In Response to Keith's Look-Ahead Article

Discussion in 'Game Design' started by Jon Perry, Dec 31, 2014.

  1. Jon Perry

    Jon Perry Well-Known Member

    This is a response to the article posted here:

    I think game designers need to stop talking in generalities about "hidden information." You have to consider the source of the information. Usually when people say hidden information they mean information that is initially (1) generated by a random process (2) or generated by player actions. These are distinct information sources and therefore should be discussed as separate game mechanics.

    Furthermore, in a single player game the word "hidden" might as well be replaced with "unknown." All the word hidden does is imply that the game *might* provide some deduction opportunities. "Hidden" makes a bit more sense in a multiplayer game where access to information is asymmetrical. Though for that we might substitute the word "private."

    Auro uses randomness. The map is generated randomly before play. This might seem like setup/input randomness but it is not. It is in-game randomness no different from drawing a card or rolling a die. Since the player doesn't see the whole map at the beginning of the game, they cannot plan accordingly. The fact that the algorithm generates the map before play begins is merely a technicality. There is in practice no difference between a map generated beforehand that I then explore, and a map that randomly generates itself AS I explore it. For all I know, moving in Auro triggers an internal die roll which then generates new tiles as they come into view. And while I have some time to react to those tiles, that is no different than having some time to react to some cards that I draw, or a die that I roll.

    Furthermore, once you see parts of the map in Auro for the first time it becomes possible for you to memorize those sections. Later if you scroll those parts off the screen, it is up to you to remember exactly what was there. Which is to say the limited view in Auro ultimately ends up just being a memorization tax placed on top of a deterministic game with a very complex lookahead tree of its own.
  2. Jon Perry

    Jon Perry Well-Known Member

    I would add that one practical way Auro does limit the lookahead tree as opposed to Chess is by having only one character to move as opposed to sixteen! I think this is a far more important distinction between the two games than the fog.
  3. Nachtfischer

    Nachtfischer Well-Known Member

    Well, the point is that it's (input) randomness that's sufficiently far away to allow for strategic planning to take place. Which is rarely the case with drawing cards or rolling dice.
    keithburgun likes this.
  4. keithburgun

    keithburgun Administrator, Lead Designer Staff Member

    So... your argument is that there's no difference between input and output randomness? That a map generated at the beginning of a game is as random as a dice roll that determines the outcome of your decision?

    Er... it kinda *is* different from having a die that you roll. You have literally 0 time to react to a die that you roll (assuming we're talking about typical dice roll stuff, like rolling for a hit in SW). It's just random information IMMEDIATELY activating.

    Well, ok, this is like a totally different thing, but one that I actually agree is a problem (the rest of your thread is totally wrong it seems to me). Yeah, I guess ideally we'd have maybe a minimap or something, or maybe the map zooms out (perhaps this could happen in the PC version) as you progress. Anyway I think it's a small problem, perhaps not worth introducing these features. But I'll admit at least that this is a problem.
    Nachtfischer likes this.
  5. keithburgun

    keithburgun Administrator, Lead Designer Staff Member

    There are things that are "unknown" and then there are things that are "intentionally unknowable". The optimal move within a given Auro board is likely unknown, but could be known. What monster will appear if I advance is intentionally and forever unknowable, no matter how much you play the game. I think this is a useful distinction for hidden information.
    Jon Perry likes this.
  6. donderper

    donderper Well-Known Member

    That is an interesting idea for a mini roguelike, possibly it already exists. A map that is deleted as it leaves the screen and each step new terrain in a given direction is procedurally generated.
  7. keithburgun

    keithburgun Administrator, Lead Designer Staff Member

    Yeah, that does already exist. At least twice I've seen it done. Actually DCSS even does that with its "void" or whatever which you can be sent to.
  8. Jon Perry

    Jon Perry Well-Known Member

    My argument is that there are two types of randomness - we can call them input & output. But my criteria for input randomness is very strict. To be input randomness:
    (1) All random information must be generated prior to the player making any decisions.
    (2) All that random information must be visible to all players.

    If both of these criteria are not met, then the player is forced to begin making decisions and building an initial strategy before key information is known. I don't see how this is any different than having to build a strategy while not knowing what cards you will draw or what dice you will roll. So for me Auro's map generation fails because of criteria (2).

