Abstract
In my technical capstone project, I describe my work as a software engineer intern at a large financial services organization. I automated the member account closure process, in which remaining funds had to be processed and reallocated according to established financial procedures. While the process had been in place for years, it was highly time-consuming, requiring thousands of labor hours annually. To streamline the process, I developed a Blazor web app using .NET and C#, along with data processing tools, automating account closure. I applied the Lean Six Sigma approach, a data-driven project management methodology, to guide the design and testing process. Prior to the solution being developed, the account closure process was estimated to require over 2,000 hours of manual work annually, equating to approximately $100,000 in labor costs. Implementation of the automation eliminates these costs. Potential errors and delays associated with the manual process are also drastically reduced. In addition to its technical impact, the project highlights the tension in how organizations balance efficiency with human oversight.
In my STS research paper, I explore why people who have all studied AI, technology, and society so extensively hold such radically differing visions of AI’s risks and possibilities. In recent discussions of AI risk, disagreements among experts are treated as factual disputes arising over how dangerous AI really is, how soon powerful AI will emerge, or how much harm it currently causes, all of which could be resolved by providing better evidence or clearer thinking. My paper identifies three ideological groups within this debate: existential risk proponents, accelerationists, and critical AI scholars. By examining works on the intellectual origins of these groups and performing discourse analysis on primary texts and policy artifacts, I argue that disagreements between them are rooted in foundational differences in values and worldviews, rather than factual matters. I highlight two foundational differences between the groups: knowledge bases and time horizons. The findings in this paper indicate that current governance structures in the United States are insufficient for addressing legitimate concerns raised by each of these groups, giving overwhelming control to the ideological group that has the most influence. Thus, reform is necessary to enable greater representation of each competing ideological group in AI policy, allowing for the creation of regulations that address current harms and existential risks without hindering fruitful development.
