Monday, May 19, 2008

Precise planning vs Conditional planning

I'm coming to realize that there are 2 different types of planning you can do... Precise planning and Conditional planning.

Precise planning is when you have all the information you can, and the only variables are player-created chaos: what will the other players do?

Conditional planning is when there is an element to plan around that you do not know for sure, based on chance. Often you can use probability to decide how to go about planning around this element.

Some gamers prefer to have all the information, they prefer to plan precisely, dealing only with the chaos of the other players' actions. Other gamers much prefer preparing for the random element as well. Let's look at some examples...

Railroad Tycoon is one of my favorite games, so I'll use it as an example. There are 2 instances of Conditional planning in RRT:

  • Random cubes which come onto the board via City Growth/Urbanization/New Industry
  • Whether or not a particular Major Line or Service Bounty card will come up.
Under the standard rules, when someone adds cubes to the board, 2 random cubes are drawn from the bag. The result of this could be that a player in the late game chooses City Growth and either gets a couple of 2 point deliveries, or a couple of 6 point deliveries. At that point in the game, 2 point deliveries are fairly worthless, while 6 point deliveries are great. The random draw could decide the result of the game for that player (1st vs 3rd). The obvious answer to that is "build your network so that no matter what cube comes up, you'll have a decent delivery." That's a good point, and an interesting consideration when planning your network, but the problem still remains - suppose my opponent doesn't plan his network as well, but lucks out on the draw, and gets an 8 point boon?

Similarly, you can build track such that you can connect Baltimore and Toledo, Atlanta and Richmond, New York and Chicago, etc... and then whether you win or lose the game can come down to whether the Major Line card you've prepared for comes up. The player in the Southeast could get an 8 point boost because the Atlanta-Richmond card came up early, while the player who's built from New York to Kansas City is in last place because that major line never came up. While it's true you can't expect any particular card to come up. just the benefit gained by a Major Line coming up early in the area you're dominating is a huge advantage, and without knowing which Major Line will come up or when, there's no way to know where to build so as to take advantage of that boon.

Recently I tried playing Railroad Tycoon with a couple of variants to turn some of the Conditional planning into Precise planning:
  • Remove Major Lines from the deck and make them all available from the outset.
  • Display 2 cubes from the bag to be used whenever the next 2 "random" cubes are needed.
I enjoyed both of these variants because I think I prefer Precise planning to Conditional. It removes some of the uncertainty and maybe the need to build your network such that any cube is good, but it adds the ability to decide if and when to take actions which add cubes to the board, and where to put them if you do. It also takes the luck-of-the-draw out of the Major Lines, which are worth a significant number of points. I might like to try beginning the game with the Service Bounties all available as well. They don't seem as bad as the Major Lines though because they aren't worth as many points, and when they come up they can often be sniped by any player, not just the player who happens to have built in the area.

Reiner Knizia tends toward Conditional planning in many of his games. Ra for example is a perfect example: you evaluate each auction lot and make the decision to bid, or to wait for a better lot. This is something of a press your luck mechanism because whether the lot increases or decreases in value and by how much is entirely based on chance, not on player actions. A friend of mine cannot stand that, and feels "why should I even play if I'm just guessing if this is the right play or not?" While it's clear I prefer Precise planning to conditional planning, I don't mind as much as that friend seems to - I like evaluating the lot in Ra and thinking about the chances it'll improve. I likened my friend's play to "being greedy."

I think the main difference between Conditional and Precise planning is the idea of "working with what you've got," and it's tied to the difference between Tactical and Strategic thinking. In a game with precise planning, you can plan out a long term strategy, and your tactical play can deal with the chaos brought in by the actions of your opponents. In a game with Conditional planning you have to play a more tactical game, and there's less (if any) semblance of a coherent long term strategy.


ekted said...

For the most part, I don't differentiate between player choices and chance, other than to call them chaos and randomness. They are both unknowns and have probabilities associated with them. You can guess at likely results, and in the chaos case may be able to control them somewhat by negotiation, but it's still an unknown. The reason people care is that "random" player choices smack of intent. But consider playing against a good AI. Its choices are based on semi-predictable decisions about the game state, but still have actual (pseudo)-randomness.

