Emergent Puzzle Solutions

A recent RAIF thread brought up the Magnetic Scrolls games, and the fact that they used a simulationist system that could produce puzzle solutions that the game authors hadn’t thought of:

“Talk of current IF development drifted on to whether it’s possible to create a game in which the player is not really constrained by the author’s intentions. Michael noted that Magnetic Scrolls games were kind of like this-for example, if an object had the “sharp shards” bit set, dropping or throwing the object would cause it to shatter into many sharp shards. In total, 128 bits were used to describe a more or less working universe that the player could interact with in ways that hadn’t been anticipated. As an example, Michael described an unintentional situation in which one could put a rat in some liquid nitrogen, snap off its tail and, for a few turns, use the tail to puncture feed sacks and obtain food.”

This raised a fair amount of interest (most of the “ZOMG that would be GREAT!!” kind). This yearning to do something the author didn’t think of is something I hear a fair amount of: Mark Bernstein has complained that, because IF games anticipate solutions, the IF player is always robbed of the pleasure of having invented a novel solution because he always knows the author was there first. Emergent-solution design might address that complaint. It might also address the frustration players often feel when a logical-seeming approach is either forbidden or not recognized by the game at all.

So I found myself thinking, again, about why more IF games don’t work this way.

It’s not that world model features like these are hard to code in modern IF languages. They were pretty doable in I6 and TADS 2, and are all the more doable now. From a coding perspective, it’s all about using attributes or classes to determine how objects interact, rather than making rules that apply only to specific items. Inform 7 and TADS 3 both, in different ways, support and even encourage this kind of code generalization.

It’s also not the case that no one has thought about or agitated for a game with emergent puzzle solutions. The idea has been floating around RAIF since at least 1999 — I’m sure there were renditions of it earlier, but I say 1999, because this was one of the topics that interested me enough to come out of lurker status, back in the day. I’d been working on a comprehensive simulation library, with modules for light levels, fire that burned and consumed, liquid that mixed and soaked and destroyed, breakable objects, cameras, smoke and gases, etc. My hope was to create a coded environment that felt deeper to the player than any other in IF to date, and that allowed for puzzle solutions that the author hadn’t thought up in advance. At one point I made up a small demo, which I posted, early in 2000; it mostly involved various ways of destroying things, by cutting, burning, or smashing to bits. I wrote then:

So with a complex, open-ended world model I envision the following IF-writing methodology:

1. Come up with the setting and the major plot junctures.
2. Figure out what the player has to do to get from one juncture to the next.
3. Fill the setting with props appropriate to the location/situation.
4. Consider: can the player use these props to solve any problems that stand between her and the next progress point? If not, figure out what’s needed and include it somewhere.

One could use this methodology now, of course, but the more complex the world-model, the more likely it will be that the game will produce its own fitting-but-unplanned solutions… my goal is to allow the player to do fairly ingenious things without breaking the feel and the flow of the story.

A few of these features made their way into my finished games — the liquid system became one of the major elements of Savoir-Faire, and I used some of the concepts in Metamorphoses. But I didn’t wind up designing those games in the way I imagined originally, and they also are not generally regarded as having achieved the emergent-solution goal, either. Nor have many other pieces of IF out there.

The Complexity Problem: what “counts” as an emergent solution. Both of those games do have emergent solutions in the sense that there are uses of objects possible that I wouldn’t have anticipated specifically. If I conceived of the puzzle solution as “use something hard to…” or “put something soft under…” or whatever, then it’s up to the player to supply an item that meets the criteria, without my having thought of all the particulars.

But that doesn’t have the degree of open-endedness that emergent-solution proponents tend to have in mind. They are, I think, looking for a multiple-step solution that relies on processes within the game, not just attributes of existing objects, even if those attributes are shared by a large number of in-game items and the author has no way to anticipate all the possibilities. What I mean by a process is some action or series of actions that changes game state. In the Magnetic Scrolls example above, the player notices two processes: dipping something in liquid nitrogen makes it fragile; breaking something produces sharp shards. Using those processes, he obtains the attributes needed to accomplish his goals.

What’s more, such processes need to apply to a large number of game items, not just one or two. In Savoir-Faire, we can apply color to a handful of objects, but for most of them this is impossible, so the color-application process does not contribute to an emergent-solution world model. Most puzzles in SF are about noticing attributes or applying at most a single action to change states.

