Alice C.W. Huang

Suppose an epistemic agent has multiple strategies available to them. Among these, one strategy maximizes their expected accuracy, while another does not maximize their own expected accuracy, but instead maximizes the expected accuracy of the epistemic group they belong to. Call an agent that takes the latter strategy ``altruistic.” The role of epistemically altruistic agents has been highlighted in many results and phenomena studied in formal social epistemology, such as jury theorems, the wisdom of crowds, information cascades, opinion aggregation, and the exploration-exploitation tradeoff. But how do some agents come to be epistemically altruistic? In this paper, I demonstrate---using a self-assembling model with simple multi-agent reinforcement learning---how a group of agents can learn to adopt cooperative epistemic strategies through basic learning rules. Counterintuitively, I show that competitive incentives at the individual level lead to cooperative choices at the group level.

Suppose an epistemic agent has multiple strategies available to them. Among these, one strategy maximizes their expected accuracy, while another does not maximize their own expected accuracy, but instead maximizes the expected accuracy of the epistemic group they belong to. Call an agent that takes the latter strategy "altruistic." The role of epistemically altruistic agents has been highlighted in many results and phenomena studied in formal social epistemology, such as jury theorems, the wisdom of crowds, information cascades, opinion aggregation, and the exploration-exploitation tradeoff. But how do some agents come to be epistemically altruistic? In this paper, I demonstrate, using a self-assembling model with simple multi-agent reinforcement learning, how a group of agents can learn to adopt cooperative epistemic strategies through basic learning rules. Counterintuitively, I show that competitive incentives at the individual level lead to cooperative choices at the group level.

Two types of formal models - landscape search tasks and two-armed bandit models - are often used to study the effects that various social factors have on epistemic performance. I argue that they can be understood within a single framework. In this unified framework, I develop a model that may be used to understand the effects of functional and demographic diversity and their interaction. Using the unified model, I find that the benefit of demographic diversity is most pronounced in a functionally homogeneous group, and decreases with the increase of functional diversity.

There is a literature across science and technology studies, sociology of science, and philosophy examining the phenomenon of hype in science and technology. While definitions of hype vary, most accounts share a critical stance, viewing them as misleading or distorting scientific and public discourse. This paper takes a different approach by asking whether certain aspects of hyping might, under specific conditions, be beneficial. Rather than thinking about hype as isolated episodes, I think about hyping as a communicative strategy that affects decision-making under uncertainty. While individual instances of hype are harmful and often short-lived, as a practice, hyping might nevertheless be efficient in allocating resources. I draw an analogy between aggressively promoting the potential of emerging technologies despite significant uncertainty, and the Upper Confidence Bound (UCB) algorithm in reinforcement learning, a strategy shown to perform well in problems involving multiple uncertain choices. By drawing this analogy, I identify the conditions under which hyping can function as a collective implementation of UCB in a multi-agent, distributed context. To do so, I present a formal model that allows us to think about hyping as a mechanism for orchestrating public attention and helping shape the direction of scientific inquiry.

Many foundational concepts in machine learning have been criticized as inadequate. Philosophers have therefore taken it upon themselves to sort out the conceptual terrain—with conceptual engineering being the method of choice. This paper takes a step back to provide theoretical and methodological grounding for future work on conceptual engineering in machine learning. To this end, we consider the functional roles of concepts in machine learning, the underlying causes and types of deficiency, and map out criteria for the successful propagation of reengineered concepts within and beyond the machine learning community. Moreover, we discuss how the space of viable conceptual revisions in machine learning is constrained by the need for operationalization, and how tensions can occur between the sociopolitical desirability and the computational implementability of relevant conceptual engineering projects. Overall, our goal is to delineate how conceptual work in philosophy ought to be if the goal is for our contribution to permeate through the science and practice of machine learning.

In the literature on expert trust, it is often assumed that track records are the gold standard for evaluating expertise, and the difficulty of expert identification arises from either the lack of access to track records, or the inability to assess them. I show, using a computational model, that even in an idealized environment where agents have a God’s eye view on track records, they may fail to identify experts. Under plausible conditions, selecting testimony based on track records ends up reducing overall accuracy, and preventing the community from identifying the real experts.

The threshold view says that a person forms an outright belief P if and only if her credence for P reaches a certain threshold. Using computer simulations, I compare different versions of the threshold view to understand how they perform under time pressure in decision problems. The results illuminate the strengths and weaknesses of the various cognitive strategies in different decision contexts. A threshold view that performs well across diverse contexts is likely to be a cognitively flexible and context-dependent fusion of several of the existing theories. The results of the simulations also cast doubts on the possibility of a threshold view that is both simple enough to streamline our reasoning while also allowing us to form good action-guiding beliefs.