Seth Jaffee said...

I can't agree with you there, Jim... Players base their choices on the board and on logic, things you can see coming, and can influence through your actions (I'm not talking negotiation or social dynamics here). Chance is just chance, maybe there are probabilities involved (dice, card distribution, etc), but you can't look at chance and make the same logical choice that it will and therefore predict it's action like you can an opponent. The best you can do is play the odds.

Shea said...

I view things much as Jim does.

Game induced randomness follows a generally calculable probability distribution.

Player created chaos can still be modeled with a fuzzy probability distribution. You have to do calculations and assumptions on the fly. The better you know your opponent, the better you can try and fill in the exact probabilities.

In fact, that's why I try and make my own actions unpredictable and unusual at times. Sometimes though it can be good to be known for a certain pattern. For example, I refuse to participate in "pass the buck" kingmaking. If you could have stopped the leader, but instead passed the problem on to me, I will gladly just let the leader win. If I do that enough, you then know what your chances of winning are if you expect me to stop the leader and change your assumptions accordingly.

Seth Jaffee said...

Hmmm... I wonder how prevalent this equation of Chaos and Chance is - because I don't think they're the same at all.

The end result may be similar, and to an outsider on the sidelines of the game looking at any given player action or random occurrence the one might look very similar to the other... but that's like looking at a black box and saying "oh look, those 2 inputs produce the same output" when in fact the process for each might be very different.

As an example, say I constructed 2 equations and didn't tell you what they were. I put each of them in a box, and then drop the number "14" into each. Each one spits out "7"...

Equating chaos and chance is like looking at that example and saying "oh, those equations must be pretty much the same thing" when in fact one could be "X/2" and the other could be "choose a random number between 1 and 10". Knowing that information it's clear that one box is highly predictable and the other is not.

So that's how I look at it - sure, Chaos can create a result that looks similar to Chance, but since Chaos is created by player choice, the motivation for the result is entirely dependent on things which are knowable or observable. That's the opposite of Chance!

Shea said...

I don't think they're exactly the same, that's why I try to differentiate them when I discuss games. Through the Desert has no "randomness", but it has plenty of "player chaos".

However, I think they serve the same purpose in games. To keep the game incalculable. I think Shogun is probably my favorite planning game. Why? Because I'm offered the choice of maximizing my potential benefit versus minimizing my potential losses. I usually try to strike a comfortable ground (I am an actuary after all :P )

Still, I think you can at least approach chaos and random with the same approach. I can calculate the probability of rolling an 8 in settlers. I can also hazard a guess at the probability of Bob cutting me off from that Oasis in Through the Desert. It's not exactly the same calculation, but many of the same principles apply.

It's pretty funny when this comes up in review commentary. I usually at least distinguish between chaos and random from a design perspective, but I think they both do a good job of adding tension and unpredictability to a game.

Seth Jaffee said...

Randomness serves the purpose of making the game incalculable - it's imposed by the design of the game. Chaos serves the purpose of attempting to gain advantage over other players or profit for ones self - it's imposed by the player in his own self interest.

Shogun has a lot of Conditional planning, too much for my tastes. I've actually not played it, but I played Wallenstein online a handful of times. It was alright, especially PBW, but it's not the kind of game I get excited about sitting down to for a couple hours.

I'm not sure it's helpful or behooves a player to treat Chaos like Chance. While it's true, you can approach it the same way... and maybe it's just semantics because if you properly adjust your assumed probabilities when dealing with Chaos then maybe it really is comparable... because of the motivations I think a player can "do better" when dealing with Chaos than with Chance.

It could be that I just see a higher accuracy of predicting Chaos than Chance, so less often you have to do your calculation, make your play, and then wait and see if it pans out for you.

ekted said...

If I was going to draw a card, and I knew the frequency distribution and what's been drawn so far, I could calculate the probability of drawing any given card.

