Optimizing Amazon Ads: Redesigning Application Log Storage for Enhanced Efficiency; Examining the Sociotechnical Intersection of AI and Criminal Justice Through the Lens of Technological Politics

The pervasive issue at the heart of this thesis is the ethical deployment and efficiency of Artificial Intelligence (AI) within significant societal systems, specifically examining criminal justice and commercial advertising technologies. The critical importance of addressing these challenges lies in AI’s dual potential: on one hand, to significantly enhance operational efficiencies across societal sectors; on the other, to exacerbate existing disparities due to inherent biases in its application and implementation. This thesis confronts a two-pronged challenge, ensuring AI operates effectively without deepening social inequalities. It investigates the optimization of Amazon’s advertising architecture to enhance efficiency and evaluates the societal implications of AI tools like COMPAS in the criminal justice system, emphasizing the need for AI technologies that are technically robust and socially equitable.

The technical portion of my research primarily focused on improving Amazon’s AI-based advertisement system architecture, a project driven by the necessity to manage data more efficiently, particularly regarding storage and retrieval operations. This enhancement was critical to improve the system’s performance, thereby reducing operational costs and optimizing resource allocation. By designing and implementing a revamped architecture, and subsequently evaluating its performance through metrics such as data retrieval times and system load handling, the findings demonstrated significant improvements in both efficiency and cost-effectiveness. These outcomes validated the architectural changes, underscoring the potential for technical innovations to boost system capabilities in environments with high demands.

Conversely, the STS study delved deeply into the ethical considerations and societal impacts of AI applications in the criminal justice system, with a focused analysis of the COMPAS tool. This exploration involved reviewing judicial use cases, examining studies on algorithmic bias, and assessing the outcomes of risk assessment implementations. It became evident that COMPAS and similar AI tools frequently exhibit racial biases, with a tendency to erroneously categorize minority groups as higher risk more often than their white counterparts. This predisposition leads to disproportionately harsher sentencing for these groups, thereby perpetuating systemic racial disparities within the criminal justice system. Such findings underscore a critical flaw in the presumed objectivity of AI systems, revealing that they not only mirror but can also amplify the societal biases present in their training data.

Reflecting on the broader objectives of this thesis, the technical research successfully achieved its aims by enhancing the efficiency of Amazon’s advertising system. This demonstrated how architectural adjustments can significantly improve AI functionalities in commercial settings, showcasing the tangible benefits of targeted technical improvements in AI systems. However, the STS research illuminated deeper, systemic challenges that are not as readily solvable. While identifying and documenting the biases in AI tools used in criminal justice was a crucial step, fully addressing these biases involves complex, ongoing efforts. To propel this research forward, future scholars should explore innovative strategies for mitigating biases in AI systems, particularly in sensitive sectors like criminal justice. This includes developing comprehensive auditing procedures for AI applications and promoting a multidisciplinary approach to AI development that incorporates ethical, technical, and societal perspectives. Such efforts are essential for creating AI systems that are not only efficient and effective but also fair and transparent.

Acknowledgments
I am immensely grateful to my advisor, Dr. Caitlin D. Wylie, for her invaluable guidance and unwavering support throughout this research journey. Her expertise and thoughtful insights have deeply influenced my work. I also wish to thank the faculty and peers at the School of Engineering and Applied Science, University of Virginia, for their stimulating discussions and encouragement. Lastly, I appreciate the broader academic and professional communities whose ongoing research into AI ethics has inspired and framed my studies.

School of Engineering and Applied Science

Bachelor of Science in Computer Science

Technical Advisor: Rosanne Vrugtman

STS Advisor: Caitlin Wylie