Daniel Cook recently posted an interesting piece on the incorporation of random noise into skill-based games.  He starts with the distinction between games of skill and games of chance and suggests that introducing noise– an element of randomness– into games of skill can potentially enhance their value. In particular, he suggests that one can develop skills at ‘mastering randomness.’  He provides a theoretical framework for thinking about randomness and noise in games, identifies different types of noise, and talks about considerations to keep in mind in terms of player skill acquisition.

At its root, he argues that players can develop a statistical model of uncertainty in a game and select actions based on that model.  So even in games of chance, there are probabilities associated with different random events and a good player learns those probabilities and makes game decisions accordingly.

What is more interesting is the introduction of a degree of randomness in games of skill.  Some of the comments to his piece objected to the introduction of an element of chance into a game of skill, arguing that a skill should be a skill and introducing probability decreases the value of developing a skill in the first place.  To an extent, this is a reasonable objection.  If the acquisition and deployment of a skill– let’s say killing the monster– is rendered pointless by an invisible dice roll that ‘cancels’ out the skill, then the game of skill is not really a game of skill, or is a game of skill only with a certain probability.

However, rarely is a skill simply a matter of emitting a precise learned action at the precise right time. There is almost always some degree of uncertainty and variance.  When the ping-pong ball comes flying at me from the other side, I can assess its velocity, trajectory and spin, but my assessment will always be within a range, not exact.  Subtle differences from one time to the next in, for example, the spin on the ball. . . which may be beyond my ability to directly assess and detect. . . will require subtle differences in how I hit the ball.    So while mastering ping pong requires the development of specific skills, there is always a degree of uncertainty in applying these skills to a specific instance of the ball coming at you.  In this sense, mastery of ping-pong requires not only the development of skills such as a good backhand, but also a model of the success of these different skills under different circumstances.  The best players maximally reduce uncertainty as to which circumstances require which subtle variation in response.  In short, even with the most highly skilled ping-pong players, there is an element of guessing, but for masters of the game it is a very refined guessing within a very narrow range of uncertainty.

So in this sense, introducing a degree of uncertainty or randomness into skill based games can make these games more like real life.  But why? As one comment pointed out, being like real-life is not a requirement for excellent games.  I would argue that building games where killing the dragon is more analogous to ping-pong– that is, where the development of skill requires working within some space of uncertainty – provides a better mechanism for titrating the difficulty and challenge of the game for players and, in general, is likely to generate more engaging games across a broader spectrum of players.

Let’s take killing the dragon as an example.  As players acquire skill, how do designers make dragon-killing more interesting and keep the game challenging? You could make it more difficult by requiring increasingly more precise blows, by requiring faster actions or by requiring the player to kill more than one dragon at a time.  At its core though, this amounts to acquiring one repetitive action and being able to perform it very well and very fast. And there is a limit to how fast and accurate it can be performed. . . both a limit for individual players (some will simply have slower processing/reaction times) as well as an overall average limit for human beings.  The player with an intrinsically faster reaction time will do better than one with an intrinsically slower reaction time, having little to do with skill acquisition per se.

When you introduce a little noise or random variation to killing the monster, you make the skill acquisition a little more like ping-pong.  Specifically, instead of ‘skill’ consisting of performing the exact same movement ever faster and more precisely– and a game that requires ever faster and more precise execution– the skill represents a range of variations on a action that generates a range of feedback.  So the number of times the monster has to be stabbed may vary, the precision of the required targeting may vary, where exactly ‘bullseye’ is may vary and so on. The skill being acquired is not the execution of a single maneuver, but a range of responses that always includes a degree of uncertainty on which exact response would be most effective.  Though some players may find the need to pound the controller ever faster to be exhilarating, undoubtedly there are some that may find this tedious.  With the introduction of noise, gameplay can be made more interesting and challenging independent of the precision and speed of a single, specific maneuver.  Even beginners tend to enjoy ping-pong.

The challenge with introducing noise is to provide in-game information that players can use to begin to discern or model that noise.  A ping-pong player may learn that when he observes his opponent smack the ball with sharp flick of the wrist, the on-coming ball is likely to have a severe spin. A game needs to provide information players can use to model uncertainty; that is, what information can I use to predict likely outcomes, what Dan Cook called learning variables: things that have some correlation with the probability distribution the players are trying to figure out. Such variables can be highly artificial and contrived, such as the color of the dragon reflecting the mean number of stabs required to kill it.  However, the learning variables can be more naturalistic as well. For example, if you have a dragon that moves and takes on different body positions during the confrontation with the player, the probabilities associated with different aspects of a successful kill could vary with the dragons body position.  For example, when the dragon’s arm is extended from his body, the target region effective for striking could enlarge and/or shift.  Alternatively, the target region could shift with each subsequent blow.

Of course, these linkages do not require noise or probability.  If the dragon’s arm is extended, only one blow will kill him.  If his arm is at his side, it will require 3.  A simple rule that players can learn.  Proficiency then requires applying this rule ever faster and more efficiently.  When noise is introduced, however, it makes it more challenging to discern the rule. Say for example that for each blow landed on the dragon, whether the dragon dies or not is drawn from a probability distribution (eg., poisson).  With each successive blow, the probability changes; however, a blow delivered while the dragon’s arm is extended increases the probability of death on the subsequent blow to a greater extent than one delivered with the arm down.  The skill here, then, becomes how to strategically battle the dragon, not simply increased precision and skill of a single movement.  This allows players to learn and improve at the game independently of just doing a single movement ever faster.  It also introduces the possibility that a player who relies on speed may initially do better than a slower player.  However, as the demands of the game increase, the slower player may excel through having a better understanding of how to effectively engage the game, much as a slower moving tennis player may defeat a quicker opponent by virtue of having greater control over the ball.

Daniel Cook’s discussion of introducing randomness into games of skill is not only a provocative idea, but prompts reflection on what we mean by ‘skill’ in the first place.   Insofar as ‘skill’ is thought of as doing something fast and precise, noise degrades skilled gameplay. However, if ‘skill’ is a matter of doing the right thing at the right time in a field of uncertainty, perhaps together with doing that faster and more efficiently, then introducing noise into games of skill may actually enhance the degree to which such games are, actually, games of skill instead of games of speed and motor repetition.

One response »

  1. froztwolf says:

    A measured amount of randomness can also make the game more engaging. If you know precisely what the dragon’s stats are and what the result of every attack is going to be, you just need to routinely execute attacks until it’s dead.

    If there’s a degree of randomness in how much damage you do, and how the dragon reacts, you need to watch the game closely for the feedback generated by your actions, and adjust your tactics on the fly. You need to give the game a lot more attention.

    Good Starcraft players can fight multiple fights at the same time, because they are deterministic and not every fight needs to be watched. Warcraft III had randomness in their combat, so each battle required the player’s full attention.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s