A Retrieval-Augmented Generation System for University Policy Querying; Relationships, Responsibility, and Research: A Care Ethics Examination of Neuralink's Animal Testing 6 views

Both my technical project and my STS research paper focus on the ethical responsibilities that come with building and deploying artificial intelligence in situations where the people, or beings, affected have limited power to protect themselves. My technical project is a CIO Policy Checker Chatbot that is designed to help University of Virginia student organization leaders navigate complicated university policies. My STS research paper applies Joan Tronto’s care ethics framework to Neuralink’s animal testing program and argues that the company failed its moral responsibilities toward the animals used in its brain-computer interface research. While these projects are very different, both are centered on accountability. In one case, student leaders depend on an LLM for accurate policy guidance and in the other, animals are dependent on researchers for ethical treatment. The CIO Policy Checker solves a common problem for student leaders at UVA. There are more than 650 Contracted Independent Organizations (CIOs), and each one must follow policies spread across dozens of websites. Something as simple as planning an event with food can require searching through multiple documents and emailing Student Affairs just to get a clear answer. Our solution uses a large language model combined with a vector database to let users ask natural-language questions like, “Can we serve food in this space?” The system retrieves the most relevant policy documents and provides a concise answer with citations showing the exact source, department, and document link. We intentionally scoped the project, so the tool provides guidance, not official policy interpretation. My STS research paper argues that Neuralink’s animal experimentation program represents a failure of care ethics, using Joan Tronto’s framework from “An Ethic of Care.” Tronto defines care through four stages: attentiveness, responsibility, competence, and responsiveness. Each stage creates moral obligations, especially when one group has power over another vulnerable group. Using investigative reporting, federal findings, and Neuralink’s own public statements, I argue that the company failed in all four stages. Neuralink prioritized speed and aggressive development timelines over attentiveness to animal suffering. Responsibility was weakened by executive pressure that pushed harmful decisions downward. Competence was compromised by conflicts of interest within its animal care committee, and responsiveness failed when the company answered criticism with deflection instead of reflection. Working on both projects at the same time changed how I think about responsible AI. Tronto’s care ethics framework became something I could directly apply to the CIO Policy Checker. It pushed my team to think carefully about who could be harmed by incorrect policy guidance, build clear disclaimers, and test the model instead of assuming it worked well. It also reminded us that deployment is not the finish line and that systems must improve with feedback and policy changes. At the same time, building the CIO Policy Checker made my STS research feel more real. It showed me how easy it is for teams to prioritize speed over careful testing, even with good intentions. Studying both projects together helped me realize that ethical engineering is not separate from technical work.

BS (Bachelor of Science)

Retrieval-Augmented Generation, Large Language Models, Brain Computer Interface, Care Ethics

School of Engineering and Applied Science Bachelor of Science in Computer Science Technical Advisor: Mark Sherriff STS Advisor: Benjamin Laugelli Technical Team Members: Anjana Raman, Elva Chen

English

2026-05-08