UC Santa Barbara

Due to the emergence of new communication and computation technologies, many existing systems and infrastructures are experiencing revolutions in their behavior and capabilities. For example, with the development of self-driving vehicles (which possess the power to automate driving decisions and coordinate with other automated vehicles on the road), new traffic patterns take place, and the opportunity to introduce safer and more efficient driving behavior presents itself. Similarly, with the proliferation of large-language AI models and access to social media, information can be garnered and exchanged in new (sometimes unreliable) ways. These systems, among many others, consist of many human and engineered entities (or agents) interacting and making decisions with the limited information they possess about one another and the environment they are in; we will refer to these systems as multi-agent systems. The main goal of this thesis is to introduce new understandings of the role information plays in several aspects of multi-agent systems, i.e., the information a system designer possesses about the agents, the agents’ knowledge of the environment, and the agents’ ability to share information and coordinate behavior among themselves. The contributions are split into two parts: the first studies the interactions of designed decision-makers in distributed autonomous systems, and the second studies behavior in social systems in which decision-makers are human users. In each setting, we can model the interactions between agents (human or engineered) through the mathematical framework of game theory. The findings of this work reveal several insights on the role of information in multi-agent systems, including 1) showing to what extent greater information on human-agent preferences and engineered agent capabilities can aid in the design phase, 2) proving that revealing information to agents (human or designed) can worsen system performance unless done carefully, and 3) quantifying the benefits and costs incurred by increasing the level of communication and collaboration among agents. These insights will be shown through rigorous analysis of several game-theoretic models.