Our thoughts behind
Recently, several projects have emerged that offer AI-assisted design of dungeons and battlemaps. In a way, we're almost in the same field, so of course we've been looking at the projects in question. A lot of it is promising and looks damn good, but as IT professionals Jan and I have of course seen the developments around AI first hand in the last years and decades and are accordingly critical.
AI supports many services and products behind the scenes today - and very successfully! For example, things like eye autofocus on cameras, image enhancement, face recognition on images and much more are possible today!
But AI is not perfect, far from it. Voice assistants have gotten better and better over the years... but you know this yourself, right? Siri just turned 10 years old and still drives me nuts on a regular basis, just like a real ten-year-old I guess. Add to that some scandals that AI had discriminated by gender or skin color when selecting applicants, or becoming a blaring racist when used as a chatbot. This is simply because AI is not really intelligent in the human sense, but merely a trained algorithm.
But what does that even mean? You feed this program data and say "right" or "wrong" to each set of data until the AI gets better and better at judging whether something is right or wrong on its own through this "wealth of experience."
We have just seen how well this works in the broad field of art in an extremely exciting post on Reddit. Reddit User Eklundz used an AI powered App by wombo.art to paint fantasy RPG character classes. The result is interesting, but not really surprising.
And that brings us to why we think AI is not a good approach for creating maps for RPGs or wargaming. You do get custom landscapes and unique rooms populated with furniture and other items, but they will very often have problems upon closer inspection.
For example, the resulting landscape may not be visually natural, or the furniture in a room may be oddly arranged in a way that no one would actually do. The stumbling blocks that an AI would not recognize due to a lack of actual intelligence are simply very numerous.
For our outdoor maps, we therefore rely almost entirely on manual work. Of course, we also use algorithms, for example to distribute vegetation. But these work rule-based and therefore deliver predictable, adjustable results.
It will be similar for our future dungeon maps. We'll probably go for a mix - hand-built rooms and rule-based generation - so that the bedroom doesn't become a pass-through room to the kitchen! This way we can ensure that a dungeon is structured in a meaningful way and that the rooms are set up in an equally reasonable way.
This keeps Infinite Realms a software for the lazy dungeon master - natural looking landscapes and convincingly placed props!