Measuring Harms and Empowering Users: Case Studies in Online Advertising and Generative AI

Abstract

Companies in the digital and generative AI ecosystems advertise their services and tools as transformative technologies that will empower people to do and create more in their everyday lives. But, these ecosystems share an important flaw: businesses have converged on strategies that leverage users’ data, time, and attention (and more), often at the cost of people's (online) safety and well-being. As researchers, one of the hardest things about trying to improve these ecosystems for users is their opacity – many online and generative AI companies engage in behavior that is often (intentionally) obfuscated from anyone trying to examine them. Because the underlying actions are often difficult to observe, the cascading effects on users can also be challenging to understand. I focus on two main threads across my research: uncovering the harms caused by online advertising platforms and generative AI companies' business practices. And second, I work to lay a foundation to empower users, and reimagine how we can establish technical solutions, influence company policies, and shape external regulations that reprioritize users’ needs and preferences. This dissertation makes the following contributions: I explore how companies in the online advertising and generative AI ecosystems make either implicit or explicit promises that they do not keep, and examine who is harmed when companies break their promises. Finally, I propose mitigations through soliciting industry cooperation and standards, regulation, and academic research to focus on empowering users. I do this through the lens of four case studies, two that focus on online advertisements, and two that are centered around generative AI. Both online ads and generative AI are increasingly ubiquitous in most people's everyday lives as they browse online, and the decisions these companies make have far-reaching effects on the quality of people's digital experiences. The four case studies in my dissertation are: 1) An exploration of the problematic behaviors of advertising platforms and the manipulative content spread opportunistically through advertisements during specific months of Russia's full-scale invasion of Ukraine; 2) Research that shows how inaccessible ads disrupt screen reader users’ web experience in ways that make people feel powerless, and prohibits their access to websites; 3) An investigation into how generative AI companies engage in practices that make it difficult for users to control how their data may be used to train models, and 4) A study of the ad-hoc strategies that people have individually developed to identify whether or not media in question has been AI generated, as part of their decision-making process towards whether or not it is reliable or trustworthy content. I contribute a unique perspective towards approaching this work by using mixed methods, combining large-scale, quantitative measurement studies to understand what problematic behaviors might be occurring as well as qualitative methods including interview studies to understand how these harms affect everyday people. I not only work to understand the different harms, I also strive to reenvision an ecosystem where companies re-prioritize users' preferences for privacy, security, and overall digital safety.