    But more fundamentally my argument is (1) that I don't think you can have input randomness once the game has started. The reason is that if the player has begun making decisions than ANY random new elements are interrupting their strategy. Now, you can argue that in some cases games generate random information, show that information to the player, and then delay the time it takes for that information to directly affect game play. For example, a face up row of cards to be drafted might seem more "delayed" than a blind draw off the top of a deck. Or seeing the contents of a map hex a few spaces away might seem more "delayed" then blindly entering a hex. There is a distinction to be made here, in terms of the fact that the "delayed" options give the player more information and complexity to grapple with, which might contribute to a deeper game. But I don't think one option is actually more or less random than the other.

    Let's say you are playing a war game and you roll a die to resolve combat. You are going to win or lose the fight and that outcome, whatever it is, is going to guide your next move and the course of your future strategy. Therefore that die roll is an input to your next decision which might be to attack somewhere else, or retreat, etc. The only real difference between a combat die roll and a hand of cards is that the hand of cards is more dense with possible information whereas a combat die roll is often binary (win/lose). So the distinction is not input/output or time to react. Instead the distinction is between high resolution information (cards) and low resolution information (a simple succeed/fail die roll).

    Anyways, going back to your original article, your example was Hearthstone which uses cards. Hearthstone gives you a random hand of cards and then asks you to respond to those cards and decide what to play and when. I still don't see how thats different from reacting to the reveal of new map tiles in Auro?
    Last edited: Jan 1, 2015
  9. keithburgun

    keithburgun Administrator, Lead Designer Staff Member

    I feel that I addressed this in the article, but yeah, if you look at everything as "100% input randomness OR it's output randomness", kind of an all-or-nothing view, then OK, sure. I admitted in the article that having "mid-game input randomness" *does* make your game more random. But rather than saying "it's output randomness", it's more nuanced and correct to say that there is a spectrum between input and output, and stuff like "revealing tiles in Auro" would be a few more steps towards the middle (from hard input randomness).

    So from what I can tell, your "it's either HARD input or it's HARD output" binary view is where we differ.

    That alone is similar. The difference is that there are necessarily a few turns (usually 3-5) between "when you see monster X" and "when monster X can affect you" (due to positioning and the fact that monsters start on-cooldown). Here's the thing in Hearthstone - a guy can play a card AND have it activate on his turn (charge or something). And even if they have to wait until the next turn, that's only 1 turn.

    Further: Auro has ~10 monsters, which do different things, but nothing that like "wipes the whole fucking screen clean of everything" or anything. Hearthstone has HUNDREDS of cards with a super super wide range in how significant of an impact they can make.

    So, it's a matter of degree, although in this case I think it's a pretty gigantic rift. Hearthstone is input randomness WAY over near output randomness, and Auro is pretty far in the other direction.
  10. Jon Perry

    Jon Perry Well-Known Member

    You're right that this is where we differ. I can't get on board with your spectrum model. The reason is I don't really know how you are defining input/output randomness. You know my definition. From what I can tell, your definition seems contingent upon this idea that with input randomness the random generated information has a "delayed activation". But I'm not sure what "activation" would mean in this context. Within a game system, the appearance of new visible information is going to affect the players instantly - the way they evaluate the game state, the choices they make, etc. So I'm not sure there's a real distinction between information becoming visible in a game system and some later moment when that information "takes effect." Perhaps you can help me unpack this and come up with a more rigorous definition of what variable is actually changing along your proposed spectrum?

    Just to be clear, even though I find the input/output distinction fuzzy and confusing, I do agree that Hearthstone is more random than Auro. Hearthstone generates a larger amount of information via a random process. Specifically the arrangement of both player's decks is 30! possible orderings of cards and those cards ARE the whole game - what powers can be played etc.

    By contrast, Auro has significant parts of the game that are largely unaffected by randomness, and stay consistent from game to game, as well as fewer game elements overall (as you pointed out). Most importantly, at a certain point you've explored the whole Auro map and the randomness stops - whereas in Hearthstone you continue drawing and encountering new cards until the game ends. So the randomness in Auro has a kind of expiration date.
    Last edited: Jan 2, 2015
  11. major_shiznick

    major_shiznick Well-Known Member

    Imagine Hearthstone where you can always look at your next draw. Or three. Or ten. You get new information from your deck throughout the game, but it doesn't "activate" until later, when you actually draw it and can use it.
    deluks917, Jon Perry and keithburgun like this.
  12. Nachtfischer

    Nachtfischer Well-Known Member

    Right, as @major_shiznick pointed out above, the concerned variable is the distance between revelation of information and it actually affecting the game state.

