Please register for the CV & Online Identity Workshop below.

As search and recommendation algorithms continue to permeate all aspects of human decision making, from how we navigate the web to how decisions are made in courts, their impact on society has become a major topic of debate. Despite the unprecedented reach and persuasive capacity of the environments these algorithms operate in, identifying their impact is challenging because they are built around existing information ecosystems, trained on past behavioral data, and rarely an individual's only source of information. Yet, as we've seen -- from a man storming a pizzeria with an assault rifle in search of a secret pedophile ring, to a man committing mass murder at a predominantly Black church after being radicalized online -- the consequences of these environments can and do spill over into the real world. Given their deep individual impact, research on these environments should not only be focused on the extent to which they have a sizable impact, but also on developing tools for identifying how, when, and why they produce biased, false, or unjust results. For it is only with such a proactive approach that we can hold the owners of these environments accountable for the information they spread.

In this thesis, I attempt to address questions about the impact of search and recommendation algorithms on society by taking an interdisciplinary approach rooted in psychology, behavioral science, and network science. In chapter one, I provide a brief introduction to online environments, ranked list interfaces, and human information seeking. In chapter two, I describe the results from a series of behavioral experiments that examine the impact of web search interfaces on political attitudes. In chapter three, I cover three papers in which I used a combination of surveys, real users, and simulated user activity to conduct algorithm audits on web search rankings and autocomplete search suggestions. In chapter four, I explore methods I developed for collecting ecologically valid data, and two forthcoming reports which use such data. Lastly, in chapter five, I conclude with recommendations for future work in this vein, emphasize the need for working with subject-matter experts, and advocate for greater infrastructure and data sharing among academics.