查看文章

We study an interesting 2-player game known as Iterated Traveler’s Dilemma. The Traveler’s Dilemma (TD) game is a non-zero sum game in which each player has a large number of possible actions or moves. In the iterated context, this means many possible actions in each round and therefore an astronomic number of possible strategies overall. What makes Iterated TD particularly interesting, is that its structure defies the usual prescriptions of classical game theory insofar as what constitutes an “optimal” or even just “good” strategy. In particular, TD has a single Nash equilibrium (which is also the game’s only evolutionarily stable equilibrium), yet that equilibrium corresponds to a very low payoff for each individual player and essentially minimizes social welfare. We study possible ways of “playing well” and seeking strategies that are good for individual players, as well as for strategy pairs that would tend to maximize, not minimize, social welfare. In that context, we propose a number of possible strategies for ITD, from some trivial and rather “dumb” ones, to generalizations of “tit-for-tat”(well-known from extensive studies of ITD’s simpler, but much more famous, cousin–(iterated) prisoners’ dilemma), to some relatively sophisticated strategies where an agent tries to non-trivially model the behavior of the other agent in order to respond (closer to) optimally in the future rounds. We perform a thorough comparison-and-contrast of 36 different strategies overall via a round-robin, everyone-against-everyone tournament in the spirit of well-known work by Axelrod [?] in the context of prisoner’s dilemma. We motivate the choices of “players”(ie, strategies …