    On a very technical level we could even keep the binary input/output thing. Input RNG happens before a decision, informing the player of what he can do. Output RNG happens after a decision, determining its actual consequences.

    The important factor though, is fairness regarding player agency. Both kinds of RNG can be unfair. Imagine a card game "Draw a card, play it". Yes, it doesn't even have decisions, but it's technically using input RNG. Yet it completely ignores agency. In actual card games we have an equivalent thing called "top decking". Drawing the optimal card can often immediately decide the match, because there's no distance between revelation and effect. This feels unfair and barely interactive.

    Auro's (input) RNG is fairer than Hearthstone's, because it's "further away".

    Output RNG is, I think, never fair in that sense.
  13. Jon Perry

    Jon Perry Well-Known Member

    I think we're mostly just having a semantic difference.

    I would describe the above situation as adding additional information and complexity to the game. Specifically you are adding information about not yet drawn cards. By making the game more complex in this way, you are increasing the burden of lookahead on the player, but also ideally making the game more deep and strategic.

    However, I would say that you've changed nothing about the level of *randomness*. The game is just as random as before, you have simply added an extra skill dimension. An extra skill dimension that comes with the player having more information. And that skill dimension is completely independently of randomness. So I'd say we should have a different name for this "delayed activation/information preview" concept that doesn't have randomness in the title - since randomness is not the real issue. You could just as easily apply delayed activation to information generated via player actions, dexterity, complexity, etc.

    So essentially, I would describe your alternate Hearthstone game as equally as random as the initial Hearthstone. But your version of Hearthstone would almost definitely be higher skill (as well as harder to play/learn, via the larger information burden on the player).

    I also think there's room to more rigorously define this "delayed activation." It sounds like we have two resources in game (a) knowledge of the identity of game elements & (b) all other game variables - score/money/hp. So we can say a game element has a delayed activation if it affects resource type (a) several decisions before it affects the values of resource type (b). Again, this has nothing to do with randomness. It's still a bit fuzzy of a definition though - since again, if I can see 3 of my upcoming Hearthstone cards, I am going to play my current cards completely differently. So in many real senses, those 3 cards have already ACTIVATED and are already affecting type (b).

    (By the way, if we're using the definition above, the higher costing cards in Hearthstone should also count as delayed activation. If something costs 6 and I get it in my opening hand, then 5 whole turns will go by before I can play that card and it "activates." So we should factor that into our analysis of Hearthstone.)
    Last edited: Jan 2, 2015
    skaugust likes this.
  14. major_shiznick

    major_shiznick Well-Known Member

    It is definitely not just a semantic difference.

    You note that my Hearthstone example is no less random than normal Hearthstone, and I agree. Probably most other people here who talk about input/output randomness would agree that one is not categorically less random than the other. The main use in differentiating the two is to talk about how to apply randomness fairly, like what Nachtfischer mentioned above ("fair" in the colloquial sense, not the statistical one).

    Your comments on delayed activation are confusing. The "interactions" between a player and a card in hand are definitely different than those between a player and the third card from the top of his deck. Not just a little bit different but very, very different. Just because you can use the word "interact" in both cases does not make them the same. Auro offers a clearer distinction. You cannot normally do anything with monsters outside your field of vision. You might know they are there, maybe even where they are, but none of the normal player-monster interactions apply until they come into view.
    Nachtfischer likes this.
  15. Waterd

    Waterd Well-Known Member

    After watching the 3 minute video I started to wonder if this difference really exist, after it was more clearly explained in this video, suddenly something start to make less sense.

    Why isn´t roll dice after attacking also input randomness?

    After all unless at the end of the dice roll, the game ended, I do in fact have time to response to the dice roll. The dice roll event, present me with a new scenario, and now I have time to respond to that scenario.
    In the case presented of battle for wesnoth. I attack that unit, and now I will have a defined scenario, now I can respond to the scenario that was developed as part of the dice roll. how is that different than having to respond to the Map setup?

    Objectively I can´t make the difference.
    Jon Perry likes this.
  16. keithburgun

    keithburgun Administrator, Lead Designer Staff Member

    Waterd, I'm gonna move this comment into the other thread where me and like 3 other people explained the difference, literally today we just kinda went through this. If you read all the posts there and still can't understand I'm not sure I can help.

