There are two kinds of bad procedural content: The first is the generic kind, made by connecting a bunch of random rooms and hallways together, all technically unique but functionally identical. The second is made by connecting pre-authored pieces together randomly – and, while certainly far more interesting than the first initially, this quickly loses its charm if there aren’t many candidate pieces to place, becoming mere remixes of largely static content. New roguelikes have largely, in lieu of actually solving these problems, found ways to bypass them. Indeed, one is hard-pressed to think of any game that has actually generated an interesting environment in its own right. Spelunky’s worlds are fascinating due to their endlessly volatile and unpredictably interacting elements: Dead Cells’ movement is enjoyable enough that even exploring generic environments holds pleasure: But, when it comes down to it, procedural spaces simply never have the mystery, the awe, or the comfort that an authored space can create.
Why is that? After all, we can haphazardly throw game mechanics together and have them create an interesting mechanical system to explore, why not combine rooms and hallways in the same way? The answer, I suspect, has much to do with the ways we regard game spaces in the first place, even in authored content. The purpose of a room in a game is to be a place where an event happens. This game-purpose supersedes any narrative idea of what function the room is purported to have for the building’s inhabitants, if any is even specified. Walls are there for the player to be blocked by, take cover behind, run on and over, and if they happen to make sense as an object a person would intentionally construct then that’s just gravy. This tight relationship between gameplay intent and physical space is nearly impossible to maintain through procedural scrambling of gameplay or space, though. Absent this connection of intent, game spaces become completely garbled and nonsensical. Thus, procedural spaces in games have largely split into two camps: Either conquering the space is the entire point, and so its specifics are always relevant (as in Spelunky or Noita) or the specific dimensions so unimportant that almost no effort is put into them (as in Hades or The Binding of Isaac).
These are both fine solutions to the problem at hand, but I believe it must be possible to have a game generate a space that’s intrinsically interesting and pleasurable to engage with – however, in order for that to happen, it must have intent. What does that mean? It means a space can’t just be an arbitrary labyrinth for monsters to be scattered in (though frankly even many pre-authored games fail by that metric). Each procedural piece must have purpose and be placed according to that purpose, with logic similar to that a level designer would use to place the same pieces. The algorithm cannot replace the role of level designer, only abstract it, distance it, so that the act of level design still happens through the instructions given to the algorithm.
That works for most rooms, but to make exploration not merely interesting but exciting there must be some chance for anomaly. Say the system is generating a bedroom: 95% of the time it will have one entrance, one bed, one closet, maybe a computer, maybe a TV, and one or two shelves and windows. Some bedrooms are different though: Maybe one has a fridge, or three beds, a balcony, a sword, a trap door, whatever. It’s the contrast between the mundane and the surprising that makes exploration interesting – most of the time we see exactly what we expect, and sometimes we are surprised, and that is where the tension, the humor, the mystery lies.
As long as games treat their environments solely as places where things will happen, rather than places where things have happened or things might happen, places with history and organizing logic, those places will feel mushy and inert. Each place must have a reason to be, or there will be no reason to make it – or to make the script by which it was made.