Fairness and the Public Good in Social Networks with Strategic Actors

点击翻译按钮获取中文摘要

Recent years saw a surge of development of artificial intelligence (AI) algorithms for decision making. These increasingly sophisticated algorithms make use of the ever-growing stores of information to make increasingly high stakes decisions. With this shift comes a pressing need to ensure that the decisions of these algorithms are fair and equitable to all parties involved, and are directed toward the public greater good. There are many ways to interpret "fairness" and "public good". This proposal focuses on fairness notions from multiagent systems and game theory. Multiagent systems is the study of communities of agents each acting independently, often according to its own interests. Game theory is the mathematics of strategic decision-making in the presence of other strategic agents. Taken together, these fields provide a rich model of the physical and digital spaces in which AI algorithms operate. They also provide fairness notions that characterize how beneficial or equitable algorithms are - for instance, an algorithm may be "truthful" where agents have no incentives to provide false reports; or, the outcome may be at a "Nash equilibrium" where agents are reasonably satisfied that they would not want to alter their actions. We focus on a particular domain for the research and application of these fairness notions - social networks. On a societal level, social networks (in the form of social media platforms) touch many aspects of our lives. They gather vast amounts of information about us and use this information to drive AI algorithms that may or may not be acting in the interest of the public good. They also control the propagation of information in digital spaces, and therefore are in a position to mediate both useful messages as well as harmful disinformation. This proposal examines how we can best leverage AI algorithms in a social network environment toward the public good, toward achieving fair and equitable outcomes for all participants.