Prisoner's Dilemma Internet Game

From WikiThor
Jump to: navigation, search

Inspired by this real-world test of the prisoner's dilemma, a large database of results of real people playing prisoner's dilemma style games would make a great test of Nash equilibrium theory.



The game needs to satisfy a number of conditions (besides scientific rigor, which goes without saying) for it to be a good source of the desired data.

  • It needs to have the structure of the prisoner's dilemma with as few addenda as possible. Complexities should be cosmetic only and not change the play of the game.
  • With the above point in mind, the game needs to be fun and engaging enough to make people want to play it. Actual monetary reward is infeasible.
  • The game should be disguised from being a scientific pursuit. As is well documented, internet trolls relish the chance to corrupt honest, transparent attempts at data-gathering.

Possible Game: Collusion

The premise is simple, yet realistic. Each player is a freelance corporate negotiator hired by a natural gas company for a quarter (game instance). Each game is a business quarter, and the player's choices are to bid high on available properties or bid low. The companies involved are huge, aggregate, mega-corps, and thus there are only two real bidders on each game's set of properties.

If both players choose to bid low (the eponymous decision, and the classic "Cooperation" game state), they take in equal assets for minimal cost, and development then yields their clients moderate profit.

If one company chooses to bid high and the other low (the first player "Defects"), that company gains all of the property. The costs are high, but the revenues are substantially higher, leading to high profit for the high bidder. The low bidder brings in no properties, yielding no revenue, and costing the company money (he still gets a paycheck, after all), resulting in negative profit.

If both agents choose to bid high (both "Defect"), again they bring in equal assets, but such a high cost means final quarterly profit for each business is low (but still positive).

Preplanned collusion is illegal and adequately enforced in the nation of business, and thus players cannot actively talk to each other. However, at the end of each round each player may leave a comment for the other to view at the end of the game.

Payoff Matrix

(A's Payoff, B's Payoff) B Bids High B Bids Low
A Bids High ($100k,$100k) ($300k,-$100k)
A Bids Low (-$100k,$300k) ($200k,$200k)


Player match-ups will be determined randomly. Game lengths (number of rounds) can be chosen from a set list (say 1, 2, 5, 10, 25, 50, 100), and players will be paired from others in their list. This should prevent actual collusion between players, as they should never be able to determine their opponent beforehand.

Each instance of the game will not be the focus for players. Instead, their profits from each game will add to their player statistics, for which they will be ranked. Players will be ranked at least by how much total profit they have brought their company (the sum of their results from all games) and how much they bring on average (sum divided by plays).

This coupled with a forum for discussions should be sufficient to mask the scientific purpose of the game.

Ficticious Companies

  • BonoboPhillips
  • Checkron Corporation
  • Coke Industries
  • Decathalon Oil
  • EndronMobile
  • Oriental Petroleum


To prevent people from making their own robots to play the game (which would skew the results), any player account can create up to a number of automated players (bots) and program them with various strategies. When playing, players can opt out (and do so by default) from playing bots.

Statistics involving robots is kept separately from statistics between human players.

Personal tools
other projects