Lessons in decentralised governance from Twitch Plays Pokemon
For 16.5 days in February of 2014, an anonymous Aussie allowed 80,000 people to play a single game of Pokémon at the same time. This particular burst of internet-era tomfoolery was predictably chaotic, yet a level of signal emerged from the noise and provides compelling, applicable lessons about the nature of mass participation.
This article proposes that Twitch Plays Pokémon (herein often referred to as TPP) is an excellent case study for the popular field of decentralised governance, and further, is a prime testbed for financially incentivising desirable outcomes of self organisation.
Why did this one go viral?
TPP combined two generational phenomenons: Twitch Chat, and one of the original Pokémon games.
Twitch
Twitch is a video streaming site. The most common form of content streamed through Twitch is video gaming. A player becomes a “streamer” when they live stream their gaming sessions to Twitch. Often this is accompanied by commentary and/or a small video feed of the player themselves, so the viewers can watch their reactions.
Alongside the video, viewers are able to participate in a chatroom. It’s interesting to note that these actively-communicating peers are now colloquially referred to as a singular noun: Chat.
On streams with few active viewers, Chat is often reasonably docile and sedated, with participants often interacting directly with the streamer and each other, asking the same kinds of questions you would ask a friend over an instant messenger.
Very active streams change the nature of Chat. Popular streamers like Ninja or Arteezy can receive tens of thousands of viewers at one time, and yearly streamed events like DOTA2’s International or the semiannual finish-lots-of-games-as-fast-as-we-can-athon Games Done Quick (GDQ) can receive viewcounts in the hundreds of thousands and sometimes tip the 1million mark.
Predictably, Chat becomes an incredibly unwieldy tool once viewer numbers cross a certain fuzzy threshold. No longer is it possible to maintain a conversation with a friend, but the velocity and ferocity of Chat becomes a force to be reckoned with. “Reading” Chat during an event like the DOTA2 International is next to impossible. It becomes a furious stream of mass consciousness wherein copy-and-paste messages which fit a particular “memetic template” have the most visual cut-through and subsequently gain the most memetic traction. Chat’s reactions to events on-screen are often colloquialised into a semi-specific memetic format before being posted, leading to a lot of similarity in the types of messages that one will see when they “read Chat”.
It is almost the perfect place to be conducting real-time sentiment analysis: truckloads of data pushed through a meme-shaped spaghetti mould.
Pokémon
Pokémon (a portmanteaux of Pocket Monsters) is a Japanese game released for the Game Boy all the way back in 1996 (1998 for America and the rest of the world). It quickly became a sensation. Pokémon is an open-world collect-em-all Role Playing Game. Basically you’re running around finding cute Tamagotchi-like creatures that you command to fight other people’s Tamagotchi creatures (as well as some wild ones) so that they get stronger and more powerful. Along the way are a series of puzzles and challenges, but the main draw of the game is encountering and collecting Pokémon, then making them as strong as possible to win battles.
You control a character moving about this open world, encountering wild Pokémon and attempting to catch some of them, as well as encountering other AI-controlled characters whose Pokémon you must defeat to progress in the game. Battles require you to choose a Pokémon to represent you and select the actions that the little monster performs in order to defeat the other player’s little monster. If you lose a battle you warp back to a checkpoint and lose (basically) no progress.
The game ends when the player completes a sequence of in-game battles known as the “Elite Four” (though there are five battles in the sequence for any sticklers out there). This requires a lot of in-game activity to train the player’s collected monsters, Pokémon is primarily a game of stats and choices, not a game of quick reactions and bravado.
Important to note is that Pokémon has a very limited set of inputs. A player can press one of the following 8 buttons at any time during the game:
- Up | Down | Left | Right (movement)
- Start | Select (open menus)
- A | B (in-game selections + actions)
Again important to note is that it is perfectly valid for the controlled character to be walking into a wall for hours at a time, the game has no time limit and doesn’t care how long you take to make a choice in a menu.
Twitch × Pokémon = Social Phenomenon²
On February 12th 2014, Twitch Plays Pokemon began streaming on Twitch. The stream was live for 24 hours a day, 7 days a week.
The aforementioned anonymous Australian allowed people to participate by creating a bot which scraped the Twitch Chat for messages. Participants were able to type which input they wanted the bot to make and it would then perform the action in game. If you want to put on your “made up phrases” hat, you could almost think of the in-game player as a Decentralised Autonomous Actor.
To begin with, anarchy ruled. Every single action typed in chat was performed by the bot, allowing for permanent freneticism and making for some near-impossible puzzles.
After four days, anarchic decisions had progressed the game to a puzzle scenario which seemed as though it would benefit from some deeper cooperation from all the “players” to complete. The developer, noticing the struggle, introduced democracy to the decision making process and inputs now went through a 30 second voting period before they were then finalised and sent to the game, which seemingly made coordination easier.
After a period of predictable Twitch Chat outrage, the dev added the ability to flip back and forth between either the initial “Anarchy” mode, which required a majority vote, and the later “Democracy” mode, which required a supermajority.
Stats
The game was completed after roughly 16 days of constant 24-hour playtime. Average concurrent viewership was 80,000(!) with at least 10% participation at any one time. Twitch estimates that roughly 1.16 million people participated, with a total of 55 million views throughout the period.
The game was completed. Anti-completionist actors were drowned out on average and the common known goal was sufficiently more attractive than permanent chaos. There are many more reasons for the game’s completion I’m sure, not least the fact that the “forward path” (progressing within the game of Pokémon) is much easier to collaborate on than a path which is anything other than forward, due to knowledge of how to progress being organically distributed amongst multiple generations over the last 20 years.
