Thursday, March 23, 2017

Episode 134: Randomness

This is episode 134 of Roguelike Radio, where Darren Grey and Mark Johnson discuss the role of Randomness in roguelikes.

You can download the mp3 of the podcast, play it in the embedded player below, or you can follow us on iTunes.



Topics Discussed:
  • Randomness in this context refers to gameplay mechanics rather than the procedurally generated elements (dungeon levels, treasure etc.)
  • Roots in Dungeons and Dragons dice based combat system
  • Randomness averaging out over time - does it even have an impact?
  • Input vs output randomness (what you see vs the results of your actions)
  • Luck vs Chance
  • Too much randomness meaning that the player can’t make an informed decision
  • Poker mechanics in roguelikes?
  • Why do we have chance to hit mechanics in roguelike combat?
  • Requiring knowledge of all the potential random outputs in a scenario
  • Communicating chance and player understanding of probability
  • The Monte Carlo fallacy makes it difficult to communicate chance to the “Average Joe” gamer
  • Risk-reward scenarios and comparisons with Poker
  • Interesting chance mechanics in board games
  • Mechanics to mitigate against randomness
  • AI and the predictability of foes
  • Interesting 'AI in Games' Youtube channel/podcast
  • Misuse of the term “Skinner Box” when referring to random game mechanics
  • The unending despair of roguelikes!

Join us next time for a look at this year's new batch of Seven Day Roguelikes.

14 comments:

  1. This comment has been removed by the author.

    ReplyDelete
  2. Nice, really better than the usual radio when i'm going at work!

    ReplyDelete
  3. I've been musing about if the Yahtzee mechanism (as it has been dubbed in some circles) of things like King of Tokyo can be harvested in a dungeon crawl board game or Roguelike video game for some time now - similar systems seem to work for cooperative disease fighting (Pandemic: The Cure), afterall, though without even an idea to show for said musing, so glad to hear that thought has at least crossed someone else's mind...

    One thing I would say, however, in response to the discussion on if a simpler or more complicated is going to be identical in experience for 99% of players, should we favour one or the other - I don't think that's as black and white as the two of you made out. If one or the other system isn't going to be noticed by the majority of players, then it's neutral for those players. As you pointed out, most players aren't going to care what's going on under the hood to generate that 97% chance vs 92% chance, and since I don't have to do the calculation myself like I would if it were a board game, it doesn't cost me, as a player, anything either way. So use that wonky formula that's tuned just right to get the distribution that's ideal for your game and took you five days of number crunching to figure out even if a simpler 3d6 or 1d20 system would be close enough for 99% of players, since unlike in analogue games those 99% of players don't have to care which you used - It's the cases where the 99% of players are going to notice and care that you should always go for the simpler of the two.

    One thing that's frustrated me in Imbroglio plays is the weapons that have random effects, particularly stuff that can go off for multiple of the same type of weapons per turn, is the lack of clarity on if it's cumulative chance (so four weapons with a 25% chance of an effect happening having a 100% chance of that effect happening) or independent chance (which would put the same scenario as a 68% chance), which is simply down to the language in games tends to be the same whichever it's going to be, and most rule the opposite direction (cumulative) to Imbroglio (independent). That's not a flaw with Imbroglio, as such, more another aspect to the problem of communicating chance to players - I think humans are more comfortable with linear stacking to the somewhat weirder of multiple random effects impacting off of a single event. And I don't think reworking from 25% to 1 in 4 would help in that case.

    It's interesting you brought up bridge, and the acceptance of chance in that, in the same conversation as you later brought up the different psychology of computer randomness vs physical randomness... Humans suck at shuffling cards, we really, really do. When I was playing bridge in secondary, the club I played at was gradually shifting from human shuffled cards to computer shuffled cards with barcodes. But because humans suck at shuffling, and bridge has a degree of sorting a deck, which while lower than some other games is definitely present, a computer shuffle gives you weirder hands more often, because these hands that should be coming up once a drive on paper are now coming up once a drive, rather than with human shuffling once every other drive or once ever three drives or whatever. And the assumption, as it is for MTG players who switch from physical to the abhorrent UX of MTG:O, that the better shuffling the computer is doing is less random, even with physical cards, because the players knew that they're now shuffled by computer. Humans suck at shuffling, but we trust computers to shuffle cards even less.