Metamorphoses includes several in-game processes (resizing objects, breaking objects, changing the material substances of objects, piercing objects with a needle), and it’s possible to string these together — change an item to glass and then break it, say, or change an item to something that isn’t too hard, then pierce it, then resize it so that the hole is large, then change the item into a heavier substance… But I did anticipate most of these sequences, in part because there weren’t that many problems available to be solved. Emergent solutions tend to happen more often when there are a large number of puzzles, so that the world model developed to account for problem A can also be leveraged, unexpectedly, against problem B. So such games also probably need to be of a reasonable size.

So I hypothesize that a game allowing emergent solutions needs all of the following:

• attributes common to most game objects that affect interaction
• processes, effective on many game items, that allow the player to change attributes (or produce an item with new attributes out of an old item, as in the case of breaking the tail off the rat)
• a selection of processes that can be used in combination (freeze rat then smash it); one way to think about this at the design phase might be to draw a chart of attributes and processes, showing which processes convert which attributes into which others; the more long chains are possible, the more complex the plans the player can execute
• sufficient number of puzzles that the solution space becomes too large for the author to anticipate at the design phase

Once we have all those features, though, we run into some other serious design problems.

Maintaining solvability. Modern IF tends to avoid designs where the player can get stuck in an unwinnable game. The predominant design these days is a game which the player can either lose unambiguously (e.g., by dying) or continue to play in the expectation of making progress. Gone (mostly) are the games where you wander around in frustration for three hours before you realize that you did something seven hundred moves back that is now hindering any forward progress.

There are a few exceptions to this design rule, especially in games that are designed to be very short, so that replaying from the beginning is not too onerous a burden. But for the most part it’s settled in as an expectation of the community.

It’s hard to reconcile this rule with a sandboxed, emergent-solution game world, though. If the player is allowed to use (and especially to use up) objects in the game world, it’s harder to make sure that there are always enough supplies remaining to let him get through to the end. It would be possible to add a game-managing routine that introduced new items into the game world when old ones were destroyed — a sort of spawning process — but it seems as though this would get quite predictable and in some cases would be narratively hard to account for. (Accidentally smash the One Amulet of the Great God Mu? No problem, we can get another from eBay.)

There are other game genres that have fewer kinds of distinct objects, and more consistent world physics, in which emergent solutions are relatively common. But IF thrives on the unique and particular. This is partly to do with the requirements of narrative. We don’t tell stories about 84 cubical crates of first aid supplies and 92 barrels mysteriously packed with ammunition.

Maintaining narrative flow. Besides the design problem just mentioned, emergent puzzle solutions can raise problems with narrative flow. IF very often uses puzzles as a gating mechanism to control the player’s access to parts of the story. That works fine, as long as the author always knows for certain when the player is going to be empowered to pass a certain puzzle. The more susceptible the world is to unintended solutions and bypasses, the harder it is to exert narrative control. If, as I’ve argued before, the key to better interactive storytelling is to make the player’s actions more relevant to what is happening, it’s not clear that emergent-solution design helps with that.

To come at this argument from another angle: there are lots of games — everything from The Sims to Europa Universalis to Grand Theft Auto — which provide challenges with multiple and versatile solutions beyond what the game authors specifically envisioned. The gameplay of these is sometimes very good. What is not so clear to me is that this kind of design produces good narrative. Europa Universalis, at least, has a fair claim on being good interactive history, if that concept makes sense. But these are not stories.

I also think that this kind of sandbox play is really, really hard — nearly impossible — to render well in text, and that therefore IF is entirely the wrong medium for it. Whenever the underlying model is dominantly quantitative rather than qualitative, it’s easier to show in graphics than to describe in words.

Testability and Special-case Responsiveness. Any time you want to allow actions the author hasn’t anticipated, that necessarily means that the author also hasn’t tested them explicitly. Which means that it’s hard to eradicate weird edge-case behavior. Suppose I freeze the rat, break its tail off, warm the tail up again, mince it, cook it in the microwave, and try to eat it. What happens then? Suppose the minced tail has inherited the MADE-OF-MEAT attribute from the rat it once belonged to, which makes it edible when cooked. Does the game respond with, “Mmm, delicious!” or the like? What if the author has coded a sarcastic retort for trying to eat the rat directly, but does permit you to eat derivative rat portions?