Self Organising
One of the most interesting things about TPP is the lack of ability for participants to understand what other actors are going to do. This became a little easier when the democracy decision making mode was active, but was still not rich enough to empower meaningful discussions about how to proceed. Off-channel communication was still necessary. Communities appeared on 4chan and Reddit amongst other sites, with players congregating to share memes inspired by bizarre in-game occurrences. Crucially, these platforms also drove participation in the game, establishing tribalistic camps of those who were for “Anarchy” or “Democracy”, and vaguely strategising about what to do next. Notably, these strategies would often not traditionally be considered “progress”.
Why Should We Take Note When Thinking About Designing Decentralised Governance?
This rapid organic self-assembly is often cited by those who are proponents of decentralised governance models as a necessary functional component of the system. “Proposal” models are dominant within the Web3 community as an effective method of surfacing items which may require decision processes (such as distribution of an organisation’s funds).
At this stage I should add that “governance” is a very loaded term within the Web3 world and I personally think it’s a very misunderstood one. I am purely referring to “governance” in this article as the way of proposing and subsequently making decisions in group contexts. The enactment of those decisions is not something I am concerned with, and neither is the conversation about “what is worth governing”.
In TPP, there was already a participatory framework in place for how decisions were proposed, decided upon, and enacted. It’s also worth being aware that the design of this framework was not conducted in a “decentralised” manner, it was pre-ordained by a single human. Whilst we’re not going to delve too deeply into these frameworks and how something equally un-opinionated could be replicated outside of this context, the fact that the entire governance framework (typing in Chat) and subsequent action framework (the bot which pushed the buttons) was itself prescribed and not subject to design by committee may be a large factor in why Twitch Chat managed to successfully complete Pokémon: they didn’t have to design the rules of engagement whilst they lacked a way to enforce them, which is what I’m seeing a lot of in the Web3 world.
Twitch Plays Pokémon is Ripe for Experimenting with Incentivisation Models
Twitch Plays Pokémon harvested no benefit for participants other than the ephemeral feeling of being part of something novel. TPP was almost primarily an art project, a labour of love and experimentation by an anonymous actor with no goal other than to allow others to create unimaginable scenarios.
To the delight of the scientist in me, open source Pokémon clones now exist, opening the door for modification of source code and the introduction of side effects. I propose that the inclusion of financial reward (maybe through cryptocurrency(!)) would be a worthwhile social experiment and could inform iterations upon economically incentivised decision frameworks.
There are simple experiments to conduct: beat the game and all participants are rewarded based on the proportion of actions that they contributed; reward participants at specific checkpoints within the game; et cetera. There are also incredibly complex scenarios which could be incentivised: go here, get this item, drop it in this particular space in the game world, et cetera. The possibilities are reasonably endless and almost verging on trivial to implement once you vaguely understand the source code. There’s so much greenfield here for experimenting with how incentives change people’s ability to coordinate and achieve a common goal within the finite decision surface of something like a Pokémon game.
Finite Decision Surfaces Also Exist In Web3
Pokémon, like the current crop of software systems which are governed in a decentralised way, has relatively few possible inputs at any one time. In this case it’s 8 buttons. If you take a look at MakerDAO’s Single Collateral DAI, one of the more successful decentralised projects deployed right now, there are roughly 12 (if I’ve interpreted the source material correctly) variables which can be updated by token holders in a plutocratic/meritocratic/technocratic/economocratic/democratic system of governance depending on what you believe.
Let’s Experiment!
Flex Dapps are heading to EDCON in Sydney next week, and are specifically going to participate in the Plasma/State channels hackathon as a team. We’re planning to build a version of TPP which utilises a Plasma chain for near-instant voting on game inputs. We don’t think that EDCON will harbour enough people or time to complete the entire game, so our intention is to allow people to participate from a specific save state, possibly the start of the “Elite Four” sequence towards the end of the game, which will start over once it’s finished so we can collect data across the entire week (if people play, that is).
We’re not sure how far we’ll get, but we think this is a fun hackathon project and an excellent opportunity to throw our hat in the research ring by doing what we’re best at: building great software. We’d love your help, so if you’re going to be at EDCON or you can provide some insight into how to build this (such as if you’ve mucked around with open source Pokémon clones before), we’d love you to get in touch. Hit up tom@flexdapps.com or @tomnash on Telegram.
Trigger Warning: Opinions Below
There’s a lot of theorising and armchair economising in the Web3 community, not least amongst projects that claim to shake up “governance” by creating some impeccable game theoretic model that is impermeable to baddies by design. I’m certainly (and happily) not a game theorist or an economist, but I have to wonder where the data to back up these claims or theoretical models comes from. Experimentation is the only way I know of to collect reliable data, and as far as I understand there aren’t many governance experiments which have yielded useful intelligence about what to do through application (though there are several useful examples about what not to do, thanks block.one).
As an aside, I think it’s fascinating that decentralised financial products which fall under the wider umbrella of DeFi are giving us all data about how financial instruments are used. Never before has the entire world had a porthole into how much I value the collateral I use to secure a loan, or how much I’m willing to bet on the fluctuations of its value, like is now possible by watching the Maker CDP contracts, for example. But there are more: Compound; Dharma; Dy/Dx. Financial products modelled on the “real world” are now creating public datasets, and if there’s anything the last decade should have taught us all, it’s that data rules and centralisation drools.
