Twitch Plays Pokemon and (accidentally) Popularizes a New Genre

This has been crossposted with my studio's website here.

Although it has only been going on for just over two weeks now, the popularity of the “Twitch Plays Pokemon” stream virtually assures that everyone in this audience has heard of it by now. For the uninitiated, “Twitch Plays Pokemon” is an ongoing stream on Twitch.tv that allows players to directly play (a ROM hack of) Pokemon Red/Blue on a GameBoy emulator by typing button commands into the attached chat. The twist here is that everyone in the chat is playing the same instance of the game. Compound the fact that there are, on average, 50,000 people playing at once with competing interests in the game (some want to go this way, others want to go that way, a few want to cause problems) with the fact that, based on how Twitch chat works player inputs in the chat are at about a 20 second delay. One only needs to watch for a few seconds to confirm the immediate suspicion that this works out to be an absolute mess (although it does seem to fair better than a random number generator on a different stream).

Now, a common joke is to suggest that Twitch play other games which are sure to result in disaster. Probably the most prominent example of this is the suggestion of “Twitch Plays Demon's Souls”. Clearly this is a bad idea if the stream ever hopes to actually finish the game based on the relative difficulty of Demon's Souls and the relative ease of Pokemon Red/Blue. However, Pokemon Red/Blue has a number of design features, other than difficulty, that make it suited towards being played in a Twitch stream like environment. In this article, we shall examine some of these game elements and examine how this unassuming joke Twitch stream can inform a just budding genre which I will call a “Massively Cooperative Game”. (Why not “Massively Cooperative Online Game” like the popular MMOG genre? There is no reason to assume this genre has to be online only as it could be played, for example, in a movie theater or at a convention.)

Before getting too deep into individual mechanics, I think it is important to point out a distinction in the way we commonly use language surrounding probabilistic events like these. Mathematicians and statisticians will generally use the words “random” and “stochastic” more or less interchangeably to talk about abstract objects that are analyzed with probability or other non-deterministic methods with “random” generally referring to variables and “stochastic” generally referring to processes. However, the more common use of “random” carries a certain connotation of something that appears completely without order or logic or is otherwise evenly distributed. (There is no non-jargon use of “stochastic” in my experience.) As such, I might suggest that “Twitch Plays Pokemon” is “random” but the immediate objection is that is can't be because it is all determined by player input who are biased toward their own goals. To avoid this confusion, I will use the term “stochastic” to describe any process that can be analyzed using these probabilistic methods. Specifically, we can best describe progress through “Twitch Plays Pokemon” as a stochastic process.

As a stochastic process, “Twitch Plays Pokemon” will experience a number of very improbable events that no usual Pokemon Red/Blue player would experience (for example, they accidentally released their starting Charmander). However, in Pokemon most of these random events will either help progress the game in some way or be completely neutral. Very few will be actively regressive (such as the a fore mentioned releasing of Pokemon). For example, if the players wander through tall grass or a cave they will experience a number of random encounters. Fighting these random encounters (battles are a stochastic process as people will want to try to use different moves or different Pokemon) will generally result in either a win, rewarding experience points(EXP), or a loss, costing money. Since the EXP gains are permanent on the player's Pokemon and Pokemon is designed to allow an inexperienced player to progress by leveling up their Pokemon. These small gains are allowed to compound until the players can progress through certain story aspects of the game. (In fact, it has led to the community nick name of “Bird Jesus” who quite often saved the day as a result of its significantly higher level.)

By comparison, Demon's Souls punishes death in a completely different way. Specifically death may end up costing the player hard won experience as well as progress through a dangerous area. As a result, small gains made by the stochastic system may be easily lost and not allowed to compound. Thus the stochastic system guided by the players will likely act for hours or even days without any notable progress.

This effect is further compounded in each game as deaths in Demon's Souls respawn monsters and reset obstacles where as trainers who are beaten in “Pokemon Red/Blue” become harmless NPCs for the remainder of the game with a scant few notable exceptions.

Another major difference is that Pokemon Red/Blue features a very discrete tile based movement system that requires no buttons to be held as opposed Demon's Souls more continuous movement system. As a result the player character in Pokemon Red/Blue will journey some distance even in a completely random environment where as in Demon's Souls it will result in small twitches in different directions but probably not getting too far from the starting location.

To understand why this is, consider this game: Two players have some coins. Each round they flip a coin each. If their coin lands heads they add points, but if it lands tails they lose points. Player 1 gains or loses 1 point with each flip and Player 2 gains or loses 2 points with each flip. The winner is the first player to get to 4 points. Who is more likely to win? Clearly Player 2 is more likely to win since his game has greater variation and thus he needs to flip heads only two more times than he flips tails (as opposed to Player 1 who needs to flip heads four more times than he flips tails). The discreteness of “Pokemon Red/Blue” ensures the greater variability over “Demon's Souls” and thus a wider variety of experiences for the players.

