Deck-based prisoner's dilemma

Abstract

A deck-based game is a modification of a game that normally permits the players to use any number of moves of any type. This freedom of choice of moves is limited by handing each player a deck of cards, each of which with a single move printed on it. The player must then play from their deck rather than simply choosing the moves. This study documents that deck-based iterated prisoner's dilemma is radically different from standard prisoner's dilemma when the entire deck must be expended during play. The restrictions imposed by the deck change the game into a coordination game or an anti-coordination game. The game is shown to transform smoothly into standard prisoner's dilemma as the fraction of the deck used in play is reduced, assuming that a constant ratio of the two types of moves are used in the deck. The size of the deck, ratio of defects to cooperates, and evolutionary algorithm parameters are all studied using a string based representation. An adaptive agent representation is also developed, based on augmented finite state machines called deck automata. Deck automata evolve to play the game more effectively than the string based agents for three different situations; experiments in which agents expend all, three-quarters, or half the available cards.