At some point these results will stop looking like awesomely flexible applications of gameworld physics and start instead to look like bugs. What’s more, the proportion of customized response in the game goes down as more and more of the game’s text is produced in reaction to standardized actions. The results feel less literary and more game-y.

Focus on the Physical… Emergent-solutions systems as suggested by IF players almost always involve a detailed physical world model in which legs can be broken from chairs and used to bludgeon prison guards; lenses can set fires; sewer ducts can be crawled through; anything made of wood can be sawn in half… All this is fine, but it’s by no means a suitable kind of interaction for all stories. If the player character is supposed to be a clever, mechanically-inclined sort who solves everything by cobbling together a gadget, fine. But for much IF, this is not going to be a kind of interaction that advances the story-telling purposes of the work; and, if anything, the presence of this intricately made but intensely substance-oriented view of the world will be a distraction. Can we imagine a version of Common Ground in which it’s possible to spend real attention on the details of repairing the toaster?

…or the Psychologically Simplistic… We can, of course, imagine emergent-solution puzzles that aren’t so object-obsessed. Arguably, that’s part of what Chris Crawford is doing: making algorithmically-behaved characters whom the player can manipulate in an assortment of ways. It’s conceivable to have a detailed system that’s about people rather than a detailed system about things. But the problem there is two-fold: the improbability of producing a good story, and the difficulty of describing it. As a solution, Crawford has all his characters talk in a simplified NPC language, stripped of all the color and nuance that lends interest to real characters.

…or Knowledge-based… My game Mystery House Possessed tried a different approach to emergent solutions: there is a randomly generated murder and series of events following the murder; each murder leaves a variety of clues behind, and the player can use any of these to narrow down who the culprit is. As the author, I have no way of knowing exactly how the player will come to a solution. Because the player is mostly finding out things, rather than changing them, there isn’t as much opportunity for the emergent-solution approach to derail the game.

On the other hand, such a large proportion of the game depends on random factors that it’s impossible to write a walkthrough, and much of the time the player doesn’t so much get a chance to execute a strategy as to happen to stumble across new information. It’s possible over time to get better at the game and solve it faster each time, but there aren’t that many opportunities for the player to spontaneously manipulate the game for his own advantage and in a way unanticipated by me.

Though the approaches to detection are less varied (and less haphazard), An Act of Murder works in similar ways: because the game parameters are always reshuffled at the start of play, the solution to the game can best be expressed in terms of a strategy — what do we ask, of whom? What should we look for? — rather than a series of actions.

…or Randomized. Similarly, the When in Rome episodes deal with partly randomized, goal-seeking NPCs who can be defeated through various approaches; again, it’s impossible to write a walkthrough for these, since the nature of the creature being fought varies from game to game, and in any given playthrough the game-state is likely to develop in its own direction. But again, while I might not be able to predict what happens in any given game, the range of options available to the player always corresponds to certain general strategies which I was aware of at the design stage; so I think this cannot be called a genuinely emergent-solution kind of work. (In beta-testing of WiR 2, I found that it was often an emergent-failure work — in a few cases a clever and malicious NPC executed its goal-seeking behavior in a way that left the player without any way to win after a certain number of moves. But this is an example of how such systems make it harder, not easier, to design a game that’s consistently fair to the player.)

Approaching Emergent Solutions in IF

We could embrace emergent-solution design as a primary goal and just accept that this may not be a direction that leads to good interactive narrative or to designs that always remain winnable. Design, instead, on the expectation that the player will need to replay; that playing a game will be more about learning successful general strategies over many play-throughs than about working out the solution to puzzles once; that in fact we’re moving away from the territory of “puzzle” entirely and towards something that might be better called a “challenge”. Replay value goes up; literary value goes down. I think that’s a completely viable choice. There will be people in the IF community who don’t like the results, but that’s their problem; I think the smart thing to do is simply to be quite clear at the outset what kind of game you’re peddling, and let people opt to play it or not, according to their preferences. Probably as many people will be attracted as are repelled.

The much harder trick is to find a way to incorporate emergent-solution design into our collection of tools for interactive story-telling: to let the player use an extensively modeled system to explore the important issues of the story/game and as a mechanism to make choices that matter.

For that, I think we’re going to need to expand our thinking about what to model beyond the realm of fragile items vs. hard items vs. flammable items, and also beyond character conversation (which I continue to think can’t usefully be abstracted in this way lest it lose all personality).

I don’t know where that leaves us.

Yet, anyway.