    Edit: moved.
  17. keithburgun

    keithburgun Administrator, Lead Designer Staff Member

    Waterd's point is basically "output randomness is input randomness for the next turn". Yeah, that's true. So if you don't acknowledge the idea of a game having a structure that adds up to more than the sum of its turns - i.e. taking advantage of form - then yeah sure. Basically what I mean is, if you think that "one turn of Summoner Wars" is a complete game, and a match of Summoner Wars is just an arbitrarily long series of SW matches, then... okay.

    But if you look at SW as a complete thing, with a beginning, middle and end, that's one holistic machine, then all of that noise getting injected between player decision and outcome is actually screwing with the outcome of the match (the game end state) like crazy.

    I just had a huge email argument with DanC about this. I think one of the biggest differences between myself and most people who are used to games is that I recognize the potential of a structured game match - one that is not just "do the same thing over and over until X happens", but one that is structured kind of like a story or a song in a smart way to maximize elegance.

    If you believe such a thing can exist, then obviously output randomness is horrible. If you, like DanC and maybe waterd believe, that a single turn in Summoner Wars is a complete match of Summoner Wars, then okay sure, the above claim makes some sense.
    Nachtfischer likes this.
  18. Jon Perry

    Jon Perry Well-Known Member

    I think we are close on some points and obviously apart on others.

    You agree that "Hearthstone" and "Alt. Hearthstone w/Lookahead added" are equally random. You say that Alt Hearthstone is more "fair." I would say Alt Hearthstone is more "complex" and more "skill testing" but sure, I'll accept fair. There is obviously a distinction between the two versions. Just the distinction is independent of randomness, which is why I think using input/output randomness to describe this distinction is misleading terminology. Alt Hearthstone does not alter the random content, it merely adds more information to the player in the form of "lookahead". This technique could be just as easily applied to a game with no RNG, such as a game where your opponent's moves are visible but do not take effect until 3 turns later. I'm arguing for different terminology to describe this phenomenon. This is the more important point I want to get across.

    Where we potentially start to disagree on a more fundamental level is on whether we can characterize this extra "lookahead" information as "delayed activation" and whether that characterization makes any sense.

    First of all how are the following two things different from the point of view of delayed activation?
    (1) Knowing what my next card draw will be
    (2) A card in my hand already, that I cannot cast yet because I don't have the requisite mana

    Aren't these functionally the same? I mean, one is in my hand and one isn't, so if some effect triggers off my hand specifically (like a discard effect) then yes there is a difference. But they both mean that several decisions will occur between when the player learns the identity of a card, and when that card begins affecting other game variables.

    Now if you can agree with me on the above case, then we move to the harder issue that is bugging me. Which is that if I know what the next card will be, then I should (if I'm playing optimally) be considering that card in my decisions. Meaning that which cards I play, which moves I make, which HP I gain/lose/etc, - actual real in game variables - are going to change as a DIRECT result of me simply *knowing* the identity of that card coming up. So isn't the card already activated? Especially since constructions "like in-hand" and "not-in-hand" are just constructions anyway?
    Last edited: Jan 3, 2015
  19. Jon Perry

    Jon Perry Well-Known Member

    Perhaps I'm now in the wrong thread so I'll go check out the other one. But I agree with Waterd, and am completely confused by your response. I would argue that YOU are the one asserting that each turn is a mini-game unto itself. You have to isolate a turn from the rest of the game to imagine that a turn has its own special input & and output independent of neighboring turns.

    A holistic look at the entire game asks "how over several turns is the player forming their strategy and how is it forced to change?" From that perspective, ANY new random information is strategy altering. Now, here's a specific compliment for Auro. What's cool about Auro is that the randomness expires (you eventually explore the whole map). So in terms of narrative arc, Auro has more randomness in the beginning of the game then at the end of the game. So if you look at the whole game, because of this expiration concept, Auro is achieving some of the "input" qualities you are suggesting. But the reason is the expiration - the randomness faucet effectively turning off - not the fact that information is being somehow previewed to you beforehand. And most importantly this effect is happening at the "game-level" and not at the "turn level."
    Last edited: Jan 2, 2015
    Dasick and Waterd like this.
  20. Jon Perry

    Jon Perry Well-Known Member

    I can't find this other thread. I'd like to read it so I'm not trodding over already discussed ground.

Share This Page