Despite these differences, however, both “Twitch Plays Pokemon” and “Twitch Plays Demon's Souls” have a quality that is beneficial to the Massive Cooperative Game format: they are both built upon games that are very well studied by gamers at large. As a result, players can form long term plans and build teams that support that plan much more easily. There is no need or drive to explore the world in “Twitch Plays Pokemon” and so the player base doesn't become too divided in how they want to act. Some divisions will occur, and that is expected, but there will never be the case where too many players want to visit areas of the game that don't have much content. A secondary benefit to this is that it allows for a sort of planning metagame where players can figure out creative solutions to difficulties they know are coming later on. Specifically in the “Twitch Plays Pokemon” case many players were concerned that the Master Ball would be wasted very quickly after they got it and so created plans to use it towards accomplishing something before it is wasted, all plans that would never occurred to people playing Pokemon Red/Blue.

Finally, we will examine one way in which Demon's Souls is slightly superior to Pokemon Red/Blue in this environment. In the later, players must transverse an area known as the Safari Zone. There, they are required to navigate a maze with no more than a set number of missteps. Due to the stochastic nature of the control scheme of “Twitch Plays Pokemon” the probability of this happening was rather low. Ordinarily this would not be a problem as an event with non-zero probability will happen almost certainly on a sufficiently long time. However, the number of tries the players have at the Safari Zone is limited by the amount of money they have which, in turn, in limited by the number of trainers they fight. As a result, it is possible (in fact, likely) that the game will enter a “walking dead” state where the game cannot be brought to any kind of win state but it does not recognize itself as being unwinnable. (Walking dead states were the bane of many early point and click adventures and are generally considered bad design in most cases, especially if the player is given no warning or ability to avoid it other than knowing the future of the game.) Demon's Souls, by comparison, does not have a walking dead state.

Clearly from the above discussion the design of a game is nontrivial in this sort of setting, but where might these settings arise and what, more specific, lessons can we take from this silly idea that has taken the internet in the last week?

In existing games, especially Massively Multiplayer Online and Free-to-Play games, there is often either an implicit or explicit community goal which often requires the teamwork of dozens (or more) players. The raid content of World of Warcraft is an example of this community goal. Often, this raid content is reserved for highly dedicated players in strictly organized and regimented guilds. A problem we might be able to answer, using design lessons learned above, is “How do we extend raid type content to more casual players and/or players who might not be able to schedule their time around guild raids?” With an understanding of what happened here in “Twitch Plays Pokemon” and a requisite understanding of stochastic processes and probability, designers of World of Warcraft or Everquest 2 could build raid like events that allow for easy jump-in/jump-out styles of play.

Alternatively we may opt to create entirely new games with the express purpose that they be played by a large community simultaneously, a game type I earlier called a Massively Cooperative Game. This make take the form of cinema type game like Renga (shown at IndieCade East) or an Augmented Reality Game like the introduction for Portal 2. However, in both example cases game play is single dimensional and rather linear: the whole world progresses when someone figures out how to decode a clue and there is exactly one right answer for each clue. It lacks the robustness of experience and choice that we might expect from many games.

With the above comparison between “Twitch Plays Pokemon” and “Twitch Plays Demon's Souls”, let's identify the elements required from a Massively Cooperative Game:

First, the control stochastic process should experience a large variance. This ensures that the game will be interesting and various player groups feel a certain level of agency within the game. This can be accomplished most readily by minimizing the number of consecutive button presses required, from any individual state, to cause an in game event.

Second, the game progress should be more or less permanent with very little ability to regress. This can be accomplished by requiring the button presses to regress slightly more complicated than the button presses required to progress. For example, assume the game is in a given state. From that state, progress can be made by pressing a sequence of 3 buttons, and regression would be the result of 4 different buttons (with any other button combination being largely neutral). From this state, progress will be made more often than regression even in a purely random state. In the directed stochastic process, progress will be made even more often, with the comparatively rare run of bad luck.

Third, walking dead states need to be eliminated entirely either by always allowing progression back towards the major goals or by explicitly stating that the game has been lost. This, of all the points here, is the most extensively talked upon condition, especially within the adventure game community. For the most part, walking dead states can be avoided by disallowing overwriting a periodic automatic save feature (perhaps when sufficient progress has been made) and through sufficient beta testing with testers being specifically instructed to try to get stuck or otherwise break the intended sequence of your game.

Finally, thanks to the success of “Twitch Plays Pokemon”, the community has felt the need to supplement the game with a number of features which any designed Massively Cooperative Game should have standard. They are: a separation of controls and communications; the ability to explore the game independently so informed decisions can be made in the game proper; the ability to share, organize or otherwise create communities around certain game features; the ability to independently see statistics about the current game state (game map, money, etc.) without disrupting the experience for other players.

Thanks to the popularity of “Twitch Plays Pokemon” in the coming months, we will certainly see a number of imitations. Many of these, of course, will simply be Twitch streams of various existing games using the bot's available source code. However, some game dev groups have already expressed interest in creating games with viewer participation and so will have to learn the lessons of “Twitch Plays Pokemon” as they start to build into this new world of Massively Cooperative Games.