I’ve been thinking a lot about procedural games lately. It’s a bit of a holy grail of the gaming industry: games that make themselves up as they go along. There are many examples of this, from the earliest versions of Rogue, which makes up a dungeon as you progressed from floor to floor, to Derek Yu’s Spelunky, which does the same thing with its caves.

Then there are games like Left 4 Dead, a zombie-infested shooter that has an AI in control of its levels. The AI collects data as you play, deciding on what enemies to place where to give you maximum challenge — and, I suspect, maximum scares. It’s a system that works well, with the AI always dishing out pain in doses that you can just narrowly survive.

The appeal of procedural game generation is obvious — games stay fresh and replayable, always playing out slightly differently. In cases like Left 4 Dead, it also keeps the difficulty in check — if you’re having a little too much trouble, the game will back off a bit so you don’t get too frustrated, and if you’re kicking ass, the game will turn around and kick yours.

So, here comes the big question: if the game procedurally generated a story as well, would you get emotionally invested in it?

Imagine a standard procedural dungeon-crawler like Rouge. But, instead of just having some generic adventurer hacking his way through a dungeon, introduce some characters, some goals, and some twists.

It would start like this: The player is given an initial goal, and possibly some characters are introduced that help (or hinder) them from reaching that goal. Relationships between the characters are decided up by an AI. Then, as the game progresses, start intelligently throwing in events based on the relationships. Maybe a much-loved character meets a tragic end. A friend turns out to be an enemy. A love interest appears. New characters are introduced, old characters leave. A villain turns into an ally, an ally turns into a villain. You get the picture.

There’s a board game with a similar setup to this — Betrayal at House on the Hill — that has the player draw a plot at random at the start of the game and then modifies that plots with characters and plot twists as the game progression. This is a similar concept, but with an AI driving thing rather than random draws. By paying attention to how the player interacts with the characters and plot events, an AI would be able to pick out appropriate plot events and twists, foreshadowing future events and ultimately delivering things to a satisfying conclusion. The AI could even get away with a running gag.

An AI-generated plot also nullifies some gameplay problems. Is the player just walking in circles, unsure of what to do next? Why not throw a plot event at them? If the player needs to do a little too much hunting to reach the next plot point, bring the plot to them instead.

So, back to the question at hand: even if a story was being generated to generate the maximum emotional response with extremely compelling characters, could you get emotionally involved knowing that it was being generated by an AI and not penned by human hand?

Of course, if you didn’t know that things were being generated by an AI, it wouldn’t be a problem. But the moment you know that it’s a computer behind the story and not a person, a immense degree of authenticity is lost. The characters become non-people, computerized approximations. The whole experience ends up coming across as a smoke-and-mirrors attempt to make characters rather than something authentically emotional. Of course, the only way to really know this for sure is to test it — but I suspect the results would be lacklustre.

That said, I do have an idea as to how to make this work: multiplayer. If what the story is missing is a human component, why not put a human component in? Start with a procedurally generated story and AI-developed characters, but introduce other player-characters as well. If your story needs a villain, and someone else is just starting a story on his own, why not start their story off by injecting them into the other player’s story as the villain? Of course, this assumes that the players are roleplaying to a degree, which is actually a pretty rare occurrence. As a bonus, there are a couple technical issues that you could even use to your advantage — if there’s a dropped connection, you can actually use that to advance the story: suddenly, the player (who just disconnected, unbeknownst to the other player) is killed from behind by a dark figure, and a new villain is introduced.

Introducing additional player-characters would have a humanizing effect on the plot, since it’s no longer purely an AI driving things forward — it’s the collective of people playing it. It also makes the plot elements seem more human, even if it’s still an AI in control; if a character dies, you’re left wondering whether it was your actions that caused their death, or the actions of another character. Just knowing that other humans are having input into the plot development means that it’s not simply the player vs. a cold AI.

Of course, this is assuming that the players are participating in at least a minimal degree of role-playing, which is a… generous assumption.

Now my brain’s just filling up with all these ideas on how to procedurally generate things. Not a story, mind you — just things: Create a room. Apply random attributes to the room. One of those attributes selected is “dark”, which is an attribute that needs to be “solved”. Randomly assign some solutions to “dark” — “candle and [match, flame, lighter]”, “flashlight”, maybe even just “lightswitch”. Find locations for one or two of those items, or determine if one of those items already exists in the game.

Right. I’ll get back to you on this.

(If you’re at all interesting in this sort of stuff, I highly recommend stalking Emily Short, another person who’s pretty obsessed with interactive storytelling. She’s also played with non-linear stories with her games Galatea and Alabaster. Alabaster in particular uses a pretty interesting system for driving the story coherently forward, though the story is still based on branches rather than procedural generation.)