, because while a lot of people simply think AI depends on a giant pile of stolen content to work, that's simply not true. If we lived in a more ethical world, it would simply have taken longer for AI to get to the same place.

AI industry's behavior being unethical doesn't make using AI unethical.

This is a very interesting argument to me because, as a romhacker, I love working around the limitations of a console and ultimately overcoming "impossible" challenges. And now, as AI improves, doing the impossible is trivial.

So I understand devs/romhackers feeling threatened by or resentful of AI.

But is there really no place for AI in this niche? I don't think it's black and white. The interesting questions are what role it played, whether the person using it understands the result, whether the result is actually useful, and what we value in the first place.

If you want to use AI for graphics and replace the ones in a game with slop, the end result will probably be crap, and I'd rather you didn't do anything at all. But I'd argue there are some instances in which I feel it's completely okay.

For instance, as mentioned, I used AI to get a "matching reimplementation" of Super Mario Land and used it as a base for my port. My code and graphics still need to get around the console's limitations (which I love!), but I'm very glad I was able to have the starting point I had. And honestly, I'm having a blast.