« Back to Stories WID Series: WID Series Archives: App Awe

App Awe: Twitch and the Infinite Player Theorem

Can chaos lead to "intelligent" behavior in video games? To find out, our App Awe editorial inspects the popular Internet experiment "Twitch Plays Pokémon."

Metal symbols photo from Thinkstock

App Awe is a WID essay series exploring the transformative potential at the intersection of gaming, science, education and society.

By Dennis Paiz-Ramirez, Project Assistant in Games+Learning+Society research group at WID. Follow him on Twitter.Dennis Paiz-Ramirez

Given enough time, 1,000 monkeys sitting at 1,000 typewriters will produce the complete works of Shakespeare. Or so goes a permutation of the infinite monkey theorem, which states that given enough time, randomly generated sequences of characters will result in a reproduction of Shakespeare’s seminal plays. While the popular saying itself is a metaphor, it seems to resonate with people given its constant citation in popular culture. While the probability of this event actually occurring is exceedingly low, it is not zero. Under a similar premise, the experiment “Twitch Plays Pokémon” set out to see if order can arise from chaos. Can 1,000 gamers at 1,000 computers collectively beat the game Pokémon? The answer seems to be, “Yes.”

Hard to control pokemon!

Credit reddit user JediJake

Twitch Plays Pokémon (TPP) made its debut on Twitch.tv (a website for streaming videogame playing) on Feb. 12, 2014. In a normal videogame stream, one player broadcasts himself/herself playing a game while spectators discuss the game and talk amongst themselves in a chatbox. Touting itself as a social experiment, TPP disrupted this format by allowing viewers to influence the game’s progress by inputting commands via chat. The system itself is simple. By parsing commands gathered in a user chat window, an IRC (Internet relay chat) bot sends commands to an emulated version of the game Pokémon. The game then executes the commands sequentially. Since its release, TPP has gathered 40 million views with upwards of 100,000 people participating at the same time.

Pokémon Red, the game used in this experiment, was released in 1996 and was the first installment of what has been noted as the most successful role-playing game (RPG) series of all time. In the game, the player is tasked with documenting, capturing, training and battling these creatures. The goal of the game is familiar to many: collect all 151 species of Pokémon (filling up an in-game encyclopedia), or become the strongest trainer in the region by defeating the regional champion in a duel. The social experiment TPP chose the latter as a goal.

As expected, parsing hundreds of commands per second results in what looks like chaos. The main character is seen spinning in circles, attempting to use items at inappropriate times, and generally doing things no competent player would ever do (such as naming a member of their team “ABBBBBBK”). In spite of this, there seems to be just enough consensus in the mind hive to move ahead. A week into the project, TPP had collected four badges, meaning contributors had completed about 50 percent of the game.

I am impressed you got here!

Credit Reddit user IAmARobot

But this is not the first time that a real life permutation of the infinite monkey problem has been attempted. In 2003, lecturers and students at the University of Plymouth placed a computer in front of Sulawesi crested macaques to see what would happen. The result was five pages of characters, mostly of the letter “s.” During the course of the experiment, the dominant male ruined the computer by bashing it with a stone. To add insult to injury, the other macaques then urinated and defecated on the computer. Needless to say, TPP seems to have a better track record of moving towards its intended goal.

Perhaps the most interesting thing about TPP is the community that has risen around it. Within days, TPP had created its own subreddit, where players converse about strategy or significant events that happened in-game. Interestingly enough, players took it upon themselves to create a narrative around the seemingly random events resulting in a collective history, and, more interestingly, a collective mythology. The Helix Fossil, a significant item in the game, cannot be used in normal circumstances but, due to the chaos, it was selected several times during gameplay to no avail. This seemingly meaningless act was interpreted by the community to hold some significance to the main character. Soon the act was interpreted as religious one, leading to the mantra “All hail Helix.” While entertaining, this act of attributing some otherworldly explanation to an unusual event has been repeated throughout our history from the Witches of Salem, to the Cargo Cults of Fiji. The narrative built around the random occurrences has become a great source of entertainment for the community and often influences its actions.

