Idk, I get we're in the frustrated zone rn but if you're going after roguelikes and "traditional procedurally generated content" as AI-adjacent where do you stop? I use an LOT of procgen animation in the game I'm working on right now because there are things that would be straight-up impossible to do without it - that doesn't make it NOT something made with intentionality. If I spend hours building a system that makes a character dynamically come to a stop, is it "not handcrafted enough" because a computer is enacting it, even though I specifically created the parameters to get what I wanted?
This kind of argument massively misses the point to me. AI is a plagiarism engine, and it's also specifically not a tool. It's a black box where idea guys put things in and get random garbage out. Imo to compare it to - to place it alongside traditional procgen is to do a massive disservice to the amount of creative effort that goes into making quality procgen work.
The basic underpinning structures behind machine learning (ML) can be applied to small data sets as well as large ones for various classes of problems. In fact, most of the ideas in machine learning work well if you feed them synthetic data - which is not difficult to generate in, say, a video game. There is lots of cool work you can do in this field which doesn't rely on a giant webcrawl of the entire Internet so all your output has a little sprinkling of nazism in it, or ripping off artist's work wholesale - and as a researcher, those ethical concerns (as well as my own interests) dictate what I choose to work on. DLSS is a good example of an ML-based algorithm that was produced ethically and without recourse to intellectual property theft, and is an improvement over what it replaced (the TAA class of anti-aliasing algorithms.) You may not like TAA/DLSS - I actually don't - but it does do what it says on the tin.
There is stuff coming down the pipe in the research pipeline that also does more cool graphics things; but at the heart of all of these methods are neural networks, variational autoencoding, and the ADAM stochastic gradient descent optimizer (which is useful for solving even non-ML optimizations!). It is cool stuff! The fact that it's lumped in with the image-stealing artist-destroying black box work being done by less ethical researchers is itself frustrating.
The fact that the AI Bros choose to fixate on algorithms requiring infringement of intellectual property is a choice, and the fact that their business application is "putting people out of work" also represents both a choice and a lack of imagination. At the end of the day we are talking about a class of function approximators and a class of optimization algorithms - that's all "machine learning" is. We need to start distinguishing between "ML" as in machine learning, and "AI" as in snake oil.
i think there's a couple threads here
edit: aura rightly noted that i focused on the roguelite/like aspect of procgen and didn't much talk about procgen as a tool for other "behind the scenes" tech like animation. i think that's because i think the OP that kicked this off was more focused on the player-facing uses of procgen (like level generation) and also because i think i agree with her without caveats on other things, like animation
procgen games and motivation
first, the "i'm uninterested in procgen games, give me intentionality" is absolutely something you can feel about games and it isn't like, a wrong thing to feel about games specifically. It's just different motivations for play and probably a cool thing i read recently about different types of perceived agency which overlaps with those motivations somewhat.
and it makes perfect sense that if you specifically crave intentional experiences - that is to say, part of your gameplay motivations are that you want to find out what the creator had to say and pick apart their ideas and how they chose to express them in a game, and you enjoy taking part in a game through the lens of what the article calls protagonism and velocity, you are going to be drawn to intentional, crafted, likely narrative heavy design
procgen (and specifically roguelike/roguelite) design by contrast feeds into a gameplay loop that lends itself more to progression-as-a-player, where you largely progress and grow more powerful at the game through learning-by-doing, accumulating knowledge and experience as a player, outside the game, that enables you to progress further in the game
and procgen isn't the only way to do this, of course, i think we're all probably thinking soulslikes at this point, and yeah. they probably differ in how much they lean on improvisation vs memorization but enumerating different shapes of games isn't really the point here so much as acknowledging that different design philosophies shape different types of experiences that cater to different types of gameplay motivations
(i have some small amount of friction with people who can't stand games that aren't open world-likes sometimes because i am the open world disliker who can't stand them and really, the answer is just that we're looking for different things in games and are motivated by different types of play and that's fine)
but to come back to the original thought, it's perfectly understandable if the modern day onslaught of exploitative generative AI being shoehorned into everything moves the needle away from procgen games for you. it makes perfect sense that oversaturation with that kind of thing and revulsion every time you see it displace human creativity and intentionality would drive you to seek more human experiences
machine learning as a technology
so yeah, i don't have a lot to add to @egotists-club here other than to say we also use machine learning in our research. it's an internally developed image classifier trained on approximately two decades (hundreds of terabytes) of meteor imagery and video we collected ourselves using our own instruments
it's not based on openai or any llm model or based on stolen art or internet sourced anything
i can't speak to the bleeding edge work coming from the computer scientists working in this field, that's not really my field. but i can tell you that the field of AI as a science has existed for decades before the LLM grifters showed up on the scene and it will exist for decades after they're dead and gone
there are actually some very simple AI techniques you can implement yourself in an afternoon! Bio-inspired algorithms like Genetic Algorithms and Particle Swarm Optimization were big in the 2000s and still have applications these days
they're probably not what you think of when people say "AI" because the likes of Sam Altman have poisoned the term by talking up the dangers and possibilities of a true sapient AI1 and positioning themselves as the only ethical answer to it2, but the field itself dates back to probably the 1940s or so. we literally call a hypothetical discriminator to discern between a human agent and an AI agent a "Turing test." Alan Turing literally did some pioneering work in the early days of this field
mundane things like simulating swarm intelligence may be kind of boring and mundane by comparison to the types of machine learning being done today by non-LLM fields (like the aforementioned graphics processing technologies) but i think it's important not to let the Sam Altmans of the world take up all the air in the room and define the boundaries of a very old and diverse field to suit their grift
-
not gonna happen in our lifetime
-
hahahahaha no
