Modeling trustworthy behavior and limiting the impact of selfishness

The major theme of the research in this dissertation is the modeling of selfish behavior and the mitigation of its effects. Game theory literature asserts that all agents behave with complete self-interest. However, this is at odds with empirical studies in behavioral economics which routinely show subjects engaging in behaviors which allow them to be taken advantage of by other agents. Despite this, the other agents rarely do so. In order to predict when and to what degree agents engage in self-serving actions, we introduce the concept of a Limited-Trust Equilibrium (LTE), a state in which all agents contribute to each other's utility, provided it is not too expensive for them personally. Each agent is motivated to do so in order to inspire reciprocity from its fellow agents and thus benefit in the long term. The LTE is then shown to exist in all finite games, and the utility of agents who play in a limited-trust manner is compared theoretically and numerically to those who play in a purely self-serving manner to illustrate why the agents prefer to interact in this way. The concept of limited trust is then applied to a social setting, in which players need to attract and form partnerships in a social network. This induces a metagame in which players must decide how much they are willing to commit to reciprocity in order to attract partners, where players who behave in a less selfish manner are naturally more attractive partners, but more selfish players benefit more per partnership formed. When other factors are not kept equal, such as when not all players are able to provide the same opportunities to their potential partners, we see the emergence of ``diva" behavior, in which talented or well-connected players are easily able to form partnerships despite behaving in a mostly or entirely selfish manner. A paper based on this work is nearing its conclusion and is expected to be submitted prior to Final Defense. As initially mentioned, our research also touches on the mitigation of the effects of selfish behavior. A major focus of research in Game Theory is on designing games in which the interests of the players align with the interest of the game's administrator or coordinator, generally maximizing the net utility or minimizing the net cost of the system the game operates in. Therefore, following our work on the LTE to better model when and how selfish behavior occurs, we pivot to focus on this area. We introduce the Prize Collecting Multiagent Orienteering Problem (PCMOP), a Game Theoretic version of the Orienteering Problem with applications to ride-sharing. We show it to be part of the class of valid utility games, then propose and analyze three policies for mitigating selfish behavior in the PCMOP. Two of these policies are broadly applicable to the class of valid utility games while the third is similarly applicable to valid utility games in extensive form