What had started as a representation of the infinite monkey theorem resulted in, what some players jokingly referred to as, a politics simulator.

On the sixth day of the experiment, the system introduced a Democracy system in order to get past a movement-based puzzle that had been holding up progress for almost 24 hours. This new system gathered votes for 5 seconds before executing the most voted for choice. In addition to voting for an action, the new system also made it possible to vote how many times that action was used. For example, “a2” would execute the “a” command twice. While this system allowed the collective to solve the puzzle in a (relatively) short amount of time, a majority of the players felt that the new spirit was contrary to the initial goal of the experiment. To their credit, the puzzle that had been impeding progress had been solved twice without the Democracy system, but due to chaos, was also restarted twice. The outrage expressed by the chat filling with “RIOT” lead to amendment of the system so that, in addition to issuing a command, players could also vote for either Democracy, or Anarchy.

This compromise did not please the collective which showed its disapproval in the form of issuing the command “start9” as a way to stop progress while in Democracy mode. In other words, TPP introduced Democracy while players introduced the filibuster. What had started as a representation of the infinite monkey theorem resulted in, what some players jokingly referred to as, a politics simulator. Those in favor of the Democracy noted that it had taken 18 hours to overcome a ledge (something which a player would spend at most a minute on). Those in favor of the classic mode of anarchy noted that Democracy slowed down the game, was not in the spirit of the original problem, and had made the random mistakes, which had been celebrated by the community, almost nonexistent. This divide still exists and has also been worked into the narrative of the community. However, since the ability to switch modes was introduced, the majority of the time has been spent in anarchy.

I will maintain anarchy

Credit reddit user JohnMarkParker

Since TPP appeared, other streams with similar goals have been created. One of these was a game of Pokémon run by a pseudo-random number generator (RNG). It’s interesting to compare the progress made by RNG to the original TPP. While both games managed to make it a fourth of the way through the game (having earned two badges) the RNG experiment ended in a fail-state where the protagonist lost all his money and released all his Pokémon — save a Magikarp — and thus was unable to continue. While impressive, the failure of RNG suggests that in spite of the interactions one might see during a minute of chat, there is more going on in TPP than chaos. Because of the community, the newly introduced Democracy system, and the collective bias towards completing the game, TPP cannot be seen as a purely random system and probably won’t give us much insight into the infinite monkey theorem, but it is a good example of emergent behavior.

Emergence is the idea that individuals following simple rules give rise to complex systems. One of the more famous examples of this can be found in agent-based models of flocking. Developed by Craig Reynolds in 1986, the artificial life program BOIDS successfully replicated the movement patterns of birds flocking by making each individual “boid” (an artificial bird) follow simple commands. These commands included how far they should be from one another and aligning towards the direction of the average flock member. In much the same way, individual participants of the TPP experiment can be seen as individual agents abiding by the same rules in order to achieve a more complex goal (progressing through the game). Individuals assign one of eight commands based on the task at hand, correct themselves if they overshoot and slowly make progress. Even if some players are there to specifically impede progress, other individuals react to their input resulting in a course correction and net positive outcome.

Three hundred and ninety hours into the game, the community of TPP beat Pokémon Red. You would think that after such an ordeal people may be hesitant to jump back in again. However, on March 2, TPP resumed playing the sequel to Pokémon Red, Pokémon Crystal. This suggests that we’ll be seeing plenty more of these community playing events, and while it’s not certain that TPP will succeed a second time, it sure is interesting to watch.


Press Contact:

WID Media
608-316-4325

More articles in App Awe:

Key image from iStock

App Awe: 2048 Variations on a Theme

How acceptable is copying in order to create new video games? One WID and video game expert weighs in through the Institute’s App Awe editorial series.

Sky and clouds image from Thinkstock

App Awe: LittleBigPlanet

Are video games tools in disguise for players to share their stories and perspectives of the world? That’s the idea behind this App Awe editorial from the Institute’s video games group.