    ReplyDelete
    Replies
    1. Thanks for these interesting thoughts! Your comments about the 99% and the experiences being generated by deep or simple systems is a good one, and I see what you're saying. Different kinds of chance in terms of communicating to the player: yeah, there's definitely something interesting there about the transparency (or lack thereof) about how we convey this information.

      However, to me, your final paragraph is the most intriguing; I'm always fascinated by how these kinds of social/technical procedures wind up (un)intentionally affecting the outcome of a game by affecting the distribution of unpredictable elements. I don't suppose you have some links to elsewhere where anyone else might have discussed these issues? I'd like to read more!

      Delete
    2. Magic forums have some debates on if human shuffling or , since there's enough of a noticeable difference in the distributions of a deck once shuffled for people who play both formats to notice - Swingier land distributions are more likely with a machine shuffle, basically, because the 1 in a thousand shuffles happen about one in a thousand hands rather than the one in three thousand or one in ten thousand they come up with a human shuffle (That's not accusing anyone of cheating or deliberately stacking the deck, simply... Humans suck at shuffling)

      For Bridge - http://www.nytimes.com/1999/07/01/technology/computerized-machine-steps-in-to-shuffle-cards-by-the-ton.html is an article on... Oh, wow, that's the right period for the version my old bridge club switched to... Had no idea that was a scaled down, commercial, version of an industrial scale card shuffling machine... But doesn't go into the difference in distributions.

      http://bridgebase.com/forums/topic/74974-hand-dealt-vs-computer-generated/ has some discussion on the topic of human vs computer hand distribution for Bridge (Apparently there's disagreement on which way hand shuffling is swingy, but the general consensus is that computer shuffling is closer to the theoretical distribution) about the direction of the problem for Bridge.

      More subtly than spectacular vs unspectacular hands is the side discussion on finesses and that AQ finesses are more likely to find the K in computer shuffled hands than human shuffled ones (at least compared to Rubber, and that someone developed a different style of finesse that seemingly exploits human shuffling!)

      Delete
  4. Excellent podcast as always.

    I'm not sure if you are hurting for ideas for new topics, but I figured I'd throw out an idea for one.

    Recently I watched a youtube video in regards to characters attributes and stats - things like Strength, Agility, Intelligence, etc. While not exclusively a Rogue-like thing (and not used in all Rogue-likes either), I think it might be worth a discussion to talk about.

    Specifically on the topic -

    Number of Different Attributes
    Realism versus Game Design
    Character Building
    Games With Multiple Races or Classes
    Overlapping Attribute Effects
    Rogue-like Games With No Attributes Vs Those With Them
    Attributes & Stats from Leveling Versus Equipment
    etc

    ReplyDelete
    Replies
    1. Glad you liked it! We do have a pretty long list, but at the same time, we are indeed always looking for suggestions. Some of those are definitely things we'd be interested in exploring in the future!

      Delete
  5. Mark brought up an interesting point about how a poker player's strategy varies depending on if he's playing in a tournament or in a cash game. Personally, I think the "cash game" paradigm is a better way of judging skill for most rogue-likes. That is, take the moving average of your performance over your last 20-40 hours of play (whatever that works out to be in terms of playthroughs). For this reason, I favor shorter rogue-like(-like)s such as FTL and Brogue because they generate more samples in the same amount of time.

    To me, randomness in rogue-likes is like the spice in your dish. You only need a little to alter the flavor and too much ruins the meal. To illustrate this fundamental idea, I partition randomness into three types: the Good, the Bad, and the Ugly.

    - The Ugly is what Mark described at the beginning of the show: a random value that is sampled to infinity so that it reaches a distribution. This isn't really so much about randomness as it is about how the game designer uses it, because randomness in anything disappears as you tend to infinity. Even the randomness in the example of the 1d20 club that Darren provided isn't loses its meaning if each enemy takes ten whacks to defeat. In that case, you're using randomness to create a dependable game mechanic. Thus, you need not strive to eliminate all forms of this sort of "ugly" randomness in your roguelikes, because it can be useful (and it would probably be hard to do so anyway), but just be aware that it won't lead to random outcomes in the greater scheme of things.

    With the Ugly out of the way, only Good and Bad randomness remain, and these are the random events which are only sampled a relatively small number of times. "Good" randomness is that whose effect is not immediately obvious: for example, say a dungeon always spawns a legendary item on the third floor. Most of the time, you'd want to keep this item, but you'll have to form your play around it. "Bad" randomness, on the other hand, is that whose consequence is immediately obvious and so consequential that it reduces the outcome of the game to a coin-flip. For example, you are low on health and you have 50% chance to kill a troll. If you miss, then it will kill you in one hit. (Conversely, you wouldn't want to include a mechanism that gives you God mode if you roll a 12, but I think that goes without saying.)

