This post by Ethan Skemp, and this comic by Randall Munroe, both comment on how humans process randomness, with a slant toward two different types of games. Through our unstoppable powers of pattern recognition, we are excellent at constructing narratives out of events that are either seemingly or literally random. As Ethan discusses, gamers are accustomed to exerting effort toward doing so, stringing together a story even when they watched the dice tumble and know that the outcomes weren't foreordained. There's another element to randomness in games, though, that I'd like to talk about in this post: randomness as a way to avert the perception that the narrative has been constructed. This is really about GM trust.
What I mean by that is that there are plenty of situations in which it is best for the flow and function of a game if the players perceive the GM as an entirely impartial arbiter. At such times, random determination of the outcome is a tried-and-true solution. Now, at its most basic level, this is just about the least innovative possible sentence as regards gaming. Everyone is familiar with non-roleplaying games that use dice to determine outcomes between players; there the dice completely take the place of an impartial arbiter.
I'm not really interested in the round-by-round combat resolution area of randomness. It is right and good that it is there, because I don't go in for entirely narrated combat. The perception of fairness is at stake more than ever when the stakes are high. Going into running Over the Edge, I mentioned to players a number of times that I, as GM, wanted combat to involve exciting description on both sides, with the potential for narrative consequences, and I hoped that they would trust me and be happy with this. Now, getting these people to trust me as a person was trivially easy; at the time, everyone there except Kainenchen had known me for four years or more.
When we talk about GM trust, though, there are a ton of issues on the table. First off, what am I even asking them to trust me to do?
Be fair. Well, what does that mean? To parse the rules as they came out of the book, and nothing more? Actually, no, this isn't quite right; I believe they expected me to put my own judgment above the words on the page when I believed it appropriate to supporting their fun.
Make the game fun. Technically, I haven't promised to make the game fun. If that were the case, I'd avoid any kind of negative consequences and just go in for the instant gratification.
Make fun possible. As I see it, this is the answer: it's up to them to decide how to feel about things that happen in the game. Individual incidents that are not fun when viewed in isolation are going to happen; I'm even going to deliberately make them happen, because challenge and threat are important to creating a fertile ground for achievement.
In the first combat, one of the players protested with some irritation when his narration of his attack included details that, in my view, made him particularly vulnerable to the opponent's next attack. This wasn't a rules-dictated (by which I mean "found in a book published by Atlas") ability - it was something that seemed appropriate to me, at a time when I needed the enemy to be a little bit more of a threat. The situation was resolved, but the player's perception stuck with me. I'm belaboring the point here, so I want to be clear on one thing: the player wasn't being a jerk, and I do get why he objected.
He wanted to understand, for the sake of future tactical choices if nothing else, why the enemy could do this. Since it wasn't printed in a Monster Manual-style stat block (because OtE is just that rules-light), I couldn't point to a book and say "because the book said so," or point at the dice and blame them. If it had been, he wouldn't have known ahead of time, but maybe he would have said to himself, "Oh, well, I guess the GM's just following the rules." Instead, the answer boiled down to because I said so, and that's a bit harder to swallow.
My reasoning at the time was something more like "what's sauce for the goose is sauce for the gander." It didn't bother any of the players that I was giving them delicious bonuses for interesting narration, regardless of whether this was printed in a book; not to mix my barnyard metaphors, but... something about gift horses. Just about every roleplaying game yet penned has mentioned bonuses for doing something cool and situationally appropriate. What's missing (and rightly so) is any opening for a penalty to PCs or bonus to enemies. The PCs can't really be expected to come up with a new and perfect combat description every time, and the PCs really don't care to hear about how awesome the NPCs are. A lot of recent indie games include mechanics to allow the GM to give NPCs just this kind of advantage. This usually awards the PCs a Plot Point or something similar.
Huh. Another way of looking at it that just occurred to me right now: everyone is free to make declarations against their own "best interests" - that is, the GM can give the PCs things that make them more likely to win, but bending the rules to help out his own side is not cool. It would be fine for the PCs to note that the GM's description was deserving and give the NPCs bonuses, or take penalties upon themselves, but even a completely non-adversarial player/GM relationship doesn't extend to the players needing to make the GM feel awesome or rewarded during play.
Anyway, where was I: I've seen other GMs use dice to determine which player comes under fire from NPCs when the NPCs have no other clear basis for their decision - or just the same for a beneficial thing. I've never been the kind of GM who is comfortable leaving the major plot decisions to a random roll, but that's a big trend among the OSR community, particularly as described here. I prefer to take a stronger hand with managing the pacing of a game than that system would allow, but even for me random determination has a right time and place.
