Yao Fan

According to Austen-Smith & Banks (1996, A-S&B hereafter), under the setup of the Condorcet Jury Theorem (the CJT hereafter), sincere voting is not fully rational in the presence of other voters. A sincere voter ignores the fact that her vote will only make a difference when she is pivotal. A fully rational voter, having this fact in view, should vote as if she were pivotal. Two related yet distinct arguments can be made based on this observation: one against the rationality of sincere voting and the other against its empirical adequacy. In this paper, I provide a two-part defense for sincere voting. The first part of my defense responds to the two arguments derived from A-S&B’s observation. I show how both arguments crucially rely on the use of Nash equilibrium as the solution concept and then reject the applicability of Nash equilibrium to the CJT based on certain features of large elections. The second part of my defense proposes to use an alternative solution concept, focal point, for understanding voter behaviors in the CJT. The focal-point analysis will highlight several attractive features of sincere voting and support it on both empirical and rational grounds.

This paper discusses the phenomenon of strategic track-record building in the context of Expert Identification. I present three mathematical models to show (1) track records do improve Novice’s epistemic performance as Goldman (2001, 2021) suggested, (2) Expert can influence Novice’s decision by strategically building track records even when Novice has full information about Expert’s track record, and (3) these influences can only be overcome by regulations at the institutional level. This enriches the existing literature on Expert Identification in two ways. First, it fills a gap in the existing literature on the strategic aspects of track-record building, despite its ubiquitous presence. Second, as the identified caveat can only be overcome via institutional arrangements, it shows that the Expert Identification problem is a problem of so-called “system-oriented social epistemology” even in the one-Expert-one-Novice setting, contra Goldman (2012), who treats it as a paradigmatic example of the study of a single doxastic agent with social evidence.

This thesis is a collection of three papers in social epistemology. Three epistemic systems of great importance for modern society are studied respectively: the expert system, the democratic system, and the financial system. Throughout the three papers, I try to pay additional attention to the interactions between individuals and institutions, which explains the title of this dissertation. In the first paper, I discuss ways in which institutional remedies are crucial for individual epistemic performance. In the second paper, I argue that the individual moral requirements of citizens (i.e., participants of democracy) should not be overlooked in presence of the epistemic aggregation effect of democracy at the institutional level. In the third paper, I analyze how individual motives in financial markets and the epistemic functions of financial markets influence and interact with each other. The proper functioning of each of the three institutions, I believe, is necessary for achieving a good society in our time. In a good society, neither the individual aspects nor the institutional aspects should be overlooked.

In this paper, we propose a novel personalized decision support system that combines Theory of Mind (ToM) modeling and explainable Reinforcement Learning (XRL) to provide effective and interpretable interventions. Our method leverages DRL to provide expert action recommendations while incorporating ToM modeling to understand users’ mental states and predict their future actions, enabling appropriate timing for intervention. To explain interventions, we use counterfactual explanations based on RL’s feature importance and users’ ToM model structure. Our proposed system generates accurate and personalized interventions that are easily interpretable by end-users. We demonstrate the effectiveness of our approach through a series of crowd-sourcing experiments in a simulated team decision-making task, where our system outperforms control baselines in terms of task performance. Our proposed approach is agnostic to task environment and RL model structure, therefore has the potential to be generalized to a wide range of applications.