Given a game state, I might assign probabilities for an opponent to choose a particular action. Maybe there's an easy choice for 5 VP's, and a much tougher choice for 8 VP's.

Say I decide the probility for drawing a card is 40%, and the probability of a player choosing an action is 40%. The former is absolute if my math is correct. The latter is based on my ability to read the game state and the psychology of my opponent. But once I calculate a probability, what else is there to either decision but the math?

You can say the probabilities of chaos are fuzzier, but I don't see how your choice can be based on anything but the probabilities. Subconsciously, you may be altering your math because of body language, etc., but it's still probabilities.

ekted said...

Your black box example doesn't really fly because it's based on a single data point. With less information, the results are going to be less predictable. But knowing the game, how to play, how to win, and the person, changes things. The black box is really the "randomizer".

Sean McCarthy said...

Leave it to someone who prefers precise planning to call the other option "conditional". :)

Conditional planning, as far as I can tell, is what you're left with if you try to do precise planning in an environment with uncertainty. By nature, it's more cumbersome than precise planning. It takes more mental activity to plan the same number of steps ahead, because you have to branch more.

There are people, myself included, who don't like precise planning that much at all. The reason we enjoy some randomness in games is because precise planning is less valuable if it has to be conditional. With enough uncertainty, it's reasonable to drop the calculation almost entirely and rely on a more intuitive understanding of the game overall. That's the other kind of planning - not conditional planning, which is just a subset of precise planning.

Personally I think of precise planning as tactical planning, and intuitive planning as strategic planning. If you are counting the exact number of something instead of just seeing it as a vaguely-sized quantity, that's tactics, not strategy. I like my games to have some strategy in them.

In conclusion, I dislike Power Grid.

etothepi said...

Jumping in...(I'm the friend mentioned above)

I'm not a fan of either randomness or chaos. Randomness, however, upsets me more, particularly when the randomness is the equivalent of a 1D6 roll (i.e. a flat probability distribution).

I would much rather play a game where my actions are consistently able to do naught but good for myself, rather than a game where I am calculating a screw-factor for another player. My girlfriend once said something which I have since treated as a gaming axiom: "20% of a good thing is better than 80% of a bad thing." It is when I lose sight of this "axiom" that I find myself performing the worst.

It is the latter of each of those two statements that lends itself most heavily to chaos. If I am playing a game in which a viable way of getting ahead is by hurting another player (Moai and Winds of Plunder were recent games I've played in which this was the case), then I need to do a "chaos check," something which I do not enjoy doing to a large extent.

But since this discussion is on the difference between randomness and chaos...

First of all, I generally make a serious chaos check only when I know a game well enough to know what things you are truly looking for. Otherwise, anything beyond superficial calculations are equivalent to randomness.

You first need to check how well the players know the game. Any new player/player who hasn't played in a long time can not be trusted to make a "good" decision.

Next you check for what random elements they know about that you do not (hidden cards, etc). This is simply randomly enforcing their decision and is not "truly" chaos.

If you've made it this far with all those checks not failing to an equivalently random result, continue with the calculation. Otherwise make a quick estimation and don't waste your time.

In the former case, the following is questionably optimial:

Calculate what the best two or three possibilities are for the next player. Assume they do the "best" case, then check the top two possibilities for the next player with this decision in mind. Continue with the third player in likewise fashion. Assuming there are four players, you have now looked at what you perceive to be everyone's best options. How does your situation look? If it's bad, can you change it simply? Does this decision affect the chain of "best" decisions you've calculated? If so, go through their new "bests" and see how you fare. If not, check second cases for disaster.

This is essentially a chess computing algorithm. This doesn't work with a normal game structure, for several reasons, primarily time and a need for paper to keep track. Your games would, properly calculated, last too long to be worthwhile anymore. Not only this, but you are still making assumptions as to players' best moves. We all miss things from time to time (largely due to a lack of thinking time spent on the problem), so the "best" for a player might not actually be the "best."