Player-driven goals are one major area for this. Let's take a hypothetical crafting system in which Karl the Crafter needs one particularly rare thing (an ingot of mithril) and two uncommon things (an amethyst and a large piece of ivory) to make the thing he wants. Within the scope of the game's actual plot, the GM neither objects to the player proceeding nor needs the player to succeed sooner. In the course of their adventures, the players kill a dragon; they didn't have any specific reason to believe that that the components would be there. As every D&D player all the way back to the Beowulf poet knows, though, a decent hoard contains a lot more than just coin. Since the hoard definitely either does or does not contain one or more of the things Karl is looking for, how should the GM make this determination? Is the GM making it too easy on the player if the hoard contains all three items? Is the GM making it too hard on the player (or, in the more common parlance, being a dick) if the hoard contains none of them?
In that case, it's a great time to go to random determination. Even if Karl doesn't get what he wants, he knows that it wasn't a sign of GM disapproval, and in other similar situations things could turn out differently. Maybe Karl can do some asking around to figure out where the components he wants are most frequently found (a bit of legwork to persuade the GM to shift the odds in his favor). In my own experience with player expectations, it doesn't take complete transparency here for players to accept that they're improving the odds, though if they continue to fail due to bad dice even after putting forth a good-faith effort and facing repeated challenges, it might be time for the GM to adjust the odds precipitously.
Sure, random rolls for what's in a dragon's hoard are nothing new. Ultimately what I'm trying to get across is that I like for randomness to interrupt and interfere with player plans in a way that doesn't represent enemy action. I discussed this idea at some length when working on a new system for skill challenges. I've also seen subsystems of games that were all about randomly introducing shifts in a battle, whether from a deck of cards or a d100 roll. At the same time, I wouldn't really want to add another step to remember and resolve to a fight scene, and if I'm deciding when I want to make a roll or draw a card, that's only marginally better than operating through fiat in the first place.
So if I like random determination for all of this stuff, what about wandering monsters? Very early in our conversations about running tabletop games, Kainenchen and I found that we disagreed on the use of wandering monsters: she swore by them, while I saw little use in them. Some years later, we talked about it more and she explained that she wasn't rolling up the encounters during play, but was rolling up encounters during session prep. This sounds a lot more reasonable to me - for all that I hear of other GMs rolling encounters at the time, my particular creative impulse isn't inclined toward rapidly incorporating new ideas into a narrative I already have in mind. In any case, it's a kind of randomized story generation that has been in D&D's lineage all along, but one I never embraced. I mostly use planned (even loosely-planned) encounters, which means that for good or for ill, what happens in the game happens by my direction.
I think this coincides with another of the dividing lines between the "modern" view and the OSR view. (To be clear, Kainenchen's point of view is not derived from the OSR.) I plan encounters and deliberately scale them to be within a certain range of what the players can handle. This doesn't mean there might not be encounters that are too tough for the PCs to handle, though they'd be very rare; part of the way player trust has developed in the "modern" view is that the GM will give some pretty strong signals if he believes that the encounter is out of the PCs' reach, and will also give the PCs ample opportunity to run away. Otherwise, the players probably believe (because all of this is directed) that the fight is meant to be a serious challenge, but within the upper limits of what they can do.
The old-school view is that the tables are demographics for the world, and the PCs should judge everything on what the characters see. If it looks too tough to fight, it probably is. It's not there because the GM thinks you can handle it; it's there because that's how the dice fell. Alternately, because "that's what the module said was there." This has the same effect of absolving the GM of responsibility for the outcomes - here again the GM is no more than a computer parsing the rules and describing the view. I have a certain amount of sympathy for the view that the players shouldn't try to read the GM's mind - doing so complicates the game immensely and distracts the players from exploring the world as presented.
I subscribe a bit too heavily to the GM-as-auteur philosophy to fully embrace random generation. On one hand the GM is absolved of being the Bad Guy when there's an encounter that is straight-up bullshit (how many medusae?). On the other, there's no plan - there are no underlying, mysterious causes driving the encounters. The dice are the mystery, and the characters are not going to resolve that. The counter to this, I think, is that the GM can string together what happens to form something cogent - or if something doesn't fit with his idea, he can reroll. For the former, it's a little too much like a stage magician explaining the trick; for the latter, the decision to reroll gets the GM involved. Let me beat this horse no further.
In summary, random determination is at its most useful whenever the GM wants any number of different bad things to happen, but fears that the players will perceive this as punishment, vindictiveness, or just "sending a message." As long as dice were rolled, players will generally accept that the outcome was one among many different outcomes. They may wonder at the odds and the other available outcomes, but I think most players would accept that something was written down ahead of time - if nothing else, because we're a literate society and accept text as authority. The reverse also applies - as GUMSHOE notes, randomization has very little place in investigative gaming.
Edited to add: After posting this, I remembered one other area where I significantly favor some form of random generation: magic item shops. If and when I decide that players can buy magic items, I don't want to open the whole catalogue to them. I want a short list of options to present to them, that they can purchase or not. Given what I inescapably know about the PCs (favored weapons, etc.), it's not really possible for me to invent a list of magic items without getting wrapped up in making them appeal to one player or another. A random generator helps me resolve this; the magic item shop really will be filled with curiosities.