    It is the job of the game designer to sprinkle in Good randomness while allowing him to mitigate the Bad randomness as much as possible.

    ReplyDelete
    Replies
    1. I agree about "overall" roguelike skill; as many have pointed out before, once you're good at a classic 10-hour or 20-hour roguelike, winning is less about playing well, and more about "making no mistakes"; or rather, one only loses by making a stupid mistake one shouldn't have made, and yet that mistake has a catastrophic effect hitting many many many hours of play.

      Your three-part definition is interesting, and one I think I'll need to ponder a little longer. I think what you call bad randomness is a crucial point; I think the other two represent a very incisive difference and overall set of three well. I think this kind of discussion always draws one's attention to precisely how systems of unpredictability are deployed; the same system can be trivial or vital depending on context, for instance. I'll think about these some more, but they're great examples of contextual deployments of similar systems...

      Delete
    2. I agree with the line of thought about mitigating randomness -- in traditional roguelike games this is typically what HP is for. You can swing with a Vorpal Blade and hope for the instant kill but if you're wrong you have some HP to afford the bad luck.

      I agree with the above poster about "Bad Randomness"; although I would expand that to include randomness that is possible but annoying to defeat. Eg. instant poison spikes in Nethack - you have a better chance (albeit not perfect) of finding spikes if you search tediously every square. If you're a good enough player you might be able to still move, kill & eat efficiently that this isn't a problem (I think? I'm not great at Nethack!). The better strategy than running through fast is an annoying and necessary one - risking poison spikes doesn't just cost you hp but your entire game. If poison spikes were more plentiful but cost you HP & hunger instead of instant kill, the choice to search or rush would be better gameplay.

      In Nethack the better solution here is to get poison resistance or levitate, which while mitigating the randomness is kind of boring in a way, there's no good reason *not* to get poison resistance.

      --

      All of that said, I'm still kind of fond of the instant kills in Nethack for the same reason that Wizards in MUD & MUD2 would trick & outright kill ascending players - only players with dedication and the skill to ascend in *multiple* playthroughs can make it. If Nethack had more "gameplay choices" than "bad randomness" it might be a better game but lose some of its charm. "Dungeons of Doom" should have some bad, unavoidable dangers for its namesake.

      Delete
  6. You talk about "luck" (no player choice) and "randomness" of map setup; I think adding "gamble" or "gambit" would make sense - the player "takes a gamble" on the die roll -- the roll is luck but the choice to roll is the player's

    ReplyDelete
    Replies
    1. This comment has been removed by the author.

      Delete
  7. A case for randomness: You can use randomness to make a game feel more real.

    In the physical game of billiards (or snooker) there is a little bit of randomness/unpredictability involved.
    Players play over four edges (or whatever they are called) but they don't play over twelve edges, because it wont work like they predicted.

    Most physical games have unpredictable factors.

    In contrast, a game like Invisible Inc. is very unrandom and predictable, which makes in feel more unrealistic. (I'm not saying it's bad!)


    Marc mentioned that boardgame players blame randomness less for losing than computer gamers. I think that's just because the board gamers Marc heard from are problably more professional tournament players or something like that whereas the computer gamers were more "casual gamers".

    These board gamers are simply correct: Just because a game involves randomness, you are still responsible for winning and losing. (As long as you can still make enough meaningful decisions. Not like the "Game of the Goose".)

    E.g. X-Com (FTL, Magic the Gathering, ...) is a game where unexperienced players will claim they lost because of bad luck, whereas very experienced players will win a majority of times.

    ReplyDelete
  8. Just got a new roguelite today called "Slay the Spire." It's a card game similar to Magic that's been augmented with rogueish elements. In this game, you are randomly dealt cards from your deck on each turn, but each card produces an effect that has a definite, known outcome. If one were to apply this usage of randomness to a traditional roguelike, it would be like giving the player a magical staff that can only hold three spells at a time, but where the effect of each spell is guaranteed to apply. Casting a spell would then charge the staff with a new spell at random.

    I highly recommend buying Slay the Spire just so you can appreciate the use of randomness.

    ReplyDelete