But you want to look at *something* to estimate the state of the game, so you estimate a player's likelihood of choosing their best actions. Again, passing the first randomness checks, you take into account if the player is a risk-taker or not, and whether this "risky" move is more likely or not. You can calculate probabilities if you want, but you're unlikely to be correct.

Let's assume I have perfect information about a given player, and their playing style, and how often they take a riskier play. Also we are assuming that one play is safer, but gives less points, while another is riskier with more points, and these two ARE the best options, unquestionably. In this situation, I can ESTIMATE the percentage of chance they will take a particular action. If they talk about their own estimation of percentage, however, it is very likely to be different than my estimation. Even with knowing how often they've taken that risk before this situation and gauging the merits of choosing one situation or another, I cannot truly estimate their likelihood for a single choice.

Said in a faster, clearer way, this is in the eye of the beholder. Random probabilities are, at least, consistent across the board, and can be known precisely. Chaotic probabilities are guesswork at best, and any probabilities guessed at are truly a waste of time and inaccurate by definition.

I do not know, but highly suspect that due to a "collapse the wavefunction" effect (due to there being inherent conscious interaction), probabilities simply do not exist within chaos, and that each element of chaos is more in line with a random 1D6 roll.

With this in mind, I will reiterate my desire to avoid games in which both flat probabilities and screw-your-neighbor/leader feature prominently in numbers greater than 2 (in "gamer's games - I mind this less in "fun games").

Seth Jaffee said...

Leave it to someone who prefers precise planning to call the other option "conditional". :)

Heh, cute ;)
By "Conditional" I mean planning based on information that is not known, and by "Precise" I meant planning based on information that is known. I think by mentioning the player induced chaos I inadvertently derailed my own thread!

In an IM etothepi mentioned this with regard to precise vs Conditional planning:
I don't think of conditional planning as really planning. You calculate the odds and roll the dice. You plan too much, and you waste your time.

I'm not sure I agree with that, I think you make a plan based on the good outcome, and make a plan based on the bad outcome, and that's Conditional planning.

etothepi said...


I'm not sure I agree with that, I think you make a plan based on the good outcome, and make a plan based on the bad outcome, and that's Conditional planning.


With enough uncertainty, it's reasonable to drop the calculation almost entirely and rely on a more intuitive understanding of the game overall.

This general idea is what I refer to.

Also, defending myself in Ra:

I "played to not be greedy" and got some large number tokens but nothing else in the first phase. I "played to be a little MORE greedily" in the second phase and didn't get anything in the second phase. I played to be "much more greedily" in the third phase and got 75% of my points (still losing considerably). Greed had very little to do with it, it was simple randomness. And yes, with as large of a sample size as Ra has in the first two rounds, I believe that the distribution is nearly a 1D6 roll.

Seth Jaffee said...

ekted said:
Your black box example doesn't really fly because it's based on a single data point. With less information, the results are going to be less predictable. But knowing the game, how to play, how to win, and the person, changes things. The black box is really the "randomizer".

Referring to player actions as "random" is like the black box example... it's looking at a player as a black box, with a known input (current game state), and a "random" output.

In reality a player should have more information than that. Opponents are not black boxes.

The deck is like a black box where you know some info about how the input turns into the output. the process is random though, unless you're Rain Man or Monk and can track the shuffling of the deck... heh.

It's been postulated that one can assign a probability to player actions ("how likely is the player to do X"), and then assuming you've assigned a probability, at that point the player's action is just as random as the deck of cards with a known distribution. If that's the case then I submit that the probabilities should be [i]much[/i] better defined... if there's a 3/5 chance of drawing a card that will give you a Red resource, that's 60%. If you're wondering if your opponent will choose to produce a Red resource, you should have that down to 80 or 90%.

So sure - you can assign approximate probabilities to opponents' actions and then treat them like randomness, but even in that case I feel like it's not the same thing - because you can act with more certainty because the probabilities are (or should be) more definite.

The reason I think they'll be more definite is because they are not random. Of course the action an opponent takes might be based on information you don't know, which could make it harder to predict, but their behavior is still not random, and therefore you should have more certainty when predicting their play.